Index: llvm/trunk/lib/Target/X86/X86.td
===================================================================
--- llvm/trunk/lib/Target/X86/X86.td
+++ llvm/trunk/lib/Target/X86/X86.td
@@ -449,6 +449,7 @@
 include "X86SchedBroadwell.td"
 include "X86ScheduleSLM.td"
 include "X86ScheduleZnver1.td"
+include "X86ScheduleBdVer2.td"
 include "X86ScheduleBtVer2.td"
 include "X86SchedSkylakeClient.td"
 include "X86SchedSkylakeServer.td"
@@ -1010,7 +1011,7 @@
 ]>;
 
 // Bulldozer
-def : Proc<"bdver1", [
+def : ProcessorModel<"bdver1", BdVer2Model, [
   FeatureX87,
   FeatureCMOV,
   FeatureXOP,
@@ -1035,7 +1036,7 @@
   FeatureMacroFusion
 ]>;
 // Piledriver
-def : Proc<"bdver2", [
+def : ProcessorModel<"bdver2", BdVer2Model, [
   FeatureX87,
   FeatureCMOV,
   FeatureXOP,
Index: llvm/trunk/lib/Target/X86/X86PfmCounters.td
===================================================================
--- llvm/trunk/lib/Target/X86/X86PfmCounters.td
+++ llvm/trunk/lib/Target/X86/X86PfmCounters.td
@@ -91,6 +91,18 @@
 }
 def : PfmCountersBinding<"skylake-avx512", SkylakeServerPfmCounters>;
 
+def BdVer2PfmCounters : ProcPfmCounters {
+  let CycleCounter = PfmCounter<"cpu_clk_unhalted">;
+  let UopsCounter = PfmCounter<"retired_uops">;
+  let IssueCounters = [
+    PfmIssueCounter<"PdFPU0", "dispatched_fpu_ops:ops_pipe0 + dispatched_fpu_ops:ops_dual_pipe0">,
+    PfmIssueCounter<"PdFPU1", "dispatched_fpu_ops:ops_pipe1 + dispatched_fpu_ops:ops_dual_pipe1">,
+    PfmIssueCounter<"PdFPU2", "dispatched_fpu_ops:ops_pipe2 + dispatched_fpu_ops:ops_dual_pipe2">,
+    PfmIssueCounter<"PdFPU3", "dispatched_fpu_ops:ops_pipe3 + dispatched_fpu_ops:ops_dual_pipe3">
+  ];
+}
+def : PfmCountersBinding<"bdver2", BdVer2PfmCounters>;
+
 def BtVer2PfmCounters : ProcPfmCounters {
   let CycleCounter = PfmCounter<"cpu_clk_unhalted">;
   let UopsCounter = PfmCounter<"retired_uops">;
Index: llvm/trunk/lib/Target/X86/X86ScheduleBdVer2.td
===================================================================
--- llvm/trunk/lib/Target/X86/X86ScheduleBdVer2.td
+++ llvm/trunk/lib/Target/X86/X86ScheduleBdVer2.td
@@ -0,0 +1,1278 @@
+//=- X86ScheduleBdVer2.td - X86 BdVer2 (Piledriver) Scheduling * tablegen -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the machine model for AMD bdver2 (Piledriver) to support
+// instruction scheduling and other instruction cost heuristics.
+// Based on:
+//  * AMD Software Optimization Guide for AMD Family 15h Processors.
+//    https://support.amd.com/TechDocs/47414_15h_sw_opt_guide.pdf
+//  * The microarchitecture of Intel, AMD and VIA CPUs, By Agner Fog
+//    http://www.agner.org/optimize/microarchitecture.pdf
+//  * https://www.realworldtech.com/bulldozer/
+//    Yes, that is for Bulldozer aka bdver1, not Piledriver aka bdver2.
+//
+//===----------------------------------------------------------------------===//
+
+def BdVer2Model : SchedMachineModel {
+  let IssueWidth = 4; // Up to 4 IPC can be decoded, issued, retired.
+  let MicroOpBufferSize = 128; // RCU reorder buffer size, which is unconfirmed.
+  let LoopMicroOpBufferSize = -1; // There does not seem to be a loop buffer.
+  let LoadLatency = 4; // L1 data cache has a 4-cycle load-to-use latency.
+  let HighLatency = 25; // FIXME: any better choice?
+  let MispredictPenalty = 20; // Minimum branch misdirection penalty.
+
+  let PostRAScheduler = 1; // Enable Post RegAlloc Scheduler pass.
+
+  // FIXME: Incomplete. This flag is set to allow the scheduler to assign
+  //        a default model to unrecognized opcodes.
+  let CompleteModel = 0;
+} // SchedMachineModel
+
+let SchedModel = BdVer2Model in {
+
+
+//===----------------------------------------------------------------------===//
+// Pipes
+//===----------------------------------------------------------------------===//
+
+// There are total of eight pipes.
+
+//===----------------------------------------------------------------------===//
+// Integer execution pipes
+//
+
+// Two EX (ALU) pipes.
+def PdEX0  : ProcResource<1>; // ALU, Integer Pipe0
+def PdEX1  : ProcResource<1>; // ALU, Integer Pipe1
+def PdEX01 : ProcResGroup<[PdEX0, PdEX1]>;
+
+// Two AGLU pipes, identical.
+def PdAGLU01 : ProcResource<2>; // AGU, Integer Pipe[23]
+
+//===----------------------------------------------------------------------===//
+// Floating point execution pipes
+//
+
+// Four FPU pipes.
+
+def PdFPU0 : ProcResource<1>; // Vector/FPU Pipe0
+def PdFPU1 : ProcResource<1>; // Vector/FPU Pipe1
+def PdFPU2 : ProcResource<1>; // Vector/FPU Pipe2
+def PdFPU3 : ProcResource<1>; // Vector/FPU Pipe3
+
+// FPU grouping
+def PdFPU01 : ProcResGroup<[PdFPU0, PdFPU1]>;
+def PdFPU23 : ProcResGroup<[PdFPU2, PdFPU3]>;
+
+
+//===----------------------------------------------------------------------===//
+// RCU
+//===----------------------------------------------------------------------===//
+
+// The Retire Control Unit on Piledriver can retire up to 4 macro-ops per cycle.
+// On the other hand, the RCU reorder buffer size for Piledriver does not
+// seem be specified in any trustworthy source.
+// But as per https://www.realworldtech.com/bulldozer/6/ the Bulldozer had
+// RCU reorder buffer size of 128. So that is a good guess for now.
+def PdRCU : RetireControlUnit<128, 4>;
+
+
+//===----------------------------------------------------------------------===//
+// Pipelines
+//===----------------------------------------------------------------------===//
+
+// There are total of two pipelines, each one with it's own scheduler.
+
+//===----------------------------------------------------------------------===//
+// Integer Pipeline Scheduling
+//
+
+// There is one Integer Scheduler per core.
+
+// Integer physical register file has 96 registers of 64-bit.
+def PdIntegerPRF : RegisterFile<96, [GR64, CCR]>;
+
+// Unified Integer, Memory Scheduler has 40 entries.
+def PdEX : ProcResGroup<[PdEX0, PdEX1, PdAGLU01]> {
+  // Up to 4 IPC can be decoded, issued, retired.
+  let BufferSize = 40;
+}
+
+
+//===----------------------------------------------------------------------===//
+// FPU Pipeline Scheduling
+//
+
+// The FPU unit is shared between the two cores.
+
+// FP physical register file has 160 registers of 128-bit.
+// Operations on 256-bit data types are cracked into two COPs.
+def PdFpuPRF : RegisterFile<160, [VR64, VR128, VR256], [1, 1, 2]>;
+
+// Unified FP Scheduler has 64 entries,
+def PdFPU : ProcResGroup<[PdFPU0, PdFPU1, PdFPU2, PdFPU3]> {
+  // Up to 4 IPC can be decoded, issued, retired.
+  let BufferSize = 64;
+}
+
+
+//===----------------------------------------------------------------------===//
+// Functional units
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// Load-Store Units
+//
+
+// FIXME: does this even make sense?
+
+def PdLoad  : ProcResGroup<[PdAGLU01]> {
+  // For Piledriver, the load queue is 40 entries deep.
+  let BufferSize = 40;
+}
+
+def PdStore : ProcResGroup<[PdAGLU01]> {
+  // For Piledriver, the store queue is 24 entries deep.
+  let BufferSize = 24;
+}
+
+//===----------------------------------------------------------------------===//
+// Integer Execution Units
+//
+
+def PdDiv    : ProcResource<1>; // PdEX0; unpipelined integer division
+def PdCount  : ProcResource<1>; // PdEX0; POPCNT, LZCOUNT
+
+def PdMul    : ProcResource<1>; // PdEX1; integer multiplication
+def PdBranch : ProcResource<1>; // PdEX1; JMP, fused branches
+
+//===----------------------------------------------------------------------===//
+// Floating-Point Units
+//
+
+// Two FMAC/FPFMA units.
+def PdFPFMA  : ProcResource<2>; // PdFPU0, PdFPU1
+
+// One 128-bit integer multiply-accumulate unit.
+def PdFPMMA  : ProcResource<1>; // PdFPU0
+
+// One fp conversion unit.
+def PdFPCVT  : ProcResource<1>; // PdFPU0
+
+// One unit for shuffles, packs, permutes, shifts.
+def PdFPXBR  : ProcResource<1>; // PdFPU1
+
+// Two 128-bit packed integer units.
+def PdFPMAL  : ProcResource<2>; // PdFPU2, PdFPU3
+
+// One FP store unit.
+def PdFPSTO  : ProcResource<1>; // PdFPU3
+
+
+//===----------------------------------------------------------------------===//
+// Basic helper classes.
+//===----------------------------------------------------------------------===//
+
+// Many SchedWrites are defined in pairs with and without a folded load.
+// Instructions with folded loads are usually micro-fused, so they only appear
+// as two micro-ops when dispatched by the schedulers.
+// This multiclass defines the resource usage for variants with and without
+// folded loads.
+multiclass PdWriteRes<SchedWrite SchedRW,
+                      list<ProcResourceKind> ExePorts, int Lat = 1,
+                      list<int> Res = [], int UOps = 1> {
+  def : WriteRes<SchedRW, ExePorts> {
+    let Latency = Lat;
+    let ResourceCycles = Res;
+    let NumMicroOps = UOps;
+  }
+}
+
+multiclass __pdWriteResPair<X86FoldableSchedWrite SchedRW,
+                            list<ProcResourceKind> ExePorts, int Lat,
+                            list<int> Res, int UOps,
+                            int LoadLat, int LoadRes, int LoadUOps> {
+  defm : PdWriteRes<SchedRW, ExePorts, Lat, Res, UOps>;
+
+  defm : PdWriteRes<SchedRW.Folded,
+                    !listconcat([PdLoad], ExePorts),
+                    !add(Lat, LoadLat),
+                    !if(!and(!empty(Res), !eq(LoadRes, 1)),
+                      [],
+                      !listconcat([LoadRes], Res)),
+                    !add(UOps, LoadUOps)>;
+}
+
+multiclass PdWriteResExPair<X86FoldableSchedWrite SchedRW,
+                            list<ProcResourceKind> ExePorts, int Lat = 1,
+                            list<int> Res = [], int UOps = 1,
+                            int LoadUOps = 0> {
+  defm : __pdWriteResPair<SchedRW, ExePorts, Lat, Res, UOps,
+                          /*LoadLat*/4, /*LoadRes*/1, LoadUOps>;
+}
+
+multiclass PdWriteResXMMPair<X86FoldableSchedWrite SchedRW,
+                             list<ProcResourceKind> ExePorts, int Lat = 1,
+                             list<int> Res = [], int UOps = 1,
+                             int LoadUOps = 0> {
+  defm : __pdWriteResPair<SchedRW, ExePorts, Lat, Res, UOps,
+                           /*LoadLat*/5, /*LoadRes*/1, LoadUOps>;
+}
+
+multiclass PdWriteResYMMPair<X86FoldableSchedWrite SchedRW,
+                             list<ProcResourceKind> ExePorts, int Lat,
+                             list<int> Res, int UOps = 2,
+                             int LoadUOps = 0> {
+  defm : __pdWriteResPair<SchedRW, ExePorts, Lat, Res, UOps,
+                           /*LoadLat*/5, /*LoadRes*/2, LoadUOps>;
+}
+
+//===----------------------------------------------------------------------===//
+// Here be dragons.
+//===----------------------------------------------------------------------===//
+
+// L1 data cache has a 4-cycle load-to-use latency, so ReadAfterLd registers
+// needn't be available until 4 cycles after the memory operand.
+def : ReadAdvance<ReadAfterLd, 4>;
+
+// Vector loads are 5 cycles, so ReadAfterVec*Ld registers needn't be available
+// until 5 cycles after the memory operand.
+def : ReadAdvance<ReadAfterVecLd, 5>;
+def : ReadAdvance<ReadAfterVecXLd, 5>;
+def : ReadAdvance<ReadAfterVecYLd, 5>;
+
+// A folded store needs a cycle on the PdStore for the store data.
+def : WriteRes<WriteRMW, [PdStore]>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Loads, stores, and moves, not folded with other operations.
+////////////////////////////////////////////////////////////////////////////////
+
+def : WriteRes<WriteLoad,    [PdLoad]> { let Latency = 5; }
+def : WriteRes<WriteStore,   [PdStore]>;
+def : WriteRes<WriteStoreNT, [PdStore]>;
+def : WriteRes<WriteMove,    [PdEX01]>;
+
+// Load/store MXCSR.
+// FIXME: These are copy and pasted from WriteLoad/Store.
+def : WriteRes<WriteLDMXCSR, [PdLoad]> { let Latency = 5; }
+def : WriteRes<WriteSTMXCSR, [PdStore]> { let NumMicroOps = 2; }
+
+// Treat misc copies as a move.
+def : InstRW<[WriteMove], (instrs COPY)>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Idioms that clear a register, like xorps %xmm0, %xmm0.
+// These can often bypass execution ports completely.
+////////////////////////////////////////////////////////////////////////////////
+
+def : WriteRes<WriteZero, [/*No ExePorts*/]>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Branches don't produce values, so they have no latency, but they still
+// consume resources. Indirect branches can fold loads.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResExPair<WriteJump,  [PdEX1, PdBranch]>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Special case scheduling classes.
+////////////////////////////////////////////////////////////////////////////////
+
+def : WriteRes<WriteSystem,     [PdEX01]> { let Latency = 100; }
+def : WriteRes<WriteMicrocoded, [PdEX01]> { let Latency = 100; }
+def : WriteRes<WriteFence,      [PdStore]>;
+
+def PdWriteXLAT : SchedWriteRes<[PdEX01]> {
+  let Latency = 6;
+}
+def : InstRW<[PdWriteXLAT], (instrs XLAT)>;
+
+def PdWriteLARrr : SchedWriteRes<[PdEX01]> {
+  let Latency = 184;
+  let NumMicroOps = 45;
+}
+def : InstRW<[PdWriteLARrr], (instregex "LAR(16|32|64)rr",
+                                        "LSL(16|32|64)rr")>;
+
+// Nops don't have dependencies, so there's no actual latency, but we set this
+// to '1' to tell the scheduler that the nop uses an ALU slot for a cycle.
+def : WriteRes<WriteNop, [PdEX01]>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Arithmetic.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResExPair<WriteALU,     [PdEX01]>;
+
+def PdWriteLXADD : SchedWriteRes<[PdEX01]> {
+  let Latency = 6;
+  let NumMicroOps = 4;
+}
+def : InstRW<[PdWriteLXADD], (instrs LXADD8, LXADD16, LXADD32, LXADD64)>;
+
+def PdWriteBMI1 : SchedWriteRes<[PdEX01]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteBMI1],
+             (instrs BLCFILL32rr, BLCFILL64rr, BLCI32rr, BLCI64rr,
+                     BLCIC32rr, BLCIC64rr, BLCMSK32rr, BLCMSK64rr,
+                     BLCS32rr, BLCS64rr, BLSFILL32rr, BLSFILL64rr,
+                     BLSIC32rr, BLSIC64rr, T1MSKC32rr, T1MSKC64rr,
+                     TZMSK32rr, TZMSK64rr)>;
+
+def PdWriteBMI1m : SchedWriteRes<[PdEX01]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteBMI1m],
+             (instrs BLCFILL32rm, BLCFILL64rm, BLCI32rm, BLCI64rm,
+                     BLCIC32rm, BLCIC64rm, BLCMSK32rm, BLCMSK64rm,
+                     BLCS32rm, BLCS64rm, BLSFILL32rm, BLSFILL64rm,
+                     BLSIC32rm, BLSIC64rm, T1MSKC32rm, T1MSKC64rm,
+                     TZMSK32rm, TZMSK64rm)>;
+
+defm : PdWriteResExPair<WriteADC,    [PdEX01],                  1,  [2]>;
+
+defm : PdWriteRes<WriteBSWAP32,      [PdEX1]>;
+defm : PdWriteRes<WriteBSWAP64,      [PdEX1]>;
+defm : PdWriteRes<WriteCMPXCHG,      [PdEX1],                   3,  [],       5>;
+defm : PdWriteRes<WriteCMPXCHGRMW,   [PdEX1, PdStore, PdLoad],  3,  [], 2>;
+defm : PdWriteRes<WriteXCHG,         [PdEX1],                   1,  [],       2>;
+
+def PdWriteCMPXCHG8rr : SchedWriteRes<[PdEX1]> {
+  let Latency = 3;
+  let NumMicroOps = 3;
+}
+def : InstRW<[PdWriteCMPXCHG8rr], (instrs CMPXCHG8rr)>;
+
+def PdWriteCMPXCHG8rm : SchedWriteRes<[PdEX1]> {
+  let Latency = 3;
+  let NumMicroOps = 5;
+}
+def : InstRW<[PdWriteCMPXCHG8rm], (instrs CMPXCHG8rm)>;
+
+def PdWriteCMPXCHG16rm_CMPXCHG32rm_CMPXCHG64rm : SchedWriteRes<[PdEX1]> {
+  let Latency = 3;
+  let NumMicroOps = 6;
+}
+def : InstRW<[PdWriteCMPXCHG16rm_CMPXCHG32rm_CMPXCHG64rm],
+             (instrs CMPXCHG16rm, CMPXCHG32rm, CMPXCHG64rm)>;
+
+def PdWriteCMPXCHG8B : SchedWriteRes<[PdEX1]> {
+  let Latency = 3;
+  let NumMicroOps = 18;
+}
+def : InstRW<[PdWriteCMPXCHG8B], (instrs CMPXCHG8B)>;
+
+def PdWriteCMPXCHG16B : SchedWriteRes<[PdEX1]> {
+  let Latency = 3;
+  let NumMicroOps = 22;
+}
+def : InstRW<[PdWriteCMPXCHG16B], (instrs CMPXCHG16B)>;
+
+def PdWriteXCHG16rr : SchedWriteRes<[PdEX1]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteXCHG16rr], (instrs XCHG16rr)>;
+
+def PdWriteXADD : SchedWriteRes<[PdEX1]> {
+  let Latency = 2;
+  let NumMicroOps = 4;
+}
+def : InstRW<[PdWriteXADD], (instrs XADD8rr, XADD16rr, XADD32rr, XADD64rr)>;
+
+def PdWriteXADDm : SchedWriteRes<[PdEX1]> {
+let Latency = 6;
+let NumMicroOps = 4;
+}
+def : InstRW<[PdWriteXADDm], (instrs XADD8rm, XADD16rm, XADD32rm, XADD64rm)>;
+
+defm : PdWriteResExPair<WriteIMul8,     [PdEX1, PdMul],          4>;
+defm : PdWriteResExPair<WriteIMul16,    [PdEX1, PdMul],          4,  [],    2>;
+defm : PdWriteResExPair<WriteIMul16Imm, [PdEX1, PdMul],          5,  [],    2>;
+defm : PdWriteResExPair<WriteIMul16Reg, [PdEX1, PdMul],          4>;
+defm : PdWriteResExPair<WriteIMul32,    [PdEX1, PdMul],          4>;
+defm : PdWriteResExPair<WriteIMul32Imm, [PdEX1, PdMul],          4,  [],    1, 1>;
+defm : PdWriteResExPair<WriteIMul32Reg, [PdEX1, PdMul],          4>;
+defm : PdWriteResExPair<WriteIMul64,    [PdEX1, PdMul],          6,  [1, 4]>;
+defm : PdWriteResExPair<WriteIMul64Imm, [PdEX1, PdMul],          6,  [1, 4],1, 1>;
+defm : PdWriteResExPair<WriteIMul64Reg, [PdEX1, PdMul],          6,  [1, 4]>;
+defm : X86WriteResUnsupported<WriteIMulH>; // BMI2 MULX
+
+defm : PdWriteResExPair<WriteDiv8,    [PdEX1, PdDiv],           12,  [1, 12]>;
+defm : PdWriteResExPair<WriteDiv16,   [PdEX1, PdDiv],           15,  [1, 15],   2>;
+defm : PdWriteResExPair<WriteDiv32,   [PdEX1, PdDiv],           14,  [1, 14],   2>;
+defm : PdWriteResExPair<WriteDiv64,   [PdEX1, PdDiv],           14,  [1, 14],   2>;
+
+defm : PdWriteResExPair<WriteIDiv8,   [PdEX1, PdDiv],           12,  [1, 12]>;
+defm : PdWriteResExPair<WriteIDiv16,  [PdEX1, PdDiv],           15,  [1, 17],   2>;
+defm : PdWriteResExPair<WriteIDiv32,  [PdEX1, PdDiv],           14,  [1, 25],   2>;
+defm : PdWriteResExPair<WriteIDiv64,  [PdEX1, PdDiv],           14,  [1, 14],   2>;
+
+defm : PdWriteResExPair<WriteCRC32,   [PdEX01],                  3,  [4],       3>;
+
+def PdWriteCRC32r32r16 : SchedWriteRes<[PdEX01]> {
+  let Latency = 5;
+  let ResourceCycles = [4];
+  let NumMicroOps = 5;
+}
+def : InstRW<[PdWriteCRC32r32r16], (instrs CRC32r32r16)>;
+
+def PdWriteCRC32r32r32 : SchedWriteRes<[PdEX01]> {
+  let Latency = 6;
+  let ResourceCycles = [4];
+  let NumMicroOps = 7;
+}
+def : InstRW<[PdWriteCRC32r32r32], (instrs CRC32r32r32)>;
+
+def PdWriteCRC32r64r64 : SchedWriteRes<[PdEX01]> {
+  let Latency = 10;
+  let ResourceCycles = [4];
+  let NumMicroOps = 11;
+}
+def : InstRW<[PdWriteCRC32r64r64], (instrs CRC32r64r64)>;
+
+defm : PdWriteResExPair<WriteCMOV,    [PdEX01]>; // Conditional move.
+defm : PdWriteResExPair<WriteCMOV2,   [PdEX01], 1, [], 1, 1>; // Conditional (CF + ZF flag) move.
+
+def : InstRW<[WriteCMOV2.Folded], (instrs CMOVG16rm, CMOVG32rm, CMOVG64rm,
+                                          CMOVGE16rm, CMOVGE32rm, CMOVGE64rm,
+                                          CMOVL16rm, CMOVL32rm, CMOVL64rm,
+                                          CMOVLE16rm, CMOVLE32rm, CMOVLE64rm)>;
+
+defm : PdWriteRes<WriteFCMOV,        [PdFPU0, PdFPFMA]>; // x87 conditional move.
+
+def : WriteRes<WriteSETCC,           [PdEX01]>; // Setcc.
+def : WriteRes<WriteSETCCStore,      [PdEX01, PdStore]>;
+
+def PdWriteSETGEmSETGmSETLEmSETLm : SchedWriteRes<[PdEX01]> {
+  let ResourceCycles = [2];
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteSETGEmSETGmSETLEmSETLm], (instrs SETGEm, SETGm,
+                                                      SETLEm, SETLm)>;
+
+defm : PdWriteRes<WriteLAHFSAHF,      [PdEX01],          2,  [],     2>;
+
+def WriteLAHF : SchedWriteRes<[PdEX01]> {
+  let Latency = 2;
+  let NumMicroOps = 4;
+}
+def : InstRW<[WriteLAHF], (instrs LAHF)>;
+
+def WriteSAHF : SchedWriteRes<[PdEX01]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+def : InstRW<[WriteSAHF], (instrs SAHF)>;
+
+defm : PdWriteRes<WriteBitTest,          [PdEX01],         1, [1],     1>;
+defm : PdWriteRes<WriteBitTestImmLd,     [PdEX01, PdLoad], 5, [1, 1],  1>;
+defm : PdWriteRes<WriteBitTestRegLd,     [PdEX01, PdLoad], 5, [1, 1],  7>;
+defm : PdWriteRes<WriteBitTestSet,       [PdEX01],         2, [1],     2>;
+defm : PdWriteRes<WriteBitTestSetImmLd,  [PdEX01, PdLoad], 6, [1, 1],  4>;
+defm : PdWriteRes<WriteBitTestSetImmRMW, [PdEX01, PdLoad], 6, [1, 1],  4>;
+defm : PdWriteRes<WriteBitTestSetRegLd,  [PdEX01, PdLoad], 6, [1, 1], 10>;
+defm : PdWriteRes<WriteBitTestSetRegRMW, [PdEX01, PdLoad], 6, [1, 1], 10>;
+
+// This is for simple LEAs with one or two input operands.
+// FIXME: SAGU 3-operand LEA
+def : WriteRes<WriteLEA,              [PdEX01]> { let NumMicroOps = 2; }
+
+// Bit counts.
+defm : PdWriteResExPair<WriteBSF,     [PdEX01],          3,  [4],     6, 2>;
+defm : PdWriteResExPair<WriteBSR,     [PdEX01],          4,  [4],     7, 2>;
+defm : PdWriteResExPair<WritePOPCNT,  [PdEX01],          4>;
+defm : PdWriteResExPair<WriteLZCNT,   [PdEX01],          2,  [],      2>;
+defm : PdWriteResExPair<WriteTZCNT,   [PdEX01],          2,  [2],     2>;
+
+// BMI1 BEXTR, BMI2 BZHI
+defm : PdWriteResExPair<WriteBEXTR,   [PdEX01],          2,  [],     2>;
+defm : PdWriteResExPair<WriteBLS,     [PdEX01],          2,  [],     2>;
+defm : PdWriteResExPair<WriteBZHI,    [PdEX01]>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Integer shifts and rotates.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResExPair<WriteShift,    [PdEX01]>;
+defm : PdWriteResExPair<WriteShiftCL,  [PdEX01]>;
+defm : PdWriteResExPair<WriteRotate,   [PdEX01]>;
+defm : PdWriteResExPair<WriteRotateCL, [PdEX01]>;
+
+def PdWriteRCL8rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 12;
+  let NumMicroOps = 26;
+}
+def : InstRW<[PdWriteRCL8rCL], (instrs RCL8rCL)>;
+
+def PdWriteRCR8ri : SchedWriteRes<[PdEX01]> {
+  let Latency = 12;
+  let NumMicroOps = 23;
+}
+def : InstRW<[PdWriteRCR8ri], (instrs RCR8ri)>;
+
+def PdWriteRCR8rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 11;
+  let NumMicroOps = 24;
+}
+def : InstRW<[PdWriteRCR8rCL], (instrs RCR8rCL)>;
+
+def PdWriteRCL16rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 10;
+  let NumMicroOps = 22;
+}
+def : InstRW<[PdWriteRCL16rCL], (instrs RCL16rCL)>;
+
+def PdWriteRCR16ri : SchedWriteRes<[PdEX01]> {
+  let Latency = 10;
+  let NumMicroOps = 19;
+}
+def : InstRW<[PdWriteRCR16ri], (instrs RCR16ri)>;
+
+def PdWriteRCL32rCLRCL64rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 7;
+  let NumMicroOps = 17;
+}
+def : InstRW<[PdWriteRCL32rCLRCL64rCL], (instrs RCL32rCL, RCL64rCL)>;
+
+def PdWriteRCR64rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 7;
+  let NumMicroOps = 16;
+}
+def : InstRW<[PdWriteRCR64rCL], (instrs RCR64rCL)>;
+
+def PdWriteRCR32rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 7;
+  let NumMicroOps = 16;
+}
+def : InstRW<[PdWriteRCR32rCL ], (instrs RCR32rCL)>;
+
+def PdWriteRCR32riRCR64ri : SchedWriteRes<[PdEX01]> {
+  let Latency = 7;
+  let NumMicroOps = 15;
+}
+def : InstRW<[PdWriteRCR32riRCR64ri], (instrs RCR32ri, RCR64ri)>;
+
+
+def PdWriteRCR16rCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 9;
+  let NumMicroOps = 20;
+}
+def : InstRW<[PdWriteRCR16rCL], (instrs RCR16rCL)>;
+
+def PdWriteRCL16ri : SchedWriteRes<[PdEX01]> {
+  let Latency = 11;
+  let NumMicroOps = 21;
+}
+def : InstRW<[PdWriteRCL16ri], (instrs RCL16ri)>;
+
+def PdWriteRCL3264ri : SchedWriteRes<[PdEX01]> {
+  let Latency = 8;
+  let NumMicroOps = 16;
+}
+def : InstRW<[PdWriteRCL3264ri], (instrs RCL32ri, RCL64ri)>;
+
+def PdWriteRCL8ri : SchedWriteRes<[PdEX01]> {
+  let Latency = 13;
+  let NumMicroOps = 25;
+}
+def : InstRW<[PdWriteRCL8ri], (instrs RCL8ri)>;
+
+// SHLD/SHRD.
+defm : PdWriteRes<WriteSHDrri,       [PdEX01],         4, [6], 6>;
+defm : PdWriteRes<WriteSHDrrcl,      [PdEX01],         4, [8], 7>;
+
+def PdWriteSHLD32rri8SHRD16rri8 : SchedWriteRes<[PdEX01]> {
+  let Latency = 3;
+  let ResourceCycles = [6];
+  let NumMicroOps = 6;
+}
+def : InstRW<[PdWriteSHLD32rri8SHRD16rri8 ], (instrs SHLD32rri8, SHRD16rri8)>;
+
+def PdWriteSHLD16rrCLSHLD32rrCLSHRD32rrCL : SchedWriteRes<[PdEX01]> {
+  let Latency = 4;
+  let ResourceCycles = [8];
+  let NumMicroOps = 7;
+}
+def : InstRW<[PdWriteSHLD16rrCLSHLD32rrCLSHRD32rrCL], (instrs SHLD16rrCL,
+                                                              SHLD32rrCL,
+                                                              SHRD32rrCL)>;
+
+defm : PdWriteRes<WriteSHDmri,       [PdLoad, PdEX01], 4, [1, 22], 8>;
+defm : PdWriteRes<WriteSHDmrcl,      [PdLoad, PdEX01], 4, [1, 22], 8>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Floating point. This covers both scalar and vector operations.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteRes<WriteFLD0,               [PdFPU1, PdFPSTO], 3>;
+defm : PdWriteRes<WriteFLD1,               [PdFPU1, PdFPSTO], 3>;
+defm : PdWriteRes<WriteFLDC,               [PdFPU1, PdFPSTO], 3>;
+
+defm : PdWriteRes<WriteFLoad,              [PdLoad, PdFPU01, PdFPFMA], 5>;
+defm : PdWriteRes<WriteFLoadX,             [PdLoad, PdFPU01, PdFPFMA], 5>;
+defm : PdWriteRes<WriteFLoadY,             [PdLoad, PdFPU01, PdFPFMA], 5, [], 2>;
+
+defm : PdWriteRes<WriteFMaskedLoad,        [PdLoad, PdFPU01, PdFPFMA], 6, [1, 1, 2]>;
+defm : PdWriteRes<WriteFMaskedLoadY,       [PdLoad, PdFPU01, PdFPFMA], 6, [2, 2, 4], 2>;
+
+defm : PdWriteRes<WriteFStore,             [PdStore, PdFPU1,  PdFPSTO], 2>;
+defm : PdWriteRes<WriteFStoreX,            [PdStore, PdFPU1,  PdFPSTO]>;
+defm : PdWriteRes<WriteFStoreY,            [PdStore, PdFPU1,  PdFPSTO], 1, [], 4>;
+
+def PdWriteMOVHPm : SchedWriteRes<[PdStore, PdFPU1,  PdFPSTO]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteMOVHPm], (instrs MOVHPDmr, MOVHPSmr, VMOVHPDmr, VMOVHPSmr)>;
+
+def PdWriteVMOVUPDYmrVMOVUPSYmr : SchedWriteRes<[PdStore, PdFPU1,  PdFPSTO]> {
+  let NumMicroOps = 8;
+}
+def : InstRW<[PdWriteVMOVUPDYmrVMOVUPSYmr], (instrs VMOVUPDYmr, VMOVUPSYmr)>;
+
+defm : PdWriteRes<WriteFStoreNT,           [PdStore, PdFPU1,  PdFPSTO], 3>;
+defm : PdWriteRes<WriteFStoreNTX,          [PdStore, PdFPU1,  PdFPSTO], 3>;
+defm : PdWriteRes<WriteFStoreNTY,          [PdStore, PdFPU1,  PdFPSTO], 3, [2, 2, 2], 4>;
+
+defm : PdWriteRes<WriteFMaskedStore,       [PdStore, PdFPU01, PdFPFMA], 6, [1, 1, 4], 18>;
+defm : PdWriteRes<WriteFMaskedStoreY,      [PdStore, PdFPU01, PdFPFMA], 6, [2, 2, 4], 34>;
+
+defm : PdWriteRes<WriteFMove,              [PdFPU01, PdFPFMA]>;
+defm : PdWriteRes<WriteFMoveX,             [PdFPU01, PdFPFMA]>;
+defm : PdWriteRes<WriteFMoveY,             [PdFPU01, PdFPFMA], 2, [2, 2], 2>;
+
+defm : PdWriteRes<WriteEMMS,               [PdFPU01, PdFPFMA], 2>;
+
+defm : PdWriteResXMMPair<WriteFAdd,         [PdFPU0, PdFPFMA],  5>;
+defm : PdWriteResXMMPair<WriteFAddX,        [PdFPU0, PdFPFMA],  5>;
+defm : PdWriteResYMMPair<WriteFAddY,        [PdFPU0, PdFPFMA],  5, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFAddZ>;
+
+defm : PdWriteResXMMPair<WriteFAdd64,       [PdFPU0, PdFPFMA],  5>;
+defm : PdWriteResXMMPair<WriteFAdd64X,      [PdFPU0, PdFPFMA],  5>;
+defm : PdWriteResYMMPair<WriteFAdd64Y,      [PdFPU0, PdFPFMA],  5, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
+
+defm : PdWriteResXMMPair<WriteFCmp,         [PdFPU0, PdFPFMA],  2>;
+defm : PdWriteResXMMPair<WriteFCmpX,        [PdFPU0, PdFPFMA],  2>;
+defm : PdWriteResYMMPair<WriteFCmpY,        [PdFPU0, PdFPFMA],  2, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFCmpZ>;
+
+defm : PdWriteResXMMPair<WriteFCmp64,       [PdFPU0, PdFPFMA],  2>;
+defm : PdWriteResXMMPair<WriteFCmp64X,      [PdFPU0, PdFPFMA],  2>;
+defm : PdWriteResYMMPair<WriteFCmp64Y,      [PdFPU0, PdFPFMA],  2, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
+
+defm : PdWriteResXMMPair<WriteFCom,         [PdFPU0, PdFPFMA, PdEX0], 1, [], 2>;
+
+def PdWriteFCOMPm : SchedWriteRes<[PdFPU1, PdFPFMA]> {
+  let Latency = 6;
+}
+def : InstRW<[PdWriteFCOMPm], (instrs FCOM32m, FCOM64m, FCOMP32m, FCOMP64m)>;
+
+def PdWriteTST_F_UCOM_FPPr : SchedWriteRes<[PdFPU1, PdFPFMA]>;
+def : InstRW<[PdWriteTST_F_UCOM_FPPr], (instrs TST_F, UCOM_FPPr)>;
+
+defm : PdWriteResXMMPair<WriteFMul,         [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResXMMPair<WriteFMulX,        [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResYMMPair<WriteFMulY,        [PdFPU1, PdFPFMA],  5, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFMulZ>;
+
+defm : PdWriteResXMMPair<WriteFMul64,       [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResXMMPair<WriteFMul64X,      [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResYMMPair<WriteFMul64Y,      [PdFPU1, PdFPFMA],  5, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFMul64Z>;
+
+defm : PdWriteResXMMPair<WriteFMA,          [PdFPU, PdFPFMA], 5>;
+defm : PdWriteResXMMPair<WriteFMAX,         [PdFPU, PdFPFMA], 5>;
+defm : PdWriteResYMMPair<WriteFMAY,         [PdFPU, PdFPFMA], 5,   [1, 1]>;
+defm : X86WriteResPairUnsupported<WriteFMAZ>;
+
+
+defm : PdWriteResXMMPair<WriteDPPD,         [PdFPU1, PdFPFMA], 15, [1, 3],  15, 2>;
+
+defm : PdWriteResXMMPair<WriteDPPS,         [PdFPU1, PdFPFMA], 25, [1, 3],  16, 2>;
+defm : PdWriteResYMMPair<WriteDPPSY,        [PdFPU1, PdFPFMA], 27, [2, 6], /*or 29*/ 25, 4>;
+defm : X86WriteResPairUnsupported<WriteDPPSZ>;
+
+def PdWriteVDPPSrri : SchedWriteRes<[PdFPU1, PdFPFMA]> {
+  let Latency = 25;
+  let ResourceCycles = [1, 3];
+  let NumMicroOps = 17;
+}
+def : InstRW<[PdWriteVDPPSrri], (instrs VDPPSrri)>;
+
+defm : PdWriteResXMMPair<WriteFRcp,         [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResXMMPair<WriteFRcpX,        [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResYMMPair<WriteFRcpY,        [PdFPU1, PdFPFMA],  5, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFRcpZ>;
+
+defm : PdWriteResXMMPair<WriteFRsqrt,       [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResXMMPair<WriteFRsqrtX,      [PdFPU1, PdFPFMA],  5>;
+defm : PdWriteResYMMPair<WriteFRsqrtY,      [PdFPU1, PdFPFMA],  5, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
+
+defm : PdWriteResXMMPair<WriteFDiv,         [PdFPU1, PdFPFMA], 9, [1, 19]>;
+defm : PdWriteResXMMPair<WriteFDivX,        [PdFPU1, PdFPFMA], 9, [1, 19]>;
+defm : PdWriteResYMMPair<WriteFDivY,        [PdFPU1, PdFPFMA], 9, [2, 38]>;
+defm : X86WriteResPairUnsupported<WriteFDivZ>;
+
+defm : PdWriteResXMMPair<WriteFDiv64,       [PdFPU1, PdFPFMA], 9, [1, 19]>;
+defm : PdWriteResXMMPair<WriteFDiv64X,      [PdFPU1, PdFPFMA], 9, [1, 19]>;
+defm : PdWriteResYMMPair<WriteFDiv64Y,      [PdFPU1, PdFPFMA], 9, [2, 38]>;
+defm : X86WriteResPairUnsupported<WriteFDiv64Z>;
+
+defm : PdWriteResXMMPair<WriteFSqrt,        [PdFPU1, PdFPFMA], 9, [1, 21]>;
+defm : PdWriteResXMMPair<WriteFSqrtX,       [PdFPU1, PdFPFMA], 9, [1, 21]>;
+defm : PdWriteResYMMPair<WriteFSqrtY,       [PdFPU1, PdFPFMA], 9, [2, 42]>;
+defm : X86WriteResPairUnsupported<WriteFSqrtZ>;
+
+defm : PdWriteResXMMPair<WriteFSqrt64,      [PdFPU1, PdFPFMA], 9, [1, 27]>;
+defm : PdWriteResXMMPair<WriteFSqrt64X,     [PdFPU1, PdFPFMA], 9, [1, 27]>;
+defm : PdWriteResYMMPair<WriteFSqrt64Y,     [PdFPU1, PdFPFMA], 9, [2, 54]>;
+defm : X86WriteResPairUnsupported<WriteFSqrt64Z>;
+
+defm : PdWriteResXMMPair<WriteFSqrt80,      [PdFPU1, PdFPFMA],  1, [1, 35]>;
+defm : PdWriteResXMMPair<WriteFSign,        [PdFPU1, PdFPFMA]>;
+
+defm : PdWriteResXMMPair<WriteFRnd,         [PdFPU1, PdFPSTO],  4>;
+defm : PdWriteResYMMPair<WriteFRndY,        [PdFPU1, PdFPSTO],  4, [2, 1], 2>;
+defm : X86WriteResPairUnsupported<WriteFRndZ>;
+
+def PdWriteVFRCZ : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 10;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteVFRCZ], (instrs VFRCZPDrr, VFRCZPSrr,
+                                     VFRCZSDrr, VFRCZSSrr)>;
+
+def PdWriteVFRCZm : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 15;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteVFRCZm], (instrs VFRCZPDrm, VFRCZPSrm,
+                                      VFRCZSDrm, VFRCZSSrm)>;
+
+def PdWriteVFRCZY : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 10;
+  let ResourceCycles = [2, 1];
+  let NumMicroOps = 4;
+}
+def : InstRW<[PdWriteVFRCZY], (instrs VFRCZPSYrr, VFRCZPDYrr)>;
+
+def PdWriteVFRCZYm : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 15;
+  let ResourceCycles = [2, 1];
+  let NumMicroOps = 8;
+}
+def : InstRW<[PdWriteVFRCZYm], (instrs VFRCZPSYrm, VFRCZPDYrm)>;
+
+defm : PdWriteResXMMPair<WriteFLogic,       [PdFPU01, PdFPFMA],  2>;
+defm : PdWriteResYMMPair<WriteFLogicY,      [PdFPU01, PdFPFMA],  2, [2, 2]>;
+defm : X86WriteResPairUnsupported<WriteFLogicZ>;
+
+defm : PdWriteResXMMPair<WriteFTest,        [PdFPU0, PdFPFMA, PdEX0],  1, [], 2>;
+defm : PdWriteResYMMPair<WriteFTestY,       [PdFPU01, PdFPFMA, PdEX0], 1, [2, 2, 1], 4, 2>;
+defm : X86WriteResPairUnsupported<WriteFTestZ>;
+
+defm : PdWriteResXMMPair<WriteFShuffle,     [PdFPU01, PdFPFMA],  2>;
+defm : PdWriteResYMMPair<WriteFShuffleY,    [PdFPU01, PdFPFMA],  2, [2, 2], 2>;
+defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
+
+def PdWriteVBROADCASTF128 : SchedWriteRes<[PdFPU01, PdFPFMA]> {
+  let Latency = 7;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteVBROADCASTF128], (instrs VBROADCASTF128)>;
+
+defm : PdWriteResXMMPair<WriteFVarShuffle,  [PdFPU01, PdFPFMA],  3, [1, 4]>;
+defm : PdWriteResYMMPair<WriteFVarShuffleY, [PdFPU01, PdFPFMA],  3, [2, 6], 2>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
+
+defm : PdWriteResXMMPair<WriteFBlend,       [PdFPU01, PdFPFMA],  2>;
+defm : PdWriteResYMMPair<WriteFBlendY,      [PdFPU01, PdFPFMA],  2, [2, 2], 2>;
+defm : X86WriteResPairUnsupported<WriteFBlendZ>;
+
+defm : PdWriteResXMMPair<WriteFVarBlend,    [PdFPU01, PdFPFMA],  2, [1, 4]>;
+defm : PdWriteResYMMPair<WriteFVarBlendY,   [PdFPU01, PdFPFMA],  2, [2, 6], 2>;
+defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
+
+defm : PdWriteResXMMPair<WriteFShuffle256,  [PdFPU01, PdFPFMA],  2, [], 2>;
+defm : X86WriteResPairUnsupported<WriteFVarShuffle256>;
+
+def PdWriteVEXTRACTF128rr : SchedWriteRes<[PdFPU01, PdFPFMA]> {
+  let Latency = 2;
+}
+def : InstRW<[PdWriteVEXTRACTF128rr], (instrs VEXTRACTF128rr)>;
+
+def PdWriteVEXTRACTF128mr : SchedWriteRes<[PdFPU01, PdFPFMA]> {
+  let Latency = 7;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteVEXTRACTF128mr], (instrs VEXTRACTF128mr)>;
+
+def PdWriteVPERM2F128rr : SchedWriteRes<[PdFPU01, PdFPFMA]> {
+  let Latency = 4;
+  let NumMicroOps = 8;
+}
+def : InstRW<[PdWriteVPERM2F128rr], (instrs VPERM2F128rr)>;
+
+def PdWriteVPERM2F128rm : SchedWriteRes<[PdFPU01, PdFPFMA]> {
+  let Latency = 8; // 4 + 4
+  let NumMicroOps = 10;
+}
+def : InstRW<[PdWriteVPERM2F128rm], (instrs VPERM2F128rm)>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Conversions.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResXMMPair<WriteCvtSS2I,   [PdFPU1, PdFPSTO, PdFPFMA, PdEX0], 13, [], 2>;
+
+defm : PdWriteResXMMPair<WriteCvtPS2I,   [PdFPU1, PdFPSTO], 4>;
+defm : PdWriteResYMMPair<WriteCvtPS2IY,  [PdFPU1, PdFPSTO], 4, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
+
+defm : PdWriteResXMMPair<WriteCvtSD2I,   [PdFPU1, PdFPSTO, PdFPFMA, PdEX0], 13, [], 2>;
+
+defm : PdWriteResXMMPair<WriteCvtPD2I,   [PdFPU1, PdFPSTO],          8, [],        2>;
+defm : PdWriteResYMMPair<WriteCvtPD2IY,  [PdFPU1, PdFPSTO, PdFPFMA], 8, [2, 1, 1], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
+
+def PdWriteMMX_CVTTPD2PIirr : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteMMX_CVTTPD2PIirr], (instrs MMX_CVTTPD2PIirr)>;
+
+// FIXME: f+3 ST, LD+STC latency
+defm : PdWriteResXMMPair<WriteCvtI2SS,   [PdFPU1, PdFPSTO], 4, [], 2>;
+// FIXME: .Folded version is one NumMicroOp *less*..
+
+defm : PdWriteResXMMPair<WriteCvtI2PS,   [PdFPU1, PdFPSTO], 4>;
+defm : PdWriteResYMMPair<WriteCvtI2PSY,  [PdFPU1, PdFPSTO], 4, [2, 1]>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
+
+defm : PdWriteResXMMPair<WriteCvtI2SD,   [PdFPU1, PdFPSTO], 4, [], 2>;
+// FIXME: .Folded version is one NumMicroOp *less*..
+
+def WriteCVTSI642SDrr : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 13;
+  let NumMicroOps = 2;
+}
+def : InstRW<[WriteCVTSI642SDrr], (instrs CVTSI642SDrr, CVTSI642SSrr)>;
+
+defm : PdWriteResXMMPair<WriteCvtI2PD,   [PdFPU1, PdFPSTO], 8, [],     2>;
+defm : PdWriteResYMMPair<WriteCvtI2PDY,  [PdFPU1, PdFPSTO], 8, [2, 1], 4, 1>;
+defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
+
+defm : PdWriteResXMMPair<WriteCvtSS2SD,  [PdFPU1, PdFPSTO], 4>;
+
+defm : PdWriteResXMMPair<WriteCvtPS2PD,  [PdFPU1, PdFPSTO], 8, [],     2>;
+defm : PdWriteResYMMPair<WriteCvtPS2PDY, [PdFPU1, PdFPSTO], 8, [2, 1], 4, 1>;
+defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
+
+defm : PdWriteResXMMPair<WriteCvtSD2SS,  [PdFPU1, PdFPSTO], 4>;
+
+defm : PdWriteResXMMPair<WriteCvtPD2PS,  [PdFPU1, PdFPSTO],          8, [],        2>;
+defm : PdWriteResYMMPair<WriteCvtPD2PSY, [PdFPU1, PdFPSTO, PdFPFMA], 8, [2, 1, 1], 4>;
+defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
+
+def WriteMMX_CVTPD2PIirrMMX_CVTPI2PDirr : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+}
+def : InstRW<[WriteMMX_CVTPD2PIirrMMX_CVTPI2PDirr], (instrs MMX_CVTPD2PIirr,
+                                                            MMX_CVTPI2PDirr)>;
+
+def WriteMMX_CVTPI2PSirr : SchedWriteRes<[PdFPU1, PdFPSTO]> {
+  let Latency = 4;
+  let NumMicroOps = 2;
+}
+def : InstRW<[WriteMMX_CVTPI2PSirr], (instrs MMX_CVTPI2PSirr)>;
+
+defm : PdWriteResXMMPair<WriteCvtPH2PS,  [PdFPU1, PdFPSTO], 8, [],     2, 1>;
+defm : PdWriteResYMMPair<WriteCvtPH2PSY, [PdFPU1, PdFPSTO], 8, [2, 1], 4, 3>;
+defm : X86WriteResPairUnsupported<WriteCvtPH2PSZ>;
+
+defm : PdWriteRes<WriteCvtPS2PH,        [PdFPU1, PdFPSTO],          8, [],        2>;
+defm : PdWriteRes<WriteCvtPS2PHY,       [PdFPU1, PdFPSTO, PdFPFMA], 8, [2, 1, 1], 4>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
+
+defm : PdWriteRes<WriteCvtPS2PHSt,      [PdFPU1, PdFPSTO, PdStore],          4, [],           3>;
+defm : PdWriteRes<WriteCvtPS2PHYSt,     [PdFPU1, PdFPSTO, PdFPFMA, PdStore], 4, [2, 1, 1, 1], 4>;
+defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Vector integer operations.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteRes<WriteVecLoad,             [PdLoad, PdFPU01, PdFPMAL], 5>;
+defm : PdWriteRes<WriteVecLoadX,            [PdLoad, PdFPU01, PdFPMAL], 5>;
+defm : PdWriteRes<WriteVecLoadY,            [PdLoad, PdFPU01, PdFPMAL], 5, [], 2>;
+
+defm : PdWriteRes<WriteVecLoadNT,           [PdLoad, PdFPU01, PdFPMAL], 5>;
+defm : PdWriteRes<WriteVecLoadNTY,          [PdLoad, PdFPU01, PdFPMAL], 5>;
+
+defm : PdWriteRes<WriteVecMaskedLoad,       [PdLoad, PdFPU01, PdFPMAL], 6, [1, 1, 2]>;
+defm : PdWriteRes<WriteVecMaskedLoadY,      [PdLoad, PdFPU01, PdFPMAL], 6, [2, 2, 4], 2>;
+
+defm : PdWriteRes<WriteVecStore,            [PdStore, PdFPU1,   PdFPSTO], 2>;
+defm : PdWriteRes<WriteVecStoreX,           [PdStore, PdFPU1,   PdFPSTO]>;
+defm : PdWriteRes<WriteVecStoreY,           [PdStore, PdFPU1,   PdFPSTO], 1, [], 4>;
+
+def PdWriteVMOVDQUYmr : SchedWriteRes<[PdStore, PdFPU1,   PdFPSTO]> {
+  let NumMicroOps = 8;
+}
+def : InstRW<[PdWriteVMOVDQUYmr], (instrs VMOVDQUYmr)>;
+
+defm : PdWriteRes<WriteVecStoreNT,          [PdStore, PdFPU1,   PdFPSTO], 2>;
+defm : PdWriteRes<WriteVecStoreNTY,         [PdStore, PdFPU1,   PdFPSTO], 2, [2, 2, 2], 4>;
+
+defm : PdWriteRes<WriteVecMaskedStore,      [PdStore, PdFPU01, PdFPMAL], 6, [1, 1, 4]>;
+defm : PdWriteRes<WriteVecMaskedStoreY,     [PdStore, PdFPU01, PdFPMAL], 6, [2, 2, 4], 2>;
+
+defm : PdWriteRes<WriteVecMove,             [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteRes<WriteVecMoveX,            [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteRes<WriteVecMoveY,            [PdFPU01, PdFPMAL], 2, [2, 2], 2>;
+
+defm : PdWriteRes<WriteVecMoveToGpr,        [PdFPU0, PdFPFMA, PdEX0], 10>;
+defm : PdWriteRes<WriteVecMoveFromGpr,      [PdFPU01, PdFPFMA], 10, [], 2>;
+
+defm : PdWriteResXMMPair<WriteVecALU,        [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteResXMMPair<WriteVecALUX,       [PdFPU01, PdFPMAL], 2>;
+defm : X86WriteResPairUnsupported<WriteVecALUY>;
+defm : X86WriteResPairUnsupported<WriteVecALUZ>;
+
+defm : PdWriteResXMMPair<WriteVecShift,      [PdFPU01, PdFPMAL], 3>;
+defm : PdWriteResXMMPair<WriteVecShiftX,     [PdFPU01, PdFPMAL], 3>;
+defm : X86WriteResPairUnsupported<WriteVecShiftY>;
+defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
+
+defm : PdWriteResXMMPair<WriteVecShiftImm,   [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteResXMMPair<WriteVecShiftImmX,  [PdFPU01, PdFPMAL], 2>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmY>;
+defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
+
+defm : PdWriteResXMMPair<WriteVecIMul,       [PdFPU0, PdFPMMA], 4>;
+defm : PdWriteResXMMPair<WriteVecIMulX,      [PdFPU0, PdFPMMA], 4>;
+defm : X86WriteResPairUnsupported<WriteVecIMulY>;
+defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
+
+defm : PdWriteResXMMPair<WritePMULLD,        [PdFPU0, PdFPU01, PdFPMMA, PdFPMAL], 5, [2, 1, 2, 1]>;
+defm : X86WriteResPairUnsupported<WritePMULLDY>;
+defm : X86WriteResPairUnsupported<WritePMULLDZ>;
+
+def JWriteVPMACS : SchedWriteRes<[PdFPU0, PdFPU01, PdFPMMA, PdFPMAL]> {
+  let Latency = 4;
+  let ResourceCycles = [2, 1, 2, 1];
+}
+def : InstRW<[JWriteVPMACS], (instrs VPMACSDQHrr, VPMACSDQLrr, VPMACSSDQHrr,
+                                     VPMACSSDQLrr)>;
+
+defm : PdWriteResXMMPair<WriteMPSAD,         [PdFPU0, PdFPMMA], 9, [1, 2], 9>;
+defm : X86WriteResPairUnsupported<WriteMPSADY>;
+defm : X86WriteResPairUnsupported<WriteMPSADZ>;
+
+defm : PdWriteResXMMPair<WritePSADBW,        [PdFPU01, PdFPMAL], 4, [], 2>;
+defm : PdWriteResXMMPair<WritePSADBWX,       [PdFPU01, PdFPMAL], 4, [], 2>;
+defm : X86WriteResPairUnsupported<WritePSADBWY>;
+defm : X86WriteResPairUnsupported<WritePSADBWZ>;
+
+defm : PdWriteResXMMPair<WritePHMINPOS,      [PdFPU0,  PdFPMAL], 4, [], 2>;
+
+defm : PdWriteResXMMPair<WriteShuffle,       [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteResXMMPair<WriteShuffleX,      [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteResYMMPair<WriteShuffleY,      [PdFPU01, PdFPMAL], 2,   [1, 1]>;
+defm : X86WriteResPairUnsupported<WriteShuffleZ>;
+
+defm : PdWriteResXMMPair<WriteVarShuffle,    [PdFPU01, PdFPMAL], 3, [1, 4]>;
+defm : PdWriteResXMMPair<WriteVarShuffleX,   [PdFPU01, PdFPMAL], 3, [1, 4]>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleY>;
+defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
+
+defm : PdWriteResXMMPair<WriteBlend,         [PdFPU01, PdFPMAL], 2>;
+defm : X86WriteResPairUnsupported<WriteBlendY>;
+defm : X86WriteResPairUnsupported<WriteBlendZ>;
+
+defm : PdWriteResXMMPair<WriteVarBlend,      [PdFPU01, PdFPMAL], 2, [1, 4]>;
+defm : X86WriteResPairUnsupported<WriteVarBlendY>;
+defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
+
+defm : PdWriteResXMMPair<WriteVecLogic,      [PdFPU01, PdFPMAL], 2>;
+defm : PdWriteResXMMPair<WriteVecLogicX,     [PdFPU01, PdFPMAL], 2>;
+defm : X86WriteResPairUnsupported<WriteVecLogicY>;
+defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
+
+defm : PdWriteResXMMPair<WriteVecTest,       [PdFPU0, PdFPFMA, PdEX0],  1, [], 2>;
+defm : PdWriteResYMMPair<WriteVecTestY,      [PdFPU01, PdFPFMA, PdEX0], 1, [2, 2, 1], 4, 2>;
+defm : X86WriteResPairUnsupported<WriteVecTestZ>;
+
+defm : PdWriteResXMMPair<WriteShuffle256,    [PdFPU01, PdFPMAL]>;
+defm : PdWriteResXMMPair<WriteVarShuffle256, [PdFPU01, PdFPMAL]>;
+
+defm : PdWriteResXMMPair<WriteVarVecShift,   [PdFPU01, PdFPMAL], 3>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftY>;
+defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Vector insert/extract operations.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteRes<WriteVecInsert,    [PdFPU01, PdFPMAL], 2, [], 2>;
+defm : PdWriteRes<WriteVecInsertLd,  [PdFPU01, PdFPMAL, PdLoad], 6, [], 2>;
+
+defm : PdWriteRes<WriteVecExtract,   [PdFPU0, PdFPFMA, PdEX0], 13, [], 2>;
+defm : PdWriteRes<WriteVecExtractSt, [PdFPU1, PdFPSTO, PdStore], 13, [], 2>;
+
+def PdWriteEXTRQ : SchedWriteRes<[PdFPU01, PdFPMAL]> {
+  let Latency = 3;
+}
+def : InstRW<[PdWriteEXTRQ], (instrs EXTRQ, EXTRQI)>;
+
+////////////////////////////////////////////////////////////////////////////////
+// SSE42 String instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResXMMPair<WritePCmpIStrI, [PdFPU1, PdFPFMA, PdEX0], 14, [1, 2, 1], 7, 1>;
+defm : PdWriteResXMMPair<WritePCmpIStrM, [PdFPU1, PdFPFMA, PdEX0],  6, [1, 2, 1], 7, 2>;
+
+defm : PdWriteResXMMPair<WritePCmpEStrI, [PdFPU1, PdStore, PdLoad, PdFPMAL, PdFPFMA, PdEX0], 15, [1, 2, 6, 4, 1, 1], 27, 1>;
+defm : PdWriteResXMMPair<WritePCmpEStrM, [PdFPU1, PdStore, PdLoad, PdFPMAL, PdFPFMA, PdEX0], 10, [1, 2, 6, 4, 1, 1], 27, 1>;
+
+////////////////////////////////////////////////////////////////////////////////
+// MOVMSK Instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteRes<WriteFMOVMSK, [PdFPU0, PdFPFMA, PdEX0],   10, [], 2>;
+
+defm : PdWriteRes<WriteVecMOVMSK, [PdFPU0, PdFPFMA, PdEX0], 13, [], 2>;
+defm : X86WriteResUnsupported<WriteVecMOVMSKY>;
+// defm : X86WriteResUnsupported<WriteVecMOVMSKZ>;
+
+defm : PdWriteRes<WriteMMXMOVMSK, [PdFPU0, PdFPFMA, PdEX0], 10, [], 2>;
+
+////////////////////////////////////////////////////////////////////////////////
+// AES Instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResXMMPair<WriteAESIMC,    [PdFPU0, PdFPMMA], 5>;
+defm : PdWriteResXMMPair<WriteAESKeyGen, [PdFPU0, PdFPMMA], 5>;
+defm : PdWriteResXMMPair<WriteAESDecEnc, [PdFPU0, PdFPMMA], 9, [], 2>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Horizontal add/sub  instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResXMMPair<WriteFHAdd,  [PdFPU0, PdFPFMA], 11, [],     3, 1>;
+defm : PdWriteResYMMPair<WriteFHAddY, [PdFPU0, PdFPFMA], 11, [2, 1], 8, 2>;
+defm : X86WriteResPairUnsupported<WriteFHAddZ>;
+
+defm : PdWriteResXMMPair<WritePHAdd,  [PdFPU01, PdFPMAL], 5, [], 3, 1>;
+defm : PdWriteResXMMPair<WritePHAddX, [PdFPU01, PdFPMAL], 2>;
+defm : X86WriteResPairUnsupported<WritePHAddY>;
+defm : X86WriteResPairUnsupported<WritePHAddZ>;
+
+def : InstRW<[WritePHAdd], (instrs PHADDDrr, PHSUBDrr,
+                                   PHADDWrr, PHSUBWrr,
+                                   PHADDSWrr, PHSUBSWrr,
+                                   VPHADDDrr, VPHSUBDrr,
+                                   VPHADDWrr, VPHSUBWrr,
+                                   VPHADDSWrr, VPHSUBSWrr)>;
+
+def : InstRW<[WritePHAdd.Folded], (instrs PHADDDrm, PHSUBDrm,
+                                          PHADDWrm, PHSUBWrm,
+                                          PHADDSWrm, PHSUBSWrm,
+                                          VPHADDDrm, VPHSUBDrm,
+                                          VPHADDWrm, VPHSUBWrm,
+                                          VPHADDSWrm, VPHSUBSWrm)>;
+
+////////////////////////////////////////////////////////////////////////////////
+// Carry-less multiplication instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+defm : PdWriteResXMMPair<WriteCLMul, [PdFPU0, PdFPMMA], 12, [], 5, 1>;
+
+def PdWriteVPCLMULQDQrr : SchedWriteRes<[PdFPU0, PdFPMMA]> {
+  let Latency = 13;
+  let NumMicroOps = 6;
+}
+def : InstRW<[PdWriteVPCLMULQDQrr], (instrs VPCLMULQDQrr)>;
+
+////////////////////////////////////////////////////////////////////////////////
+// SSE4A instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+def PdWriteINSERTQ : SchedWriteRes<[PdFPU01, PdFPMAL]> {
+  let Latency = 3;
+  let ResourceCycles = [1, 4];
+}
+def : InstRW<[PdWriteINSERTQ], (instrs INSERTQ, INSERTQI)>;
+
+////////////////////////////////////////////////////////////////////////////////
+// AVX instructions.
+////////////////////////////////////////////////////////////////////////////////
+
+def PdWriteVBROADCASTYLd : SchedWriteRes<[PdLoad, PdFPU01, PdFPFMA]> {
+  let Latency = 6;
+  let ResourceCycles = [1, 2, 4];
+  let NumMicroOps = 2;
+}
+def : InstRW<[PdWriteVBROADCASTYLd, ReadAfterLd], (instrs VBROADCASTSDYrm,
+                                                          VBROADCASTSSYrm)>;
+
+def PdWriteVZEROALL : SchedWriteRes<[]> {
+  let Latency = 90;
+  let NumMicroOps = 32;
+}
+def : InstRW<[PdWriteVZEROALL], (instrs VZEROALL)>;
+
+def PdWriteVZEROUPPER : SchedWriteRes<[]> {
+  let Latency = 46;
+  let NumMicroOps = 16;
+}
+def : InstRW<[PdWriteVZEROUPPER], (instrs VZEROUPPER)>;
+
+///////////////////////////////////////////////////////////////////////////////
+//  SchedWriteVariant definitions.
+///////////////////////////////////////////////////////////////////////////////
+
+def PdWriteZeroLatency : SchedWriteRes<[]> {
+  let Latency = 0;
+}
+
+def PdWriteZeroIdiom : SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [PdWriteZeroLatency]>,
+  SchedVar<MCSchedPredicate<TruePred>,           [WriteALU]>
+]>;
+def : InstRW<[PdWriteZeroIdiom], (instrs SUB32rr, SUB64rr,
+                                         XOR32rr, XOR64rr)>;
+
+def PdWriteFZeroIdiom : SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [PdWriteZeroLatency]>,
+  SchedVar<MCSchedPredicate<TruePred>,           [WriteFLogic]>
+]>;
+def : InstRW<[PdWriteFZeroIdiom], (instrs XORPSrr,  VXORPSrr,
+                                          XORPDrr,  VXORPDrr,
+                                          ANDNPSrr, VANDNPSrr,
+                                          ANDNPDrr, VANDNPDrr)>;
+
+// VXORPSYrr, VXORPDYrr, VANDNPSYrr, VANDNPDYrr "zero-idioms" have latency of 1.
+
+def PdWriteVZeroIdiomLogic : SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [PdWriteZeroLatency]>,
+  SchedVar<MCSchedPredicate<TruePred>,           [WriteVecLogic]>
+]>;
+def : InstRW<[PdWriteVZeroIdiomLogic], (instrs MMX_PXORirr, MMX_PANDNirr)>;
+
+def PdWriteVZeroIdiomLogicX : SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [PdWriteZeroLatency]>,
+  SchedVar<MCSchedPredicate<TruePred>,           [WriteVecLogicX]>
+]>;
+def : InstRW<[PdWriteVZeroIdiomLogicX], (instrs PXORrr,  VPXORrr,
+                                                PANDNrr, VPANDNrr)>;
+
+def PdWriteVZeroIdiomALU : SchedWriteVariant<[
+  SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [PdWriteZeroLatency]>,
+  SchedVar<MCSchedPredicate<TruePred>,           [WriteVecALU]>
+]>;
+def : InstRW<[PdWriteVZeroIdiomALU], (instrs MMX_PSUBBirr,   MMX_PSUBDirr,
+                                             MMX_PSUBQirr,   MMX_PSUBWirr,
+                                             MMX_PCMPGTBirr,
+                                             MMX_PCMPGTDirr,
+                                             MMX_PCMPGTWirr)>;
+
+def PdWriteVZeroIdiomALUX : SchedWriteVariant<[
+    SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [PdWriteZeroLatency]>,
+    SchedVar<MCSchedPredicate<TruePred>,           [WriteVecALUX]>
+]>;
+def : InstRW<[PdWriteVZeroIdiomALUX], (instrs PSUBBrr, VPSUBBrr,
+                                              PSUBDrr, VPSUBDrr,
+                                              PSUBQrr, VPSUBQrr,
+                                              PSUBWrr, VPSUBWrr,
+                                              PCMPGTBrr, VPCMPGTBrr,
+                                              PCMPGTDrr, VPCMPGTDrr,
+                                              PCMPGTWrr, VPCMPGTWrr)>;
+
+///////////////////////////////////////////////////////////////////////////////
+// Dependency breaking instructions.
+///////////////////////////////////////////////////////////////////////////////
+
+// VPCMPGTQ, but not PCMPGTQ!
+
+def : IsZeroIdiomFunction<[
+  // GPR Zero-idioms.
+  DepBreakingClass<[ SUB32rr, SUB64rr, XOR32rr, XOR64rr ], ZeroIdiomPredicate>,
+
+  // MMX Zero-idioms.
+  DepBreakingClass<[
+    MMX_PXORirr, MMX_PANDNirr, MMX_PSUBBirr,
+    MMX_PSUBDirr, MMX_PSUBQirr, MMX_PSUBWirr,
+    MMX_PSUBSBirr, MMX_PSUBSWirr, MMX_PSUBUSBirr, MMX_PSUBUSWirr,
+    MMX_PCMPGTBirr, MMX_PCMPGTDirr, MMX_PCMPGTWirr
+  ], ZeroIdiomPredicate>,
+
+  // SSE Zero-idioms.
+  DepBreakingClass<[
+    // fp variants.
+    XORPSrr, XORPDrr, ANDNPSrr, ANDNPDrr,
+
+    // int variants.
+    PXORrr, PANDNrr,
+    PSUBBrr, PSUBWrr, PSUBDrr, PSUBQrr,
+    PSUBSBrr, PSUBSWrr, PSUBUSBrr, PSUBUSWrr,
+    PCMPGTBrr, PCMPGTDrr, PCMPGTWrr
+  ], ZeroIdiomPredicate>,
+
+  // AVX Zero-idioms.
+  DepBreakingClass<[
+    // xmm fp variants.
+    VXORPSrr, VXORPDrr, VANDNPSrr, VANDNPDrr,
+
+    // xmm int variants.
+    VPXORrr, VPANDNrr,
+    VPSUBBrr, VPSUBWrr, VPSUBDrr, VPSUBQrr,
+    VPSUBSBrr, VPSUBSWrr, VPSUBUSBrr, VPSUBUSWrr,
+    VPCMPGTBrr, VPCMPGTWrr, VPCMPGTDrr, VPCMPGTQrr,
+
+    // ymm variants.
+    VXORPSYrr, VXORPDYrr, VANDNPSYrr, VANDNPDYrr
+  ], ZeroIdiomPredicate>
+]>;
+
+def : IsDepBreakingFunction<[
+  // GPR
+  DepBreakingClass<[ SBB32rr, SBB64rr ], ZeroIdiomPredicate>,
+  DepBreakingClass<[ CMP32rr, CMP64rr ], CheckSameRegOperand<0, 1> >,
+
+  // MMX
+  DepBreakingClass<[
+    MMX_PCMPEQBirr, MMX_PCMPEQDirr, MMX_PCMPEQWirr
+  ], ZeroIdiomPredicate>,
+
+  // SSE
+  DepBreakingClass<[
+    PCMPEQBrr, PCMPEQWrr, PCMPEQDrr
+    // But not PCMPEQQrr.
+  ], ZeroIdiomPredicate>,
+
+  // AVX
+  DepBreakingClass<[
+    VPCMPEQBrr, VPCMPEQWrr, VPCMPEQDrr
+    // But not VPCMPEQQrr.
+  ], ZeroIdiomPredicate>
+]>;
+
+
+} // SchedModel
Index: llvm/trunk/test/CodeGen/X86/aes-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/aes-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/aes-schedule.ll
@@ -14,8 +14,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=+aes,-avx2 | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=+aes,-avx  | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=+aes,-avx2 | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+aes,-avx  | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=+aes,-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=+aes,-avx  | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=+aes,-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=+aes,-avx  | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=+aes,-avx2 | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=+aes,-avx  | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
@@ -96,15 +96,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_aesdec:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    aesdec %xmm1, %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    aesdec (%rdi), %xmm0 # sched: [13:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    aesdec %xmm1, %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    aesdec (%rdi), %xmm0 # sched: [14:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_aesdec:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaesdec %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaesdec (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaesdec %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    vaesdec (%rdi), %xmm0, %xmm0 # sched: [14:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_aesdec:
 ; BTVER2-SSE:       # %bb.0:
@@ -211,15 +211,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_aesdeclast:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    aesdeclast %xmm1, %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    aesdeclast (%rdi), %xmm0 # sched: [13:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    aesdeclast %xmm1, %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    aesdeclast (%rdi), %xmm0 # sched: [14:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_aesdeclast:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaesdeclast %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaesdeclast (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaesdeclast %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    vaesdeclast (%rdi), %xmm0, %xmm0 # sched: [14:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_aesdeclast:
 ; BTVER2-SSE:       # %bb.0:
@@ -326,15 +326,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_aesenc:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    aesenc %xmm1, %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    aesenc (%rdi), %xmm0 # sched: [13:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    aesenc %xmm1, %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    aesenc (%rdi), %xmm0 # sched: [14:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_aesenc:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaesenc %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaesenc (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaesenc %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    vaesenc (%rdi), %xmm0, %xmm0 # sched: [14:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_aesenc:
 ; BTVER2-SSE:       # %bb.0:
@@ -441,15 +441,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_aesenclast:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    aesenclast %xmm1, %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    aesenclast (%rdi), %xmm0 # sched: [13:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    aesenclast %xmm1, %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    aesenclast (%rdi), %xmm0 # sched: [14:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_aesenclast:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaesenclast %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaesenclast (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaesenclast %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    vaesenclast (%rdi), %xmm0, %xmm0 # sched: [14:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_aesenclast:
 ; BTVER2-SSE:       # %bb.0:
@@ -569,17 +569,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_aesimc:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    aesimc %xmm0, %xmm1 # sched: [12:2.00]
-; BDVER2-SSE-NEXT:    aesimc (%rdi), %xmm0 # sched: [18:2.00]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    aesimc %xmm0, %xmm1 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    aesimc (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_aesimc:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaesimc %xmm0, %xmm0 # sched: [12:2.00]
-; BDVER2-NEXT:    vaesimc (%rdi), %xmm1 # sched: [18:2.00]
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaesimc (%rdi), %xmm1 # sched: [10:1.00]
+; BDVER2-NEXT:    vaesimc %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_aesimc:
 ; BTVER2-SSE:       # %bb.0:
@@ -703,17 +703,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_aeskeygenassist:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    aeskeygenassist $7, %xmm0, %xmm1 # sched: [8:3.67]
-; BDVER2-SSE-NEXT:    aeskeygenassist $7, (%rdi), %xmm0 # sched: [8:3.33]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    aeskeygenassist $7, %xmm0, %xmm1 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    aeskeygenassist $7, (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_aeskeygenassist:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaeskeygenassist $7, %xmm0, %xmm0 # sched: [8:3.67]
-; BDVER2-NEXT:    vaeskeygenassist $7, (%rdi), %xmm1 # sched: [8:3.33]
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaeskeygenassist $7, (%rdi), %xmm1 # sched: [10:1.00]
+; BDVER2-NEXT:    vaeskeygenassist $7, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_aeskeygenassist:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/avx-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/avx-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/avx-schedule.ll
@@ -6,7 +6,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -49,9 +49,9 @@
 ;
 ; BDVER2-LABEL: test_addpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vaddpd (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_addpd:
 ; BTVER2:       # %bb.0:
@@ -109,9 +109,9 @@
 ;
 ; BDVER2-LABEL: test_addps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vaddps (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_addps:
 ; BTVER2:       # %bb.0:
@@ -169,9 +169,9 @@
 ;
 ; BDVER2-LABEL: test_addsubpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddsubpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsubpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vaddsubpd (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_addsubpd:
 ; BTVER2:       # %bb.0:
@@ -230,9 +230,9 @@
 ;
 ; BDVER2-LABEL: test_addsubps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddsubps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsubps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vaddsubps (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_addsubps:
 ; BTVER2:       # %bb.0:
@@ -297,10 +297,10 @@
 ;
 ; BDVER2-LABEL: test_andnotpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandnpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandnpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandnpd %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vandnpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_andnotpd:
 ; BTVER2:       # %bb.0:
@@ -373,10 +373,10 @@
 ;
 ; BDVER2-LABEL: test_andnotps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandnps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandnps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandnps %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vandnps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_andnotps:
 ; BTVER2:       # %bb.0:
@@ -449,10 +449,10 @@
 ;
 ; BDVER2-LABEL: test_andpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandpd %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vandpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_andpd:
 ; BTVER2:       # %bb.0:
@@ -523,10 +523,10 @@
 ;
 ; BDVER2-LABEL: test_andps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandps %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vandps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_andps:
 ; BTVER2:       # %bb.0:
@@ -597,10 +597,10 @@
 ;
 ; BDVER2-LABEL: test_blendpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3] sched: [1:0.50]
-; BDVER2-NEXT:    vblendpd {{.*#+}} ymm1 = ymm0[0,1],mem[2,3] sched: [8:0.50]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3] sched: [2:1.00]
+; BDVER2-NEXT:    vblendpd {{.*#+}} ymm1 = ymm0[0,1],mem[2,3] sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blendpd:
 ; BTVER2:       # %bb.0:
@@ -667,10 +667,10 @@
 ;
 ; BDVER2-LABEL: test_blendps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3,4,5,6,7] sched: [1:0.50]
-; BDVER2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],mem[2],ymm1[3],mem[4,5,6],ymm1[7] sched: [8:0.50]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3,4,5,6,7] sched: [2:1.00]
+; BDVER2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],mem[2],ymm1[3],mem[4,5,6],ymm1[7] sched: [7:1.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blendps:
 ; BTVER2:       # %bb.0:
@@ -731,9 +731,9 @@
 ;
 ; BDVER2-LABEL: test_blendvpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vblendvpd %ymm2, (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:3.00]
+; BDVER2-NEXT:    vblendvpd %ymm2, (%rdi), %ymm0, %ymm0 # sched: [7:3.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blendvpd:
 ; BTVER2:       # %bb.0:
@@ -792,9 +792,9 @@
 ;
 ; BDVER2-LABEL: test_blendvps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vblendvps %ymm2, (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:3.00]
+; BDVER2-NEXT:    vblendvps %ymm2, (%rdi), %ymm0, %ymm0 # sched: [7:3.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blendvps:
 ; BTVER2:       # %bb.0:
@@ -847,8 +847,8 @@
 ;
 ; BDVER2-LABEL: test_broadcastf128:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_broadcastf128:
 ; BTVER2:       # %bb.0:
@@ -897,8 +897,8 @@
 ;
 ; BDVER2-LABEL: test_broadcastsd_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vbroadcastsd (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vbroadcastsd (%rdi), %ymm0 # sched: [6:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_broadcastsd_ymm:
 ; BTVER2:       # %bb.0:
@@ -948,8 +948,8 @@
 ;
 ; BDVER2-LABEL: test_broadcastss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vbroadcastss (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vbroadcastss (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_broadcastss:
 ; BTVER2:       # %bb.0:
@@ -999,8 +999,8 @@
 ;
 ; BDVER2-LABEL: test_broadcastss_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vbroadcastss (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vbroadcastss (%rdi), %ymm0 # sched: [6:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_broadcastss_ymm:
 ; BTVER2:       # %bb.0:
@@ -1062,10 +1062,10 @@
 ;
 ; BDVER2-LABEL: test_cmppd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm1 # sched: [3:1.00]
-; BDVER2-NEXT:    vcmpeqpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    vorpd %ymm0, %ymm1, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm1 # sched: [2:2.00]
+; BDVER2-NEXT:    vcmpeqpd (%rdi), %ymm0, %ymm0 # sched: [7:2.00]
+; BDVER2-NEXT:    vorpd %ymm0, %ymm1, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmppd:
 ; BTVER2:       # %bb.0:
@@ -1135,10 +1135,10 @@
 ;
 ; BDVER2-LABEL: test_cmpps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcmpeqps %ymm1, %ymm0, %ymm1 # sched: [3:1.00]
-; BDVER2-NEXT:    vcmpeqps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    vorps %ymm0, %ymm1, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcmpeqps %ymm1, %ymm0, %ymm1 # sched: [2:2.00]
+; BDVER2-NEXT:    vcmpeqps (%rdi), %ymm0, %ymm0 # sched: [7:2.00]
+; BDVER2-NEXT:    vorps %ymm0, %ymm1, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmpps:
 ; BTVER2:       # %bb.0:
@@ -1208,10 +1208,10 @@
 ;
 ; BDVER2-LABEL: test_cvtdq2pd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtdq2pd %xmm0, %ymm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvtdq2pd (%rdi), %ymm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtdq2pd (%rdi), %ymm1 # sched: [13:2.00]
+; BDVER2-NEXT:    vcvtdq2pd %xmm0, %ymm0 # sched: [8:2.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtdq2pd:
 ; BTVER2:       # %bb.0:
@@ -1280,10 +1280,10 @@
 ;
 ; BDVER2-LABEL: test_cvtdq2ps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtdq2ps %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vcvtdq2ps (%rdi), %ymm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtdq2ps (%rdi), %ymm1 # sched: [9:2.00]
+; BDVER2-NEXT:    vcvtdq2ps %ymm0, %ymm0 # sched: [4:2.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtdq2ps:
 ; BTVER2:       # %bb.0:
@@ -1350,10 +1350,10 @@
 ;
 ; BDVER2-LABEL: test_cvtpd2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtpd2dq %ymm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvtpd2dqy (%rdi), %xmm1 # sched: [11:1.00]
-; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtpd2dqy (%rdi), %xmm1 # sched: [13:2.00]
+; BDVER2-NEXT:    vcvtpd2dq %ymm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtpd2dq:
 ; BTVER2:       # %bb.0:
@@ -1421,10 +1421,10 @@
 ;
 ; BDVER2-LABEL: test_cvttpd2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttpd2dq %ymm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvttpd2dqy (%rdi), %xmm1 # sched: [11:1.00]
-; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttpd2dqy (%rdi), %xmm1 # sched: [13:2.00]
+; BDVER2-NEXT:    vcvttpd2dq %ymm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvttpd2dq:
 ; BTVER2:       # %bb.0:
@@ -1491,10 +1491,10 @@
 ;
 ; BDVER2-LABEL: test_cvtpd2ps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtpd2ps %ymm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvtpd2psy (%rdi), %xmm1 # sched: [11:1.00]
-; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtpd2psy (%rdi), %xmm1 # sched: [13:2.00]
+; BDVER2-NEXT:    vcvtpd2ps %ymm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtpd2ps:
 ; BTVER2:       # %bb.0:
@@ -1561,10 +1561,10 @@
 ;
 ; BDVER2-LABEL: test_cvtps2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtps2dq %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vcvtps2dq (%rdi), %ymm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtps2dq (%rdi), %ymm1 # sched: [9:2.00]
+; BDVER2-NEXT:    vcvtps2dq %ymm0, %ymm0 # sched: [4:2.00]
+; BDVER2-NEXT:    vorpd %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtps2dq:
 ; BTVER2:       # %bb.0:
@@ -1632,10 +1632,10 @@
 ;
 ; BDVER2-LABEL: test_cvttps2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttps2dq %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vcvttps2dq (%rdi), %ymm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttps2dq (%rdi), %ymm1 # sched: [9:2.00]
+; BDVER2-NEXT:    vcvttps2dq %ymm0, %ymm0 # sched: [4:2.00]
+; BDVER2-NEXT:    vorps %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvttps2dq:
 ; BTVER2:       # %bb.0:
@@ -1696,9 +1696,9 @@
 ;
 ; BDVER2-LABEL: test_divpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdivpd %ymm1, %ymm0, %ymm0 # sched: [45:44.00]
-; BDVER2-NEXT:    vdivpd (%rdi), %ymm0, %ymm0 # sched: [52:44.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdivpd %ymm1, %ymm0, %ymm0 # sched: [9:19.00]
+; BDVER2-NEXT:    vdivpd (%rdi), %ymm0, %ymm0 # sched: [14:19.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_divpd:
 ; BTVER2:       # %bb.0:
@@ -1756,9 +1756,9 @@
 ;
 ; BDVER2-LABEL: test_divps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdivps %ymm1, %ymm0, %ymm0 # sched: [29:28.00]
-; BDVER2-NEXT:    vdivps (%rdi), %ymm0, %ymm0 # sched: [36:28.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdivps %ymm1, %ymm0, %ymm0 # sched: [9:19.00]
+; BDVER2-NEXT:    vdivps (%rdi), %ymm0, %ymm0 # sched: [14:19.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_divps:
 ; BTVER2:       # %bb.0:
@@ -1816,9 +1816,9 @@
 ;
 ; BDVER2-LABEL: test_dpps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [12:2.00]
-; BDVER2-NEXT:    vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [19:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [27:3.00]
+; BDVER2-NEXT:    vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [32:3.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dpps:
 ; BTVER2:       # %bb.0:
@@ -1883,10 +1883,10 @@
 ;
 ; BDVER2-LABEL: test_extractf128:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [7:0.50]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_extractf128:
 ; BTVER2:       # %bb.0:
@@ -1945,9 +1945,9 @@
 ;
 ; BDVER2-LABEL: test_haddpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhaddpd (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0 # sched: [11:2.00]
+; BDVER2-NEXT:    vhaddpd (%rdi), %ymm0, %ymm0 # sched: [16:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_haddpd:
 ; BTVER2:       # %bb.0:
@@ -2006,9 +2006,9 @@
 ;
 ; BDVER2-LABEL: test_haddps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhaddps (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhaddps %ymm1, %ymm0, %ymm0 # sched: [11:2.00]
+; BDVER2-NEXT:    vhaddps (%rdi), %ymm0, %ymm0 # sched: [16:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_haddps:
 ; BTVER2:       # %bb.0:
@@ -2067,9 +2067,9 @@
 ;
 ; BDVER2-LABEL: test_hsubpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhsubpd (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0 # sched: [11:2.00]
+; BDVER2-NEXT:    vhsubpd (%rdi), %ymm0, %ymm0 # sched: [16:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_hsubpd:
 ; BTVER2:       # %bb.0:
@@ -2128,9 +2128,9 @@
 ;
 ; BDVER2-LABEL: test_hsubps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhsubps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhsubps (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhsubps %ymm1, %ymm0, %ymm0 # sched: [11:2.00]
+; BDVER2-NEXT:    vhsubps (%rdi), %ymm0, %ymm0 # sched: [16:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_hsubps:
 ; BTVER2:       # %bb.0:
@@ -2195,10 +2195,10 @@
 ;
 ; BDVER2-LABEL: test_insertf128:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1 # sched: [1:1.00]
+; BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vinsertf128 $1, (%rdi), %ymm0, %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_insertf128:
 ; BTVER2:       # %bb.0:
@@ -2255,8 +2255,8 @@
 ;
 ; BDVER2-LABEL: test_lddqu:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vlddqu (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vlddqu (%rdi), %ymm0 # sched: [5:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lddqu:
 ; BTVER2:       # %bb.0:
@@ -2317,10 +2317,10 @@
 ;
 ; BDVER2-LABEL: test_maskmovpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaskmovpd (%rdi), %xmm0, %xmm2 # sched: [8:1.00]
-; BDVER2-NEXT:    vmaskmovpd %xmm1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    vmovapd %xmm2, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaskmovpd (%rdi), %xmm0, %xmm2 # sched: [6:1.00]
+; BDVER2-NEXT:    vmaskmovpd %xmm1, %xmm0, (%rdi) # sched: [6:2.00]
+; BDVER2-NEXT:    vmovapd %xmm2, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maskmovpd:
 ; BTVER2:       # %bb.0:
@@ -2387,10 +2387,10 @@
 ;
 ; BDVER2-LABEL: test_maskmovpd_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaskmovpd (%rdi), %ymm0, %ymm2 # sched: [9:1.00]
-; BDVER2-NEXT:    vmaskmovpd %ymm1, %ymm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    vmovapd %ymm2, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaskmovpd (%rdi), %ymm0, %ymm2 # sched: [6:2.00]
+; BDVER2-NEXT:    vmaskmovpd %ymm1, %ymm0, (%rdi) # sched: [6:2.00]
+; BDVER2-NEXT:    vmovapd %ymm2, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maskmovpd_ymm:
 ; BTVER2:       # %bb.0:
@@ -2457,10 +2457,10 @@
 ;
 ; BDVER2-LABEL: test_maskmovps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaskmovps (%rdi), %xmm0, %xmm2 # sched: [8:1.00]
-; BDVER2-NEXT:    vmaskmovps %xmm1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    vmovaps %xmm2, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaskmovps (%rdi), %xmm0, %xmm2 # sched: [6:1.00]
+; BDVER2-NEXT:    vmaskmovps %xmm1, %xmm0, (%rdi) # sched: [6:2.00]
+; BDVER2-NEXT:    vmovaps %xmm2, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maskmovps:
 ; BTVER2:       # %bb.0:
@@ -2527,10 +2527,10 @@
 ;
 ; BDVER2-LABEL: test_maskmovps_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaskmovps (%rdi), %ymm0, %ymm2 # sched: [9:1.00]
-; BDVER2-NEXT:    vmaskmovps %ymm1, %ymm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    vmovaps %ymm2, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaskmovps (%rdi), %ymm0, %ymm2 # sched: [6:2.00]
+; BDVER2-NEXT:    vmaskmovps %ymm1, %ymm0, (%rdi) # sched: [6:2.00]
+; BDVER2-NEXT:    vmovaps %ymm2, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maskmovps_ymm:
 ; BTVER2:       # %bb.0:
@@ -2591,9 +2591,9 @@
 ;
 ; BDVER2-LABEL: test_maxpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaxpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmaxpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaxpd %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vmaxpd (%rdi), %ymm0, %ymm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maxpd:
 ; BTVER2:       # %bb.0:
@@ -2652,9 +2652,9 @@
 ;
 ; BDVER2-LABEL: test_maxps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaxps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmaxps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaxps %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vmaxps (%rdi), %ymm0, %ymm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maxps:
 ; BTVER2:       # %bb.0:
@@ -2713,9 +2713,9 @@
 ;
 ; BDVER2-LABEL: test_minpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vminpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vminpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vminpd %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vminpd (%rdi), %ymm0, %ymm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_minpd:
 ; BTVER2:       # %bb.0:
@@ -2774,9 +2774,9 @@
 ;
 ; BDVER2-LABEL: test_minps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vminps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vminps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vminps %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vminps (%rdi), %ymm0, %ymm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_minps:
 ; BTVER2:       # %bb.0:
@@ -2841,10 +2841,10 @@
 ;
 ; BDVER2-LABEL: test_movapd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovapd (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovapd (%rdi), %ymm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm0, %ymm0 # sched: [5:2.00]
 ; BDVER2-NEXT:    vmovapd %ymm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movapd:
 ; BTVER2:       # %bb.0:
@@ -2910,10 +2910,10 @@
 ;
 ; BDVER2-LABEL: test_movaps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovaps (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovaps (%rdi), %ymm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm0, %ymm0 # sched: [5:2.00]
 ; BDVER2-NEXT:    vmovaps %ymm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movaps:
 ; BTVER2:       # %bb.0:
@@ -2979,10 +2979,10 @@
 ;
 ; BDVER2-LABEL: test_movddup:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2] sched: [1:1.00]
-; BDVER2-NEXT:    vmovddup {{.*#+}} ymm1 = mem[0,0,2,2] sched: [7:0.50]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovddup {{.*#+}} ymm1 = mem[0,0,2,2] sched: [7:1.00]
+; BDVER2-NEXT:    vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2] sched: [2:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movddup:
 ; BTVER2:       # %bb.0:
@@ -3043,9 +3043,9 @@
 ;
 ; BDVER2-LABEL: test_movmskpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovmskpd %ymm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovmskpd %ymm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movmskpd:
 ; BTVER2:       # %bb.0:
@@ -3101,9 +3101,9 @@
 ;
 ; BDVER2-LABEL: test_movmskps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovmskps %ymm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovmskps %ymm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movmskps:
 ; BTVER2:       # %bb.0:
@@ -3172,10 +3172,10 @@
 ; BDVER2-LABEL: test_movntdq:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
+; BDVER2-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [2:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movntdq:
 ; BTVER2:       # %bb.0:
@@ -3234,9 +3234,9 @@
 ;
 ; BDVER2-LABEL: test_movntpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovntpd %ymm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovntpd %ymm0, (%rdi) # sched: [3:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movntpd:
 ; BTVER2:       # %bb.0:
@@ -3293,9 +3293,9 @@
 ;
 ; BDVER2-LABEL: test_movntps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovntps %ymm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovntps %ymm0, (%rdi) # sched: [3:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movntps:
 ; BTVER2:       # %bb.0:
@@ -3358,10 +3358,10 @@
 ;
 ; BDVER2-LABEL: test_movshdup:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7] sched: [1:1.00]
-; BDVER2-NEXT:    vmovshdup {{.*#+}} ymm1 = mem[1,1,3,3,5,5,7,7] sched: [7:0.50]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovshdup {{.*#+}} ymm1 = mem[1,1,3,3,5,5,7,7] sched: [7:1.00]
+; BDVER2-NEXT:    vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7] sched: [2:1.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movshdup:
 ; BTVER2:       # %bb.0:
@@ -3428,10 +3428,10 @@
 ;
 ; BDVER2-LABEL: test_movsldup:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovsldup {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6] sched: [1:1.00]
-; BDVER2-NEXT:    vmovsldup {{.*#+}} ymm1 = mem[0,0,2,2,4,4,6,6] sched: [7:0.50]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovsldup {{.*#+}} ymm1 = mem[0,0,2,2,4,4,6,6] sched: [7:1.00]
+; BDVER2-NEXT:    vmovsldup {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6] sched: [2:1.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movsldup:
 ; BTVER2:       # %bb.0:
@@ -3500,10 +3500,10 @@
 ;
 ; BDVER2-LABEL: test_movupd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovupd (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovupd (%rdi), %ymm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm0, %ymm0 # sched: [5:2.00]
 ; BDVER2-NEXT:    vmovupd %ymm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movupd:
 ; BTVER2:       # %bb.0:
@@ -3571,10 +3571,10 @@
 ;
 ; BDVER2-LABEL: test_movups:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovups (%rdi), %ymm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovups (%rdi), %ymm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm0, %ymm0 # sched: [5:2.00]
 ; BDVER2-NEXT:    vmovups %ymm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movups:
 ; BTVER2:       # %bb.0:
@@ -3634,9 +3634,9 @@
 ;
 ; BDVER2-LABEL: test_mulpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmulpd %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulpd (%rdi), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmulpd (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_mulpd:
 ; BTVER2:       # %bb.0:
@@ -3694,9 +3694,9 @@
 ;
 ; BDVER2-LABEL: test_mulps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmulps %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulps (%rdi), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmulps (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_mulps:
 ; BTVER2:       # %bb.0:
@@ -3760,10 +3760,10 @@
 ;
 ; BDVER2-LABEL: orpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vorpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vorpd %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vorpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: orpd:
 ; BTVER2:       # %bb.0:
@@ -3834,10 +3834,10 @@
 ;
 ; BDVER2-LABEL: test_orps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vorps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vorps %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vorps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_orps:
 ; BTVER2:       # %bb.0:
@@ -3908,10 +3908,10 @@
 ;
 ; BDVER2-LABEL: test_perm2f128:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3],ymm1[0,1] sched: [1:1.00]
-; BDVER2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[0,1] sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3],ymm1[0,1] sched: [4:0.50]
+; BDVER2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[0,1] sched: [8:0.50]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_perm2f128:
 ; BTVER2:       # %bb.0:
@@ -3978,10 +3978,10 @@
 ;
 ; BDVER2-LABEL: test_permilpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0] sched: [1:1.00]
-; BDVER2-NEXT:    vpermilpd {{.*#+}} xmm1 = mem[1,0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilpd {{.*#+}} xmm1 = mem[1,0] sched: [7:0.50]
+; BDVER2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0] sched: [2:0.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilpd:
 ; BTVER2:       # %bb.0:
@@ -4048,10 +4048,10 @@
 ;
 ; BDVER2-LABEL: test_permilpd_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3] sched: [1:1.00]
-; BDVER2-NEXT:    vpermilpd {{.*#+}} ymm1 = mem[1,0,2,3] sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilpd {{.*#+}} ymm1 = mem[1,0,2,3] sched: [7:1.00]
+; BDVER2-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3] sched: [2:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilpd_ymm:
 ; BTVER2:       # %bb.0:
@@ -4118,10 +4118,10 @@
 ;
 ; BDVER2-LABEL: test_permilps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0] sched: [1:1.00]
-; BDVER2-NEXT:    vpermilps {{.*#+}} xmm1 = mem[3,2,1,0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilps {{.*#+}} xmm1 = mem[3,2,1,0] sched: [7:0.50]
+; BDVER2-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0] sched: [2:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilps:
 ; BTVER2:       # %bb.0:
@@ -4188,10 +4188,10 @@
 ;
 ; BDVER2-LABEL: test_permilps_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4] sched: [1:1.00]
-; BDVER2-NEXT:    vpermilps {{.*#+}} ymm1 = mem[3,2,1,0,7,6,5,4] sched: [8:1.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilps {{.*#+}} ymm1 = mem[3,2,1,0,7,6,5,4] sched: [7:1.00]
+; BDVER2-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4] sched: [2:1.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilps_ymm:
 ; BTVER2:       # %bb.0:
@@ -4252,9 +4252,9 @@
 ;
 ; BDVER2-LABEL: test_permilvarpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vpermilpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilpd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BDVER2-NEXT:    vpermilpd (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarpd:
 ; BTVER2:       # %bb.0:
@@ -4313,9 +4313,9 @@
 ;
 ; BDVER2-LABEL: test_permilvarpd_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vpermilpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
+; BDVER2-NEXT:    vpermilpd (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarpd_ymm:
 ; BTVER2:       # %bb.0:
@@ -4374,9 +4374,9 @@
 ;
 ; BDVER2-LABEL: test_permilvarps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vpermilps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilps %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BDVER2-NEXT:    vpermilps (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarps:
 ; BTVER2:       # %bb.0:
@@ -4435,9 +4435,9 @@
 ;
 ; BDVER2-LABEL: test_permilvarps_ymm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpermilps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vpermilps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpermilps %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
+; BDVER2-NEXT:    vpermilps (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_permilvarps_ymm:
 ; BTVER2:       # %bb.0:
@@ -4502,10 +4502,10 @@
 ;
 ; BDVER2-LABEL: test_rcpps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps (%rdi), %ymm1 # sched: [14:2.00]
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [7:2.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vrcpps (%rdi), %ymm1 # sched: [10:2.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rcpps:
 ; BTVER2:       # %bb.0:
@@ -4573,10 +4573,10 @@
 ;
 ; BDVER2-LABEL: test_roundpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vroundpd $7, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vroundpd $7, (%rdi), %ymm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vroundpd $7, (%rdi), %ymm1 # sched: [9:2.00]
+; BDVER2-NEXT:    vroundpd $7, %ymm0, %ymm0 # sched: [4:2.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_roundpd:
 ; BTVER2:       # %bb.0:
@@ -4644,10 +4644,10 @@
 ;
 ; BDVER2-LABEL: test_roundps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vroundps $7, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vroundps $7, (%rdi), %ymm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vroundps $7, (%rdi), %ymm1 # sched: [9:2.00]
+; BDVER2-NEXT:    vroundps $7, %ymm0, %ymm0 # sched: [4:2.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_roundps:
 ; BTVER2:       # %bb.0:
@@ -4715,10 +4715,10 @@
 ;
 ; BDVER2-LABEL: test_rsqrtps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrsqrtps (%rdi), %ymm1 # sched: [14:2.00]
-; BDVER2-NEXT:    vrsqrtps %ymm0, %ymm0 # sched: [7:2.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vrsqrtps (%rdi), %ymm1 # sched: [10:2.00]
+; BDVER2-NEXT:    vrsqrtps %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rsqrtps:
 ; BTVER2:       # %bb.0:
@@ -4786,10 +4786,10 @@
 ;
 ; BDVER2-LABEL: test_shufpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[1],ymm1[0],ymm0[2],ymm1[3] sched: [1:1.00]
-; BDVER2-NEXT:    vshufpd {{.*#+}} ymm1 = ymm1[1],mem[0],ymm1[2],mem[3] sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[1],ymm1[0],ymm0[2],ymm1[3] sched: [2:1.00]
+; BDVER2-NEXT:    vshufpd {{.*#+}} ymm1 = ymm1[1],mem[0],ymm1[2],mem[3] sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_shufpd:
 ; BTVER2:       # %bb.0:
@@ -4856,10 +4856,10 @@
 ;
 ; BDVER2-LABEL: test_shufps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,0],ymm1[0,0],ymm0[4,4],ymm1[4,4] sched: [1:1.00]
-; BDVER2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,3],mem[0,0],ymm1[4,7],mem[4,4] sched: [8:1.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,0],ymm1[0,0],ymm0[4,4],ymm1[4,4] sched: [2:1.00]
+; BDVER2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,3],mem[0,0],ymm1[4,7],mem[4,4] sched: [7:1.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_shufps:
 ; BTVER2:       # %bb.0:
@@ -4926,10 +4926,10 @@
 ;
 ; BDVER2-LABEL: test_sqrtpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsqrtpd (%rdi), %ymm1 # sched: [52:44.00]
-; BDVER2-NEXT:    vsqrtpd %ymm0, %ymm0 # sched: [45:44.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsqrtpd (%rdi), %ymm1 # sched: [14:27.00]
+; BDVER2-NEXT:    vsqrtpd %ymm0, %ymm0 # sched: [9:27.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sqrtpd:
 ; BTVER2:       # %bb.0:
@@ -4997,10 +4997,10 @@
 ;
 ; BDVER2-LABEL: test_sqrtps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsqrtps (%rdi), %ymm1 # sched: [36:28.00]
-; BDVER2-NEXT:    vsqrtps %ymm0, %ymm0 # sched: [29:28.00]
-; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsqrtps (%rdi), %ymm1 # sched: [14:21.00]
+; BDVER2-NEXT:    vsqrtps %ymm0, %ymm0 # sched: [9:21.00]
+; BDVER2-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sqrtps:
 ; BTVER2:       # %bb.0:
@@ -5062,9 +5062,9 @@
 ;
 ; BDVER2-LABEL: test_subpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vsubpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsubpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vsubpd (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_subpd:
 ; BTVER2:       # %bb.0:
@@ -5122,9 +5122,9 @@
 ;
 ; BDVER2-LABEL: test_subps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vsubps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsubps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vsubps (%rdi), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_subps:
 ; BTVER2:       # %bb.0:
@@ -5203,9 +5203,9 @@
 ; BDVER2-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; BDVER2-NEXT:    vtestpd %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setb %al # sched: [1:0.50]
-; BDVER2-NEXT:    vtestpd (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vtestpd (%rdi), %xmm0 # sched: [6:1.00]
+; BDVER2-NEXT:    adcl $0, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_testpd:
 ; BTVER2:       # %bb.0:
@@ -5298,10 +5298,10 @@
 ; BDVER2-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; BDVER2-NEXT:    vtestpd %ymm1, %ymm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setb %al # sched: [1:0.50]
-; BDVER2-NEXT:    vtestpd (%rdi), %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vtestpd (%rdi), %ymm0 # sched: [6:1.00]
+; BDVER2-NEXT:    adcl $0, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_testpd_ymm:
 ; BTVER2:       # %bb.0:
@@ -5389,9 +5389,9 @@
 ; BDVER2-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; BDVER2-NEXT:    vtestps %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setb %al # sched: [1:0.50]
-; BDVER2-NEXT:    vtestps (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vtestps (%rdi), %xmm0 # sched: [6:1.00]
+; BDVER2-NEXT:    adcl $0, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_testps:
 ; BTVER2:       # %bb.0:
@@ -5484,10 +5484,10 @@
 ; BDVER2-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; BDVER2-NEXT:    vtestps %ymm1, %ymm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setb %al # sched: [1:0.50]
-; BDVER2-NEXT:    vtestps (%rdi), %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vtestps (%rdi), %ymm0 # sched: [6:1.00]
+; BDVER2-NEXT:    adcl $0, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_testps_ymm:
 ; BTVER2:       # %bb.0:
@@ -5560,10 +5560,10 @@
 ;
 ; BDVER2-LABEL: test_unpckhpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3] sched: [1:1.00]
-; BDVER2-NEXT:    vunpckhpd {{.*#+}} ymm1 = ymm1[1],mem[1],ymm1[3],mem[3] sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3] sched: [2:1.00]
+; BDVER2-NEXT:    vunpckhpd {{.*#+}} ymm1 = ymm1[1],mem[1],ymm1[3],mem[3] sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_unpckhpd:
 ; BTVER2:       # %bb.0:
@@ -5624,9 +5624,9 @@
 ;
 ; BDVER2-LABEL: test_unpckhps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7] sched: [1:1.00]
-; BDVER2-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7] sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7] sched: [2:1.00]
+; BDVER2-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7] sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_unpckhps:
 ; BTVER2:       # %bb.0:
@@ -5690,10 +5690,10 @@
 ;
 ; BDVER2-LABEL: test_unpcklpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2] sched: [1:1.00]
-; BDVER2-NEXT:    vunpcklpd {{.*#+}} ymm1 = ymm1[0],mem[0],ymm1[2],mem[2] sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2] sched: [2:1.00]
+; BDVER2-NEXT:    vunpcklpd {{.*#+}} ymm1 = ymm1[0],mem[0],ymm1[2],mem[2] sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_unpcklpd:
 ; BTVER2:       # %bb.0:
@@ -5754,9 +5754,9 @@
 ;
 ; BDVER2-LABEL: test_unpcklps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5] sched: [1:1.00]
-; BDVER2-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5] sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5] sched: [2:1.00]
+; BDVER2-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5] sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_unpcklps:
 ; BTVER2:       # %bb.0:
@@ -5820,10 +5820,10 @@
 ;
 ; BDVER2-LABEL: test_xorpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vxorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vxorpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vxorpd %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vxorpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xorpd:
 ; BTVER2:       # %bb.0:
@@ -5894,10 +5894,10 @@
 ;
 ; BDVER2-LABEL: test_xorps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vxorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vxorps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vxorps %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vxorps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xorps:
 ; BTVER2:       # %bb.0:
@@ -5956,8 +5956,8 @@
 ;
 ; BDVER2-LABEL: test_zeroall:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vzeroall # sched: [9:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroall # sched: [90:8.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_zeroall:
 ; BTVER2:       # %bb.0:
@@ -6006,8 +6006,8 @@
 ;
 ; BDVER2-LABEL: test_zeroupper:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_zeroupper:
 ; BTVER2:       # %bb.0:
@@ -6086,12 +6086,12 @@
 ; BDVER2-LABEL: test_avx256_zero_idioms:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    vxorps %ymm0, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vxorpd %ymm1, %ymm1, %ymm1 # sched: [1:1.00]
-; BDVER2-NEXT:    vandnps %ymm2, %ymm2, %ymm2 # sched: [1:1.00]
-; BDVER2-NEXT:    vandnpd %ymm3, %ymm3, %ymm3 # sched: [1:1.00]
+; BDVER2-NEXT:    vxorps %ymm0, %ymm0, %ymm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vxorpd %ymm1, %ymm1, %ymm1 # sched: [2:1.00]
+; BDVER2-NEXT:    vandnps %ymm2, %ymm2, %ymm2 # sched: [2:1.00]
+; BDVER2-NEXT:    vandnpd %ymm3, %ymm3, %ymm3 # sched: [2:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_avx256_zero_idioms:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/avx-vzeroupper.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/avx-vzeroupper.ll
+++ llvm/trunk/test/CodeGen/X86/avx-vzeroupper.ll
@@ -2,7 +2,7 @@
 ; RUN: llc < %s -x86-use-vzeroupper -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=VZ --check-prefix=AVX
 ; RUN: llc < %s -x86-use-vzeroupper -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=VZ --check-prefix=AVX512
 ; RUN: llc < %s -x86-use-vzeroupper -mtriple=x86_64-unknown-unknown -mattr=+avx,+fast-partial-ymm-or-zmm-write | FileCheck %s --check-prefix=ALL --check-prefix=NO-VZ --check-prefix=FAST-ymm-zmm
-; RUN: llc < %s -x86-use-vzeroupper -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=NO-VZ --check-prefix=BDVER2
+; RUN: llc < %s -x86-use-vzeroupper -mtriple=x86_64-unknown-unknown -mcpu=bdver2 | FileCheck %s --check-prefix=ALL --check-prefix=NO-VZ --check-prefix=BDVER2
 ; RUN: llc < %s -x86-use-vzeroupper -mtriple=x86_64-unknown-unknown -mcpu=btver2 | FileCheck %s --check-prefix=ALL --check-prefix=NO-VZ --check-prefix=BTVER2
 
 declare i32 @foo()
@@ -60,8 +60,8 @@
 ; BDVER2-LABEL: test01:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    subq $56, %rsp
-; BDVER2-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
 ; BDVER2-NEXT:    vmovaps {{.*}}(%rip), %xmm0
+; BDVER2-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
 ; BDVER2-NEXT:    vzeroupper
 ; BDVER2-NEXT:    callq do_sse
 ; BDVER2-NEXT:    vmovaps %xmm0, {{.*}}(%rip)
@@ -203,8 +203,8 @@
 ; BDVER2-NEXT:    testl %eax, %eax
 ; BDVER2-NEXT:    jne .LBB3_1
 ; BDVER2-NEXT:  # %bb.2: # %for.body.preheader
-; BDVER2-NEXT:    movl $4, %ebx
 ; BDVER2-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
+; BDVER2-NEXT:    movl $4, %ebx
 ; BDVER2-NEXT:    .p2align 4, 0x90
 ; BDVER2-NEXT:  .LBB3_3: # %for.body
 ; BDVER2-NEXT:    # =>This Inner Loop Header: Depth=1
@@ -214,7 +214,7 @@
 ; BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; BDVER2-NEXT:    vzeroupper
 ; BDVER2-NEXT:    callq do_sse
-; BDVER2-NEXT:    addl $-1, %ebx
+; BDVER2-NEXT:    decl %ebx
 ; BDVER2-NEXT:    jne .LBB3_3
 ; BDVER2-NEXT:  # %bb.4: # %for.end
 ; BDVER2-NEXT:    addq $16, %rsp
Index: llvm/trunk/test/CodeGen/X86/bmi-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/bmi-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/bmi-schedule.ll
@@ -4,7 +4,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=knl     | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+bmi  | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2  | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2  | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1  | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -39,10 +39,10 @@
 ;
 ; BDVER2-LABEL: test_andn_i32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    andnl %esi, %edi, %ecx # sched: [1:0.33]
-; BDVER2-NEXT:    andnl (%rdx), %edi, %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andnl (%rdx), %edi, %eax # sched: [5:0.50]
+; BDVER2-NEXT:    andnl %esi, %edi, %ecx # sched: [1:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_andn_i32:
 ; BTVER2:       # %bb.0:
@@ -96,10 +96,10 @@
 ;
 ; BDVER2-LABEL: test_andn_i64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    andnq %rsi, %rdi, %rcx # sched: [1:0.33]
-; BDVER2-NEXT:    andnq (%rdx), %rdi, %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andnq (%rdx), %rdi, %rax # sched: [5:0.50]
+; BDVER2-NEXT:    andnq %rsi, %rdi, %rcx # sched: [1:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_andn_i64:
 ; BTVER2:       # %bb.0:
@@ -153,10 +153,10 @@
 ;
 ; BDVER2-LABEL: test_bextr_i32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    bextrl %edi, (%rdx), %ecx # sched: [7:1.00]
-; BDVER2-NEXT:    bextrl %edi, %esi, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    bextrl %edi, (%rdx), %ecx # sched: [6:0.50]
+; BDVER2-NEXT:    bextrl %edi, %esi, %eax # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bextr_i32:
 ; BTVER2:       # %bb.0:
@@ -210,10 +210,10 @@
 ;
 ; BDVER2-LABEL: test_bextr_i64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    bextrq %rdi, (%rdx), %rcx # sched: [7:1.00]
-; BDVER2-NEXT:    bextrq %rdi, %rsi, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    bextrq %rdi, (%rdx), %rcx # sched: [6:0.50]
+; BDVER2-NEXT:    bextrq %rdi, %rsi, %rax # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bextr_i64:
 ; BTVER2:       # %bb.0:
@@ -268,9 +268,9 @@
 ; BDVER2-LABEL: test_blsi_i32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    blsil (%rsi), %ecx # sched: [6:0.50]
-; BDVER2-NEXT:    blsil %edi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsil %edi, %eax # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blsi_i32:
 ; BTVER2:       # %bb.0:
@@ -326,9 +326,9 @@
 ; BDVER2-LABEL: test_blsi_i64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    blsiq (%rsi), %rcx # sched: [6:0.50]
-; BDVER2-NEXT:    blsiq %rdi, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsiq %rdi, %rax # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blsi_i64:
 ; BTVER2:       # %bb.0:
@@ -384,9 +384,9 @@
 ; BDVER2-LABEL: test_blsmsk_i32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    blsmskl (%rsi), %ecx # sched: [6:0.50]
-; BDVER2-NEXT:    blsmskl %edi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsmskl %edi, %eax # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blsmsk_i32:
 ; BTVER2:       # %bb.0:
@@ -442,9 +442,9 @@
 ; BDVER2-LABEL: test_blsmsk_i64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    blsmskq (%rsi), %rcx # sched: [6:0.50]
-; BDVER2-NEXT:    blsmskq %rdi, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsmskq %rdi, %rax # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blsmsk_i64:
 ; BTVER2:       # %bb.0:
@@ -500,9 +500,9 @@
 ; BDVER2-LABEL: test_blsr_i32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    blsrl (%rsi), %ecx # sched: [6:0.50]
-; BDVER2-NEXT:    blsrl %edi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsrl %edi, %eax # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blsr_i32:
 ; BTVER2:       # %bb.0:
@@ -558,9 +558,9 @@
 ; BDVER2-LABEL: test_blsr_i64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    blsrq (%rsi), %rcx # sched: [6:0.50]
-; BDVER2-NEXT:    blsrq %rdi, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsrq %rdi, %rax # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_blsr_i64:
 ; BTVER2:       # %bb.0:
@@ -619,11 +619,11 @@
 ;
 ; BDVER2-LABEL: test_cttz_i16:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    tzcntw (%rsi), %cx # sched: [8:1.00]
-; BDVER2-NEXT:    tzcntw %di, %ax # sched: [3:1.00]
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
+; BDVER2-NEXT:    tzcntw (%rsi), %cx # sched: [6:1.00]
+; BDVER2-NEXT:    tzcntw %di, %ax # sched: [2:1.00]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BDVER2-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cttz_i16:
 ; BTVER2:       # %bb.0:
@@ -679,10 +679,10 @@
 ;
 ; BDVER2-LABEL: test_cttz_i32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    tzcntl (%rsi), %ecx # sched: [8:1.00]
-; BDVER2-NEXT:    tzcntl %edi, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    tzcntl (%rsi), %ecx # sched: [6:1.00]
+; BDVER2-NEXT:    tzcntl %edi, %eax # sched: [2:1.00]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cttz_i32:
 ; BTVER2:       # %bb.0:
@@ -736,10 +736,10 @@
 ;
 ; BDVER2-LABEL: test_cttz_i64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    tzcntq (%rsi), %rcx # sched: [8:1.00]
-; BDVER2-NEXT:    tzcntq %rdi, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    tzcntq (%rsi), %rcx # sched: [6:1.00]
+; BDVER2-NEXT:    tzcntq %rdi, %rax # sched: [2:1.00]
+; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cttz_i64:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/cmov-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/cmov-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/cmov-schedule.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -544,68 +544,68 @@
 ; BDVER2-LABEL: test_cmov_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmovow %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnow %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovaew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovaew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovaew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbew %si, %di # sched: [3:1.00]
-; BDVER2-NEXT:    cmovbew %si, %di # sched: [3:1.00]
-; BDVER2-NEXT:    cmovaw %si, %di # sched: [3:1.00]
-; BDVER2-NEXT:    cmovaw %si, %di # sched: [3:1.00]
-; BDVER2-NEXT:    cmovsw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnsw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovpw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovpw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnpw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnpw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlew %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgw %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    cmovow (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnow (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovaew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovaew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovaew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbew (%rdx), %di # sched: [8:1.00]
-; BDVER2-NEXT:    cmovbew (%rdx), %di # sched: [8:1.00]
-; BDVER2-NEXT:    cmovaw (%rdx), %di # sched: [8:1.00]
-; BDVER2-NEXT:    cmovaw (%rdx), %di # sched: [8:1.00]
-; BDVER2-NEXT:    cmovsw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnsw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovpw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovpw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnpw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnpw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlew (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgw (%rdx), %di # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgw (%rdx), %di # sched: [7:0.67]
+; BDVER2-NEXT:    cmovow %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnow %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovsw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnsw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovpw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovpw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnpw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnpw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlew %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgw %si, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmovow (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnow (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovsw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnsw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovpw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovpw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnpw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnpw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlew (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgw (%rdx), %di # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgw (%rdx), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmov_16:
 ; BTVER2:       # %bb.0:
@@ -1274,68 +1274,68 @@
 ; BDVER2-LABEL: test_cmov_32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmovol %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnol %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovael %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovael %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovael %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbel %esi, %edi # sched: [3:1.00]
-; BDVER2-NEXT:    cmovbel %esi, %edi # sched: [3:1.00]
-; BDVER2-NEXT:    cmoval %esi, %edi # sched: [3:1.00]
-; BDVER2-NEXT:    cmoval %esi, %edi # sched: [3:1.00]
-; BDVER2-NEXT:    cmovsl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnsl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovpl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovpl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnpl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnpl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovll %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovll %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlel %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgl %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovol (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnol (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovael (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovael (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovael (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbel (%rdx), %edi # sched: [8:1.00]
-; BDVER2-NEXT:    cmovbel (%rdx), %edi # sched: [8:1.00]
-; BDVER2-NEXT:    cmoval (%rdx), %edi # sched: [8:1.00]
-; BDVER2-NEXT:    cmoval (%rdx), %edi # sched: [8:1.00]
-; BDVER2-NEXT:    cmovsl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnsl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovpl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovpl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnpl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnpl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovll (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovll (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlel (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgl (%rdx), %edi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgl (%rdx), %edi # sched: [7:0.67]
+; BDVER2-NEXT:    cmovol %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnol %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovael %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovael %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovael %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmoval %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmoval %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovsl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnsl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovpl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovpl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnpl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnpl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovll %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovll %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlel %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgl %esi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovol (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnol (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovael (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovael (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovael (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmoval (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmoval (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovsl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnsl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovpl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovpl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnpl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnpl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovll (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovll (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlel (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgl (%rdx), %edi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgl (%rdx), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmov_32:
 ; BTVER2:       # %bb.0:
@@ -2004,68 +2004,68 @@
 ; BDVER2-LABEL: test_cmov_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmovoq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnoq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovaeq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovaeq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovaeq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmoveq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmoveq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovneq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovneq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovbeq %rsi, %rdi # sched: [3:1.00]
-; BDVER2-NEXT:    cmovbeq %rsi, %rdi # sched: [3:1.00]
-; BDVER2-NEXT:    cmovaq %rsi, %rdi # sched: [3:1.00]
-; BDVER2-NEXT:    cmovaq %rsi, %rdi # sched: [3:1.00]
-; BDVER2-NEXT:    cmovsq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnsq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovpq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovpq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnpq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovnpq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovlq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgeq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgeq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovleq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovleq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovgq %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    cmovoq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnoq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovaeq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovaeq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovaeq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmoveq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmoveq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovneq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovneq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovbeq (%rdx), %rdi # sched: [8:1.00]
-; BDVER2-NEXT:    cmovbeq (%rdx), %rdi # sched: [8:1.00]
-; BDVER2-NEXT:    cmovaq (%rdx), %rdi # sched: [8:1.00]
-; BDVER2-NEXT:    cmovaq (%rdx), %rdi # sched: [8:1.00]
-; BDVER2-NEXT:    cmovsq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnsq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovpq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovpq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnpq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovnpq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovlq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgeq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgeq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovleq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovleq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgq (%rdx), %rdi # sched: [7:0.67]
-; BDVER2-NEXT:    cmovgq (%rdx), %rdi # sched: [7:0.67]
+; BDVER2-NEXT:    cmovoq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnoq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmoveq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmoveq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovneq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovneq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovbeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovaq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovsq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnsq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovpq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovpq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnpq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovnpq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovlq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgeq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovleq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovleq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovgq %rsi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmovoq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnoq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmoveq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmoveq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovneq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovneq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovbeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovaq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovsq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnsq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovpq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovpq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnpq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovnpq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovlq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgeq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovleq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovleq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgq (%rdx), %rdi # sched: [5:0.50]
+; BDVER2-NEXT:    cmovgq (%rdx), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmov_64:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/f16c-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/f16c-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/f16c-schedule.ll
@@ -4,7 +4,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+f16c | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -46,10 +46,10 @@
 ;
 ; BDVER2-LABEL: test_vcvtph2ps_128:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtph2ps (%rdi), %xmm1 # sched: [8:1.00]
-; BDVER2-NEXT:    vcvtph2ps %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtph2ps (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    vcvtph2ps %xmm0, %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_vcvtph2ps_128:
 ; BTVER2:       # %bb.0:
@@ -110,10 +110,10 @@
 ;
 ; BDVER2-LABEL: test_vcvtph2ps_256:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtph2ps (%rdi), %ymm1 # sched: [8:1.00]
-; BDVER2-NEXT:    vcvtph2ps %xmm0, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtph2ps (%rdi), %ymm1 # sched: [13:2.00]
+; BDVER2-NEXT:    vcvtph2ps %xmm0, %ymm0 # sched: [8:2.00]
+; BDVER2-NEXT:    vaddps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_vcvtph2ps_256:
 ; BTVER2:       # %bb.0:
@@ -169,9 +169,9 @@
 ;
 ; BDVER2-LABEL: test_vcvtps2ph_128:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtps2ph $0, %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vcvtps2ph $0, %xmm0, %xmm0 # sched: [8:1.00]
 ; BDVER2-NEXT:    vcvtps2ph $0, %xmm1, (%rdi) # sched: [4:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_vcvtps2ph_128:
 ; BTVER2:       # %bb.0:
@@ -230,10 +230,10 @@
 ;
 ; BDVER2-LABEL: test_vcvtps2ph_256:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [4:1.00]
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [4:2.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_vcvtps2ph_256:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/fma-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/fma-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/fma-schedule.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+fma | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+fma | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
@@ -35,7 +35,7 @@
 ; BDVER2-NEXT:    vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmadd231pd {{.*#+}} xmm0 = (xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddpd_128:
 ; HASWELL:       # %bb.0:
@@ -132,12 +132,12 @@
 ; BDVER2-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddpd_256:
 ; HASWELL:       # %bb.0:
@@ -242,7 +242,7 @@
 ; BDVER2-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmadd231ps {{.*#+}} xmm0 = (xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddps_128:
 ; HASWELL:       # %bb.0:
@@ -339,12 +339,12 @@
 ; BDVER2-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddps_256:
 ; HASWELL:       # %bb.0:
@@ -449,7 +449,7 @@
 ; BDVER2-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmadd231sd {{.*#+}} xmm0 = (xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddsd_128:
 ; HASWELL:       # %bb.0:
@@ -549,7 +549,7 @@
 ; BDVER2-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmadd231ss {{.*#+}} xmm0 = (xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddss_128:
 ; HASWELL:       # %bb.0:
@@ -653,7 +653,7 @@
 ; BDVER2-NEXT:    vfmaddsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) +/- mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddsub231pd {{.*#+}} xmm0 = (xmm1 * mem) +/- xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddsubpd_128:
 ; HASWELL:       # %bb.0:
@@ -750,12 +750,12 @@
 ; BDVER2-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddsubpd_256:
 ; HASWELL:       # %bb.0:
@@ -860,7 +860,7 @@
 ; BDVER2-NEXT:    vfmaddsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) +/- mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddsub231ps {{.*#+}} xmm0 = (xmm1 * mem) +/- xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddsubps_128:
 ; HASWELL:       # %bb.0:
@@ -957,12 +957,12 @@
 ; BDVER2-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmaddsubps_256:
 ; HASWELL:       # %bb.0:
@@ -1071,7 +1071,7 @@
 ; BDVER2-NEXT:    vfmsubadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmsubadd231pd {{.*#+}} xmm0 = (xmm1 * mem) -/+ xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubaddpd_128:
 ; HASWELL:       # %bb.0:
@@ -1168,12 +1168,12 @@
 ; BDVER2-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubaddpd_256:
 ; HASWELL:       # %bb.0:
@@ -1278,7 +1278,7 @@
 ; BDVER2-NEXT:    vfmsubadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmsubadd231ps {{.*#+}} xmm0 = (xmm1 * mem) -/+ xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubaddps_128:
 ; HASWELL:       # %bb.0:
@@ -1375,12 +1375,12 @@
 ; BDVER2-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubaddps_256:
 ; HASWELL:       # %bb.0:
@@ -1489,7 +1489,7 @@
 ; BDVER2-NEXT:    vfmsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmsub231pd {{.*#+}} xmm0 = (xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubpd_128:
 ; HASWELL:       # %bb.0:
@@ -1586,12 +1586,12 @@
 ; BDVER2-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubpd_256:
 ; HASWELL:       # %bb.0:
@@ -1696,7 +1696,7 @@
 ; BDVER2-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmsub231ps {{.*#+}} xmm0 = (xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubps_128:
 ; HASWELL:       # %bb.0:
@@ -1793,12 +1793,12 @@
 ; BDVER2-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [10:0.50]
-; BDVER2-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [10:1.00]
+; BDVER2-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubps_256:
 ; HASWELL:       # %bb.0:
@@ -1903,7 +1903,7 @@
 ; BDVER2-NEXT:    vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmsub231sd {{.*#+}} xmm0 = (xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubsd_128:
 ; HASWELL:       # %bb.0:
@@ -2003,7 +2003,7 @@
 ; BDVER2-NEXT:    vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfmsub231ss {{.*#+}} xmm0 = (xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfmsubss_128:
 ; HASWELL:       # %bb.0:
@@ -2107,7 +2107,7 @@
 ; BDVER2-NEXT:    vfnmadd213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmadd231pd {{.*#+}} xmm0 = -(xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmaddpd_128:
 ; HASWELL:       # %bb.0:
@@ -2204,12 +2204,12 @@
 ; BDVER2-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [10:0.50]
-; BDVER2-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [10:1.00]
+; BDVER2-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmaddpd_256:
 ; HASWELL:       # %bb.0:
@@ -2314,7 +2314,7 @@
 ; BDVER2-NEXT:    vfnmadd213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmadd231ps {{.*#+}} xmm0 = -(xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmaddps_128:
 ; HASWELL:       # %bb.0:
@@ -2411,12 +2411,12 @@
 ; BDVER2-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [10:0.50]
-; BDVER2-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [10:1.00]
+; BDVER2-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmaddps_256:
 ; HASWELL:       # %bb.0:
@@ -2521,7 +2521,7 @@
 ; BDVER2-NEXT:    vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmadd231sd {{.*#+}} xmm0 = -(xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmaddsd_128:
 ; HASWELL:       # %bb.0:
@@ -2621,7 +2621,7 @@
 ; BDVER2-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmadd231ss {{.*#+}} xmm0 = -(xmm1 * mem) + xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmaddss_128:
 ; HASWELL:       # %bb.0:
@@ -2725,7 +2725,7 @@
 ; BDVER2-NEXT:    vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmsub231pd {{.*#+}} xmm0 = -(xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmsubpd_128:
 ; HASWELL:       # %bb.0:
@@ -2822,12 +2822,12 @@
 ; BDVER2-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [10:0.50]
-; BDVER2-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [10:1.00]
+; BDVER2-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmsubpd_256:
 ; HASWELL:       # %bb.0:
@@ -2932,7 +2932,7 @@
 ; BDVER2-NEXT:    vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmsub231ps {{.*#+}} xmm0 = -(xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmsubps_128:
 ; HASWELL:       # %bb.0:
@@ -3029,12 +3029,12 @@
 ; BDVER2-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
-; BDVER2-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [10:0.50]
-; BDVER2-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [10:0.50]
-; BDVER2-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [10:0.50]
+; BDVER2-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [10:1.00]
+; BDVER2-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [10:1.00]
+; BDVER2-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmsubps_256:
 ; HASWELL:       # %bb.0:
@@ -3139,7 +3139,7 @@
 ; BDVER2-NEXT:    vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmsub231sd {{.*#+}} xmm0 = -(xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmsubsd_128:
 ; HASWELL:       # %bb.0:
@@ -3239,7 +3239,7 @@
 ; BDVER2-NEXT:    vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - mem sched: [10:0.50]
 ; BDVER2-NEXT:    vfnmsub231ss {{.*#+}} xmm0 = -(xmm1 * mem) - xmm0 sched: [10:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; HASWELL-LABEL: test_vfnmsubss_128:
 ; HASWELL:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/fma.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/fma.ll
+++ llvm/trunk/test/CodeGen/X86/fma.ll
@@ -247,76 +247,6 @@
 ; FMA32-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
 ; FMA32-NEXT:    retl ## encoding: [0xc3]
 ;
-; FMACALL32-LABEL: test_v4f32:
-; FMACALL32:       ## %bb.0: ## %entry
-; FMACALL32-NEXT:    subl $108, %esp ## encoding: [0x83,0xec,0x6c]
-; FMACALL32-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x40]
-; FMACALL32-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x30]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x20]
-; FMACALL32-NEXT:    vextractps $2, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vextractps $2, %xmm1, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x4c,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x60]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x30]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x20]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x54]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x30]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x20]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x30]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x20]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x1c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x54]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x18]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x60]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x14]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x10]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfa,0x10,0x44,0x24,0x1c]
-; FMACALL32-NEXT:    ## xmm0 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x18,0x10]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0],xmm0[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x14,0x20]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x10,0x30]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0,1,2],mem[0]
-; FMACALL32-NEXT:    addl $108, %esp ## encoding: [0x83,0xc4,0x6c]
-; FMACALL32-NEXT:    retl ## encoding: [0xc3]
-;
 ; FMA64-LABEL: test_v4f32:
 ; FMA64:       ## %bb.0: ## %entry
 ; FMA64-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
@@ -407,6 +337,76 @@
 ; AVX512VL-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
 ; AVX512VL-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
 ; AVX512VL-NEXT:    retq ## encoding: [0xc3]
+;
+; FMACALL32_BDVER2-LABEL: test_v4f32:
+; FMACALL32_BDVER2:       ## %bb.0: ## %entry
+; FMACALL32_BDVER2-NEXT:    subl $108, %esp ## encoding: [0x83,0xec,0x6c]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm1, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x4c,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x54,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x4c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x54]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x54,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x4c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x1c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x54]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x18]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x14]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfa,0x10,0x44,0x24,0x1c]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x18,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x14,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x10,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    addl $108, %esp ## encoding: [0x83,0xc4,0x6c]
+; FMACALL32_BDVER2-NEXT:    retl ## encoding: [0xc3]
 entry:
   %call = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c)
   ret <4 x float> %call
@@ -419,165 +419,6 @@
 ; FMA32-NEXT:    ## ymm0 = (ymm1 * ymm0) + ymm2
 ; FMA32-NEXT:    retl ## encoding: [0xc3]
 ;
-; FMACALL32-LABEL: test_v8f32:
-; FMACALL32:       ## %bb.0: ## %entry
-; FMACALL32-NEXT:    subl $316, %esp ## encoding: [0x81,0xec,0x3c,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## imm = 0x13C
-; FMACALL32-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x94,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x8c,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm2, %xmm3 ## encoding: [0xc4,0xe3,0x7d,0x19,0xd3,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x5c,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $2, %xmm3, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x5c,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm1, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xca,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x50]
-; FMACALL32-NEXT:    vextractps $2, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm1 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc1,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x40]
-; FMACALL32-NEXT:    vextractps $2, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x02]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xb4,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xa8,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x9c,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x90,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x84,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x78]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x3c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x78]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x38]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x84,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x34]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x90,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x30]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x9c,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x2c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xa8,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x28]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xb4,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x24]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x20]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfa,0x10,0x44,0x24,0x3c]
-; FMACALL32-NEXT:    ## xmm0 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x38,0x10]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0],xmm0[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x34,0x20]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x30,0x30]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0,1,2],mem[0]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfa,0x10,0x4c,0x24,0x2c]
-; FMACALL32-NEXT:    ## xmm1 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x28,0x10]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0],mem[0],xmm1[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x24,0x20]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0,1],mem[0],xmm1[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x20,0x30]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0,1,2],mem[0]
-; FMACALL32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
-; FMACALL32-NEXT:    addl $316, %esp ## encoding: [0x81,0xc4,0x3c,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## imm = 0x13C
-; FMACALL32-NEXT:    retl ## encoding: [0xc3]
-;
 ; FMA64-LABEL: test_v8f32:
 ; FMA64:       ## %bb.0: ## %entry
 ; FMA64-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
@@ -745,6 +586,165 @@
 ; AVX512VL-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
 ; AVX512VL-NEXT:    ## ymm0 = (ymm1 * ymm0) + ymm2
 ; AVX512VL-NEXT:    retq ## encoding: [0xc3]
+;
+; FMACALL32_BDVER2-LABEL: test_v8f32:
+; FMACALL32_BDVER2:       ## %bb.0: ## %entry
+; FMACALL32_BDVER2-NEXT:    subl $316, %esp ## encoding: [0x81,0xec,0x3c,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## imm = 0x13C
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm2, %xmm3 ## encoding: [0xc4,0xe3,0x7d,0x19,0xd3,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x94,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm1, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xca,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x8c,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm1 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc1,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x5c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm3, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x5c,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xb4,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x54,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x4c,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xa8,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x9c,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x90,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x84,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x78]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x54,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x4c,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x3c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x78]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x38]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x84,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x34]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x90,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x9c,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x2c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xa8,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x28]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xb4,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfa,0x10,0x44,0x24,0x3c]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x38,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfa,0x10,0x4c,0x24,0x2c]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x28,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x34,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x30,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x20,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
+; FMACALL32_BDVER2-NEXT:    addl $316, %esp ## encoding: [0x81,0xc4,0x3c,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## imm = 0x13C
+; FMACALL32_BDVER2-NEXT:    retl ## encoding: [0xc3]
 entry:
   %call = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c)
   ret <8 x float> %call
@@ -765,321 +765,6 @@
 ; FMA32-NEXT:    popl %ebp ## encoding: [0x5d]
 ; FMA32-NEXT:    retl ## encoding: [0xc3]
 ;
-; FMACALL32-LABEL: test_v16f32:
-; FMACALL32:       ## %bb.0: ## %entry
-; FMACALL32-NEXT:    pushl %ebp ## encoding: [0x55]
-; FMACALL32-NEXT:    movl %esp, %ebp ## encoding: [0x89,0xe5]
-; FMACALL32-NEXT:    andl $-32, %esp ## encoding: [0x83,0xe4,0xe0]
-; FMACALL32-NEXT:    subl $448, %esp ## encoding: [0x81,0xec,0xc0,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## imm = 0x1C0
-; FMACALL32-NEXT:    vmovaps %ymm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x5c,0x24,0x60]
-; FMACALL32-NEXT:    vmovaps %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x94,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x8c,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x84,0x24,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm3, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xd8,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0xb0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm1, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc8,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x54,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xb0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x48,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xb0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x3c,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x30,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x24,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x18,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x0c,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xf4,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xe8,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x60]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x50]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xdc,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xb0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x2c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xdc,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x28]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x50]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x24]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x60]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x20]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x1c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xe8,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x18]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xf4,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x14]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x10]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x0c,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x4c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x18,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x48]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x24,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x44]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x30,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x40]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x3c,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x3c]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x48,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x38]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x54,0x01,0x00,0x00]
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x34]
-; FMACALL32-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x30]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfa,0x10,0x44,0x24,0x2c]
-; FMACALL32-NEXT:    ## xmm0 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x28,0x10]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0],xmm0[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x24,0x20]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x20,0x30]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0,1,2],mem[0]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfa,0x10,0x4c,0x24,0x1c]
-; FMACALL32-NEXT:    ## xmm1 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x18,0x10]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0],mem[0],xmm1[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x14,0x20]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0,1],mem[0],xmm1[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x10,0x30]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0,1,2],mem[0]
-; FMACALL32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfa,0x10,0x4c,0x24,0x4c]
-; FMACALL32-NEXT:    ## xmm1 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x48,0x10]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0],mem[0],xmm1[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x44,0x20]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0,1],mem[0],xmm1[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x40,0x30]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0,1,2],mem[0]
-; FMACALL32-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm2 ## encoding: [0xc5,0xfa,0x10,0x54,0x24,0x3c]
-; FMACALL32-NEXT:    ## xmm2 = mem[0],zero,zero,zero
-; FMACALL32-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc4,0xe3,0x69,0x21,0x54,0x24,0x38,0x10]
-; FMACALL32-NEXT:    ## xmm2 = xmm2[0],mem[0],xmm2[2,3]
-; FMACALL32-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc4,0xe3,0x69,0x21,0x54,0x24,0x34,0x20]
-; FMACALL32-NEXT:    ## xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; FMACALL32-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc4,0xe3,0x69,0x21,0x54,0x24,0x30,0x30]
-; FMACALL32-NEXT:    ## xmm2 = xmm2[0,1,2],mem[0]
-; FMACALL32-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1 ## encoding: [0xc4,0xe3,0x75,0x18,0xca,0x01]
-; FMACALL32-NEXT:    movl %ebp, %esp ## encoding: [0x89,0xec]
-; FMACALL32-NEXT:    popl %ebp ## encoding: [0x5d]
-; FMACALL32-NEXT:    retl ## encoding: [0xc3]
-;
 ; FMA64-LABEL: test_v16f32:
 ; FMA64:       ## %bb.0: ## %entry
 ; FMA64-NEXT:    vfmadd213ps %ymm4, %ymm2, %ymm0 ## encoding: [0xc4,0xe2,0x6d,0xa8,0xc4]
@@ -1378,6 +1063,321 @@
 ; AVX512VL-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa8,0xc2]
 ; AVX512VL-NEXT:    ## zmm0 = (zmm1 * zmm0) + zmm2
 ; AVX512VL-NEXT:    retq ## encoding: [0xc3]
+;
+; FMACALL32_BDVER2-LABEL: test_v16f32:
+; FMACALL32_BDVER2:       ## %bb.0: ## %entry
+; FMACALL32_BDVER2-NEXT:    pushl %ebp ## encoding: [0x55]
+; FMACALL32_BDVER2-NEXT:    movl %esp, %ebp ## encoding: [0x89,0xe5]
+; FMACALL32_BDVER2-NEXT:    andl $-32, %esp ## encoding: [0x83,0xe4,0xe0]
+; FMACALL32_BDVER2-NEXT:    subl $448, %esp ## encoding: [0x81,0xec,0xc0,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## imm = 0x1C0
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x84,0x24,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x94,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm3, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xda,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x8c,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm1, %xmm1 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc9,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x5c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x94,0x24,0xb0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x8c,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x54,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x94,0x24,0xb0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x8c,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x48,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xb0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x3c,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x30,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x24,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x18,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x0c,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x54,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x4c,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xf4,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x54,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x4c,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xe8,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x02]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $2, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x02]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x04,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $1, %xmm0, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x04,0x24,0x01]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xdc,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x08]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xfa,0x11,0x44,0x24,0x04]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovss %xmm0, (%esp) ## encoding: [0xc5,0xfa,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x94,0x24,0xb0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x8c,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x44,0x24,0x08,0x03]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x54,0x24,0x04,0x03]
+; FMACALL32_BDVER2-NEXT:    vextractps $3, %xmm1, (%esp) ## encoding: [0xc4,0xe3,0x79,0x17,0x0c,0x24,0x03]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x2c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xdc,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x28]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x24]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x1c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xe8,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x18]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xf4,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x14]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x0c,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x4c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x18,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x48]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x24,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x44]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x30,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x3c,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x3c]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x48,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x38]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x54,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x34]
+; FMACALL32_BDVER2-NEXT:    calll _fmaf ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fmaf-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstps {{[0-9]+}}(%esp) ## encoding: [0xd9,0x5c,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfa,0x10,0x44,0x24,0x2c]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x28,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfa,0x10,0x4c,0x24,0x1c]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x18,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm2 ## encoding: [0xc5,0xfa,0x10,0x54,0x24,0x3c]
+; FMACALL32_BDVER2-NEXT:    ## xmm2 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc4,0xe3,0x69,0x21,0x54,0x24,0x38,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm2 = xmm2[0],mem[0],xmm2[2,3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x14,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc4,0xe3,0x69,0x21,0x54,0x24,0x34,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm2 = xmm2[0,1],mem[0],xmm2[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0x44,0x24,0x20,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x10,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc4,0xe3,0x69,0x21,0x54,0x24,0x30,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm2 = xmm2[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovss {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfa,0x10,0x4c,0x24,0x4c]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = mem[0],zero,zero,zero
+; FMACALL32_BDVER2-NEXT:    vinsertps $16, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x48,0x10]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $32, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x44,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; FMACALL32_BDVER2-NEXT:    vinsertps $48, {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x21,0x4c,0x24,0x40,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0,1,2],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1 ## encoding: [0xc4,0xe3,0x75,0x18,0xca,0x01]
+; FMACALL32_BDVER2-NEXT:    movl %ebp, %esp ## encoding: [0x89,0xec]
+; FMACALL32_BDVER2-NEXT:    popl %ebp ## encoding: [0x5d]
+; FMACALL32_BDVER2-NEXT:    retl ## encoding: [0xc3]
 entry:
   %call = call <16 x float> @llvm.fma.v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c)
   ret <16 x float> %call
@@ -1390,41 +1390,6 @@
 ; FMA32-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
 ; FMA32-NEXT:    retl ## encoding: [0xc3]
 ;
-; FMACALL32-LABEL: test_v2f64:
-; FMACALL32:       ## %bb.0: ## %entry
-; FMACALL32-NEXT:    subl $108, %esp ## encoding: [0x83,0xec,0x6c]
-; FMACALL32-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x30]
-; FMACALL32-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x40]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vmovlps %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x54,0x24,0x10]
-; FMACALL32-NEXT:    vmovlhps %xmm1, %xmm0, %xmm2 ## encoding: [0xc5,0xf8,0x16,0xd1]
-; FMACALL32-NEXT:    ## xmm2 = xmm0[0],xmm1[0]
-; FMACALL32-NEXT:    vmovups %xmm2, (%esp) ## encoding: [0xc5,0xf8,0x11,0x14,0x24]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    vmovapd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x28,0x44,0x24,0x30]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
-; FMACALL32-NEXT:    vmovlps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x12,0x44,0x24,0x58]
-; FMACALL32-NEXT:    ## xmm0 = mem[0,1],xmm0[2,3]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x28]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x20]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfb,0x10,0x44,0x24,0x28]
-; FMACALL32-NEXT:    ## xmm0 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x16,0x44,0x24,0x20]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    addl $108, %esp ## encoding: [0x83,0xc4,0x6c]
-; FMACALL32-NEXT:    retl ## encoding: [0xc3]
-;
 ; FMA64-LABEL: test_v2f64:
 ; FMA64:       ## %bb.0: ## %entry
 ; FMA64-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
@@ -1477,6 +1442,41 @@
 ; AVX512VL-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
 ; AVX512VL-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
 ; AVX512VL-NEXT:    retq ## encoding: [0xc3]
+;
+; FMACALL32_BDVER2-LABEL: test_v2f64:
+; FMACALL32_BDVER2:       ## %bb.0: ## %entry
+; FMACALL32_BDVER2-NEXT:    subl $108, %esp ## encoding: [0x83,0xec,0x6c]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovlhps %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf8,0x16,0xc1]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],xmm1[0]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm2, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x54,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovapd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x28,0x44,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x40]
+; FMACALL32_BDVER2-NEXT:    vmovlps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x12,0x44,0x24,0x58]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0,1],xmm0[2,3]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x28]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfb,0x10,0x44,0x24,0x28]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x16,0x44,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    addl $108, %esp ## encoding: [0x83,0xc4,0x6c]
+; FMACALL32_BDVER2-NEXT:    retl ## encoding: [0xc3]
 entry:
   %call = call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c)
   ret <2 x double> %call
@@ -1489,90 +1489,6 @@
 ; FMA32-NEXT:    ## ymm0 = (ymm1 * ymm0) + ymm2
 ; FMA32-NEXT:    retl ## encoding: [0xc3]
 ;
-; FMACALL32-LABEL: test_v4f64:
-; FMACALL32:       ## %bb.0: ## %entry
-; FMACALL32-NEXT:    subl $252, %esp ## encoding: [0x81,0xec,0xfc,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x94,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x8c,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm2, %xmm3 ## encoding: [0xc4,0xe3,0x7d,0x19,0xd3,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x5c,0x24,0x70]
-; FMACALL32-NEXT:    vmovlps %xmm3, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x5c,0x24,0x10]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm1, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xca,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x50]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm1 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc1,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x4c,0x24,0x60]
-; FMACALL32-NEXT:    vmovlhps %xmm2, %xmm1, %xmm0 ## encoding: [0xc5,0xf0,0x16,0xc2]
-; FMACALL32-NEXT:    ## xmm0 = xmm1[0],xmm2[0]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x44]
-; FMACALL32-NEXT:    vmovupd {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfd,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vunpckhpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x15,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[1],mem[1]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x38]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    vmovapd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x28,0x44,0x24,0x70]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
-; FMACALL32-NEXT:    vmovlps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x12,0x44,0x24,0x68]
-; FMACALL32-NEXT:    ## xmm0 = mem[0,1],xmm0[2,3]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x30]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x38]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x28]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x44]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x20]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x18]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfb,0x10,0x44,0x24,0x30]
-; FMACALL32-NEXT:    ## xmm0 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x16,0x44,0x24,0x28]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfb,0x10,0x4c,0x24,0x20]
-; FMACALL32-NEXT:    ## xmm1 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x16,0x4c,0x24,0x18]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0],mem[0]
-; FMACALL32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
-; FMACALL32-NEXT:    addl $252, %esp ## encoding: [0x81,0xc4,0xfc,0x00,0x00,0x00]
-; FMACALL32-NEXT:    retl ## encoding: [0xc3]
-;
 ; FMA64-LABEL: test_v4f64:
 ; FMA64:       ## %bb.0: ## %entry
 ; FMA64-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
@@ -1664,6 +1580,90 @@
 ; AVX512VL-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
 ; AVX512VL-NEXT:    ## ymm0 = (ymm1 * ymm0) + ymm2
 ; AVX512VL-NEXT:    retq ## encoding: [0xc3]
+;
+; FMACALL32_BDVER2-LABEL: test_v4f64:
+; FMACALL32_BDVER2:       ## %bb.0: ## %entry
+; FMACALL32_BDVER2-NEXT:    subl $252, %esp ## encoding: [0x81,0xec,0xfc,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm2, %xmm3 ## encoding: [0xc4,0xe3,0x7d,0x19,0xd3,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x94,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm1, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xca,0x01]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x11,0x8c,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    vmovlhps %xmm2, %xmm0, %xmm0 ## encoding: [0xc5,0xf8,0x16,0xc2]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],xmm2[0]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x5c,0x24,0x70]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm3, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x5c,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x54,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x44]
+; FMACALL32_BDVER2-NEXT:    vmovupd {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfd,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vunpckhpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x15,0x84,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[1],mem[1]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x38]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovups {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x10,0x84,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovapd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x28,0x44,0x24,0x70]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    vmovlps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x12,0x44,0x24,0x68]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0,1],xmm0[2,3]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x38]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x28]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x44]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x18]
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfb,0x10,0x44,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfb,0x10,0x4c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x16,0x44,0x24,0x28]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x16,0x4c,0x24,0x18]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
+; FMACALL32_BDVER2-NEXT:    addl $252, %esp ## encoding: [0x81,0xc4,0xfc,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    retl ## encoding: [0xc3]
 entry:
   %call = call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c)
   ret <4 x double> %call
@@ -1684,179 +1684,6 @@
 ; FMA32-NEXT:    popl %ebp ## encoding: [0x5d]
 ; FMA32-NEXT:    retl ## encoding: [0xc3]
 ;
-; FMACALL32-LABEL: test_v8f64:
-; FMACALL32:       ## %bb.0: ## %entry
-; FMACALL32-NEXT:    pushl %ebp ## encoding: [0x55]
-; FMACALL32-NEXT:    movl %esp, %ebp ## encoding: [0x89,0xe5]
-; FMACALL32-NEXT:    andl $-32, %esp ## encoding: [0x83,0xe4,0xe0]
-; FMACALL32-NEXT:    subl $384, %esp ## encoding: [0x81,0xec,0x80,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## imm = 0x180
-; FMACALL32-NEXT:    vmovaps %ymm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x9c,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x94,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x8c,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm3, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xda,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x94,0x24,0x40,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm1, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc8,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0x50,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovlhps %xmm2, %xmm0, %xmm0 ## encoding: [0xc5,0xf8,0x16,0xc2]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],xmm2[0]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x94,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovapd 40(%ebp), %ymm0 ## encoding: [0xc5,0xfd,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vunpckhpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x15,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[1],mem[1]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x88,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
-; FMACALL32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x29,0x44,0x24,0x30]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm1 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc1,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x8c,0x24,0x30,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
-; FMACALL32-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x20]
-; FMACALL32-NEXT:    vunpckhpd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x15,0xc1]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[1],xmm1[1]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x30]
-; FMACALL32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x20]
-; FMACALL32-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0x30,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x30]
-; FMACALL32-NEXT:    vmovapd 8(%ebp), %ymm0 ## encoding: [0xc5,0xfd,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vunpckhpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x15,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[1],mem[1]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x20]
-; FMACALL32-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
-; FMACALL32-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xe0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0x00,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    vmovapd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf9,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
-; FMACALL32-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x40,0x01,0x00,0x00]
-; FMACALL32-NEXT:    vmovlps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xc5,0xf8,0x12,0x84,0x24,0x58,0x01,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm0 = mem[0,1],xmm0[2,3]
-; FMACALL32-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x60]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x20]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x58]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x30]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x50]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xa0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x48]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0xc0,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x9c,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x88,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x78]
-; FMACALL32-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
-; FMACALL32-NEXT:    ## encoding: [0xdb,0xac,0x24,0x94,0x00,0x00,0x00]
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x70]
-; FMACALL32-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
-; FMACALL32-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
-; FMACALL32-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x68]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfb,0x10,0x44,0x24,0x60]
-; FMACALL32-NEXT:    ## xmm0 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x16,0x44,0x24,0x58]
-; FMACALL32-NEXT:    ## xmm0 = xmm0[0],mem[0]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfb,0x10,0x4c,0x24,0x50]
-; FMACALL32-NEXT:    ## xmm1 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x16,0x4c,0x24,0x48]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0],mem[0]
-; FMACALL32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfb,0x10,0x8c,0x24,0x80,0x00,0x00,0x00]
-; FMACALL32-NEXT:    ## xmm1 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x16,0x4c,0x24,0x78]
-; FMACALL32-NEXT:    ## xmm1 = xmm1[0],mem[0]
-; FMACALL32-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm2 ## encoding: [0xc5,0xfb,0x10,0x54,0x24,0x70]
-; FMACALL32-NEXT:    ## xmm2 = mem[0],zero
-; FMACALL32-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0x16,0x54,0x24,0x68]
-; FMACALL32-NEXT:    ## xmm2 = xmm2[0],mem[0]
-; FMACALL32-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1 ## encoding: [0xc4,0xe3,0x75,0x18,0xca,0x01]
-; FMACALL32-NEXT:    movl %ebp, %esp ## encoding: [0x89,0xec]
-; FMACALL32-NEXT:    popl %ebp ## encoding: [0x5d]
-; FMACALL32-NEXT:    retl ## encoding: [0xc3]
-;
 ; FMA64-LABEL: test_v8f64:
 ; FMA64:       ## %bb.0: ## %entry
 ; FMA64-NEXT:    vfmadd213pd %ymm4, %ymm2, %ymm0 ## encoding: [0xc4,0xe2,0xed,0xa8,0xc4]
@@ -2011,6 +1838,179 @@
 ; AVX512VL-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa8,0xc2]
 ; AVX512VL-NEXT:    ## zmm0 = (zmm1 * zmm0) + zmm2
 ; AVX512VL-NEXT:    retq ## encoding: [0xc3]
+;
+; FMACALL32_BDVER2-LABEL: test_v8f64:
+; FMACALL32_BDVER2:       ## %bb.0: ## %entry
+; FMACALL32_BDVER2-NEXT:    pushl %ebp ## encoding: [0x55]
+; FMACALL32_BDVER2-NEXT:    movl %esp, %ebp ## encoding: [0x89,0xe5]
+; FMACALL32_BDVER2-NEXT:    andl $-32, %esp ## encoding: [0x83,0xe4,0xe0]
+; FMACALL32_BDVER2-NEXT:    subl $384, %esp ## encoding: [0x81,0xec,0x80,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## imm = 0x180
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x94,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm3, %xmm2 ## encoding: [0xc4,0xe3,0x7d,0x19,0xda,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm3, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x9c,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps %ymm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 32-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x29,0x8c,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x94,0x24,0x40,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm1, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc8,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x84,0x24,0x50,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovlhps %xmm2, %xmm0, %xmm0 ## encoding: [0xc5,0xf8,0x16,0xc2]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],xmm2[0]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovapd 40(%ebp), %ymm0 ## encoding: [0xc5,0xfd,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x94,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vunpckhpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x15,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[1],mem[1]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 40(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x28]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0x88,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x29,0x44,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm1 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc1,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x8c,0x24,0x30,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vextractf128 $1, %ymm0, %xmm0 ## encoding: [0xc4,0xe3,0x7d,0x19,0xc0,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) ## 16-byte Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x29,0x44,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vunpckhpd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x15,0xc1]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[1],xmm1[1]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xbc,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x44,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0x30,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovapd 8(%ebp), %ymm0 ## encoding: [0xc5,0xfd,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vunpckhpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x15,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[1],mem[1]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovaps 8(%ebp), %ymm0 ## encoding: [0xc5,0xfc,0x28,0x45,0x08]
+; FMACALL32_BDVER2-NEXT:    fstpt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Spill
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x7c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    vmovlps %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf8,0x13,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %ymm0 ## 32-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x84,0x24,0xe0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vunpcklpd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x14,0x84,0x24,0x00,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    vmovapd {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf9,0x28,0x84,0x24,0x60,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovhpd %xmm0, {{[0-9]+}}(%esp) ## encoding: [0xc5,0xf9,0x17,0x44,0x24,0x10]
+; FMACALL32_BDVER2-NEXT:    vmovaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 ## 16-byte Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x84,0x24,0x40,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    vmovlps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0, %xmm0 ## 16-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xc5,0xf8,0x12,0x84,0x24,0x58,0x01,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0,1],xmm0[2,3]
+; FMACALL32_BDVER2-NEXT:    vmovups %xmm0, (%esp) ## encoding: [0xc5,0xf8,0x11,0x04,0x24]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x20]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x58]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0x6c,0x24,0x30]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xa0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x48]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0xc0,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x9c,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x88,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x78]
+; FMACALL32_BDVER2-NEXT:    fldt {{[-0-9]+}}(%e{{[sb]}}p) ## 10-byte Folded Reload
+; FMACALL32_BDVER2-NEXT:    ## encoding: [0xdb,0xac,0x24,0x94,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x70]
+; FMACALL32_BDVER2-NEXT:    calll _fma ## encoding: [0xe8,A,A,A,A]
+; FMACALL32_BDVER2-NEXT:    ## fixup A - offset: 1, value: _fma-4, kind: FK_PCRel_4
+; FMACALL32_BDVER2-NEXT:    fstpl {{[0-9]+}}(%esp) ## encoding: [0xdd,0x5c,0x24,0x68]
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm0 ## encoding: [0xc5,0xfb,0x10,0x44,0x24,0x60]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfb,0x10,0x4c,0x24,0x50]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x16,0x44,0x24,0x58]
+; FMACALL32_BDVER2-NEXT:    ## xmm0 = xmm0[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x16,0x4c,0x24,0x48]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm2 ## encoding: [0xc5,0xfb,0x10,0x54,0x24,0x70]
+; FMACALL32_BDVER2-NEXT:    ## xmm2 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0x16,0x54,0x24,0x68]
+; FMACALL32_BDVER2-NEXT:    ## xmm2 = xmm2[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe3,0x7d,0x18,0xc1,0x01]
+; FMACALL32_BDVER2-NEXT:    vmovsd {{[0-9]+}}(%esp), %xmm1 ## encoding: [0xc5,0xfb,0x10,0x8c,0x24,0x80,0x00,0x00,0x00]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = mem[0],zero
+; FMACALL32_BDVER2-NEXT:    vmovhpd {{[0-9]+}}(%esp), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x16,0x4c,0x24,0x78]
+; FMACALL32_BDVER2-NEXT:    ## xmm1 = xmm1[0],mem[0]
+; FMACALL32_BDVER2-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1 ## encoding: [0xc4,0xe3,0x75,0x18,0xca,0x01]
+; FMACALL32_BDVER2-NEXT:    movl %ebp, %esp ## encoding: [0x89,0xec]
+; FMACALL32_BDVER2-NEXT:    popl %ebp ## encoding: [0x5d]
+; FMACALL32_BDVER2-NEXT:    retl ## encoding: [0xc3]
 entry:
   %call = call <8 x double> @llvm.fma.v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c)
   ret <8 x double> %call
Index: llvm/trunk/test/CodeGen/X86/fma4-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/fma4-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/fma4-schedule.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+fma4 | FileCheck %s --check-prefixes=CHECK,GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+fma4 | FileCheck %s --check-prefixes=CHECK,BDVER,BDVER12,BDVER1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+fma4 -mattr=-fma  | FileCheck %s --check-prefixes=CHECK,BDVER,BDVER12,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver1              | FileCheck %s --check-prefixes=CHECK,BDVER,BDVER12,BDVER1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-fma  | FileCheck %s --check-prefixes=CHECK,BDVER,BDVER12,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver3 -mattr=-fma  | FileCheck %s --check-prefixes=CHECK,BDVER,BDVER34,BDVER3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver4 -mattr=-fma  | FileCheck %s --check-prefixes=CHECK,BDVER,BDVER34,BDVER4
 
@@ -26,7 +26,7 @@
 ; BDVER12-NEXT:    vfmaddpd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmaddpd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddpd_128:
 ; BDVER34:       # %bb.0:
@@ -55,11 +55,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmaddpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddpd_256:
 ; BDVER34:       # %bb.0:
@@ -91,7 +91,7 @@
 ; BDVER12-NEXT:    vfmaddps (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmaddps %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddps_128:
 ; BDVER34:       # %bb.0:
@@ -120,11 +120,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddps_256:
 ; BDVER34:       # %bb.0:
@@ -156,7 +156,7 @@
 ; BDVER12-NEXT:    vfmaddsd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmaddsd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddsd_128:
 ; BDVER34:       # %bb.0:
@@ -187,7 +187,7 @@
 ; BDVER12-NEXT:    vfmaddss (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmaddss %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddss_128:
 ; BDVER34:       # %bb.0:
@@ -222,7 +222,7 @@
 ; BDVER12-NEXT:    vfmaddsubpd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmaddsubpd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddsubpd_128:
 ; BDVER34:       # %bb.0:
@@ -251,11 +251,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmaddsubpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmaddsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmaddsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmaddsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmaddsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddsubpd_256:
 ; BDVER34:       # %bb.0:
@@ -287,7 +287,7 @@
 ; BDVER12-NEXT:    vfmaddsubps (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmaddsubps %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddsubps_128:
 ; BDVER34:       # %bb.0:
@@ -316,11 +316,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmaddsubps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmaddsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmaddsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmaddsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmaddsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmaddsubps_256:
 ; BDVER34:       # %bb.0:
@@ -356,7 +356,7 @@
 ; BDVER12-NEXT:    vfmsubaddpd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmsubaddpd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubaddpd_128:
 ; BDVER34:       # %bb.0:
@@ -385,11 +385,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmsubaddpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmsubaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmsubaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmsubaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmsubaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubaddpd_256:
 ; BDVER34:       # %bb.0:
@@ -421,7 +421,7 @@
 ; BDVER12-NEXT:    vfmsubaddps (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmsubaddps %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubaddps_128:
 ; BDVER34:       # %bb.0:
@@ -450,11 +450,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmsubaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmsubaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmsubaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmsubaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmsubaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubaddps_256:
 ; BDVER34:       # %bb.0:
@@ -490,7 +490,7 @@
 ; BDVER12-NEXT:    vfmsubpd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmsubpd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubpd_128:
 ; BDVER34:       # %bb.0:
@@ -519,11 +519,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmsubpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubpd_256:
 ; BDVER34:       # %bb.0:
@@ -555,7 +555,7 @@
 ; BDVER12-NEXT:    vfmsubps (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmsubps %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubps_128:
 ; BDVER34:       # %bb.0:
@@ -584,11 +584,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfmsubps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfmsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfmsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubps_256:
 ; BDVER34:       # %bb.0:
@@ -620,7 +620,7 @@
 ; BDVER12-NEXT:    vfmsubsd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmsubsd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubsd_128:
 ; BDVER34:       # %bb.0:
@@ -651,7 +651,7 @@
 ; BDVER12-NEXT:    vfmsubss (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfmsubss %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfmsubss_128:
 ; BDVER34:       # %bb.0:
@@ -686,7 +686,7 @@
 ; BDVER12-NEXT:    vfnmaddpd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmaddpd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmaddpd_128:
 ; BDVER34:       # %bb.0:
@@ -715,11 +715,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfnmaddpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfnmaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfnmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfnmaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfnmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmaddpd_256:
 ; BDVER34:       # %bb.0:
@@ -751,7 +751,7 @@
 ; BDVER12-NEXT:    vfnmaddps (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmaddps %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmaddps_128:
 ; BDVER34:       # %bb.0:
@@ -780,11 +780,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfnmaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfnmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfnmaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfnmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmaddps_256:
 ; BDVER34:       # %bb.0:
@@ -816,7 +816,7 @@
 ; BDVER12-NEXT:    vfnmaddsd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmaddsd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmaddsd_128:
 ; BDVER34:       # %bb.0:
@@ -847,7 +847,7 @@
 ; BDVER12-NEXT:    vfnmaddss (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmaddss %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmaddss_128:
 ; BDVER34:       # %bb.0:
@@ -882,7 +882,7 @@
 ; BDVER12-NEXT:    vfnmsubpd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmsubpd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmsubpd_128:
 ; BDVER34:       # %bb.0:
@@ -911,11 +911,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfnmsubpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfnmsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfnmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfnmsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfnmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmsubpd_256:
 ; BDVER34:       # %bb.0:
@@ -947,7 +947,7 @@
 ; BDVER12-NEXT:    vfnmsubps (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmsubps %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmsubps_128:
 ; BDVER34:       # %bb.0:
@@ -976,11 +976,11 @@
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; BDVER12-NEXT:    vfnmsubps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER12-NEXT:    vfnmsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
-; BDVER12-NEXT:    vfnmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER12-NEXT:    vfnmsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfnmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmsubps_256:
 ; BDVER34:       # %bb.0:
@@ -1012,7 +1012,7 @@
 ; BDVER12-NEXT:    vfnmsubsd (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmsubsd %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmsubsd_128:
 ; BDVER34:       # %bb.0:
@@ -1043,7 +1043,7 @@
 ; BDVER12-NEXT:    vfnmsubss (%rdi), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    vfnmsubss %xmm1, (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER34-LABEL: test_vfnmsubss_128:
 ; BDVER34:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/lea32-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/lea32-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/lea32-schedule.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell   | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake     | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=knl         | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=-slow-3ops-lea      | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2      | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2      | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1      | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -65,7 +65,7 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal -24(%rdi), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_offset:
 ; BTVER2:       # %bb.0:
@@ -135,7 +135,7 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal 1024(%rdi), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_offset_big:
 ; BTVER2:       # %bb.0:
@@ -214,7 +214,7 @@
 ; BDVER2-NEXT:    # kill: def $esi killed $esi def $rsi
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal (%rdi,%rsi), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add:
 ; BTVER2:       # %bb.0:
@@ -299,7 +299,7 @@
 ; BDVER2-NEXT:    # kill: def $esi killed $esi def $rsi
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal 16(%rdi,%rsi), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_offset:
 ; BTVER2:       # %bb.0:
@@ -390,7 +390,7 @@
 ; BDVER2-NEXT:    # kill: def $esi killed $esi def $rsi
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal -4096(%rdi,%rsi), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_offset_big:
 ; BTVER2:       # %bb.0:
@@ -463,7 +463,7 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_mul:
 ; BTVER2:       # %bb.0:
@@ -538,7 +538,7 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal -32(%rdi,%rdi,2), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_mul_offset:
 ; BTVER2:       # %bb.0:
@@ -619,7 +619,7 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal 10000(%rdi,%rdi,8), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_mul_offset_big:
 ; BTVER2:       # %bb.0:
@@ -698,7 +698,7 @@
 ; BDVER2-NEXT:    # kill: def $esi killed $esi def $rsi
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal (%rdi,%rsi,2), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_scale:
 ; BTVER2:       # %bb.0:
@@ -784,7 +784,7 @@
 ; BDVER2-NEXT:    # kill: def $esi killed $esi def $rsi
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal 96(%rdi,%rsi,4), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_scale_offset:
 ; BTVER2:       # %bb.0:
@@ -876,7 +876,7 @@
 ; BDVER2-NEXT:    # kill: def $esi killed $esi def $rsi
 ; BDVER2-NEXT:    # kill: def $edi killed $edi def $rdi
 ; BDVER2-NEXT:    leal -1200(%rdi,%rsi,8), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_scale_offset_big:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/lea64-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/lea64-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/lea64-schedule.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell   | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake     | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=knl         | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=-slow-3ops-lea      | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2      | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2      | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1      | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -57,7 +57,7 @@
 ; BDVER2-LABEL: test_lea_offset:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq -24(%rdi), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_offset:
 ; BTVER2:       # %bb.0:
@@ -117,7 +117,7 @@
 ; BDVER2-LABEL: test_lea_offset_big:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq 1024(%rdi), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_offset_big:
 ; BTVER2:       # %bb.0:
@@ -178,7 +178,7 @@
 ; BDVER2-LABEL: test_lea_add:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq (%rdi,%rsi), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add:
 ; BTVER2:       # %bb.0:
@@ -243,7 +243,7 @@
 ; BDVER2-LABEL: test_lea_add_offset:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq 16(%rdi,%rsi), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_offset:
 ; BTVER2:       # %bb.0:
@@ -314,7 +314,7 @@
 ; BDVER2-LABEL: test_lea_add_offset_big:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq -4096(%rdi,%rsi), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_offset_big:
 ; BTVER2:       # %bb.0:
@@ -375,7 +375,7 @@
 ; BDVER2-LABEL: test_lea_mul:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_mul:
 ; BTVER2:       # %bb.0:
@@ -440,7 +440,7 @@
 ; BDVER2-LABEL: test_lea_mul_offset:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq -32(%rdi,%rdi,2), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_mul_offset:
 ; BTVER2:       # %bb.0:
@@ -511,7 +511,7 @@
 ; BDVER2-LABEL: test_lea_mul_offset_big:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq 10000(%rdi,%rdi,8), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_mul_offset_big:
 ; BTVER2:       # %bb.0:
@@ -572,7 +572,7 @@
 ; BDVER2-LABEL: test_lea_add_scale:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq (%rdi,%rsi,2), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_scale:
 ; BTVER2:       # %bb.0:
@@ -638,7 +638,7 @@
 ; BDVER2-LABEL: test_lea_add_scale_offset:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq 96(%rdi,%rsi,4), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_scale_offset:
 ; BTVER2:       # %bb.0:
@@ -710,7 +710,7 @@
 ; BDVER2-LABEL: test_lea_add_scale_offset_big:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    leaq -1200(%rdi,%rsi,8), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lea_add_scale_offset_big:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/lwp-intrinsics.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/lwp-intrinsics.ll
+++ llvm/trunk/test/CodeGen/X86/lwp-intrinsics.ll
@@ -40,14 +40,41 @@
 }
 
 define i8 @test_lwpins32_rri(i32 %a0, i32 %a1) nounwind {
-; X86-LABEL: test_lwpins32_rri:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    addl %ecx, %ecx
-; X86-NEXT:    lwpins $-1985229329, %ecx, %eax # imm = 0x89ABCDEF
-; X86-NEXT:    setb %al
-; X86-NEXT:    retl
+; X86_BDVER1-LABEL: test_lwpins32_rri:
+; X86_BDVER1:       # %bb.0:
+; X86_BDVER1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER1-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER1-NEXT:    addl %ecx, %ecx
+; X86_BDVER1-NEXT:    lwpins $-1985229329, %ecx, %eax # imm = 0x89ABCDEF
+; X86_BDVER1-NEXT:    setb %al
+; X86_BDVER1-NEXT:    retl
+;
+; X86_BDVER2-LABEL: test_lwpins32_rri:
+; X86_BDVER2:       # %bb.0:
+; X86_BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER2-NEXT:    addl %ecx, %ecx
+; X86_BDVER2-NEXT:    lwpins $-1985229329, %ecx, %eax # imm = 0x89ABCDEF
+; X86_BDVER2-NEXT:    setb %al
+; X86_BDVER2-NEXT:    retl
+;
+; X86_BDVER3-LABEL: test_lwpins32_rri:
+; X86_BDVER3:       # %bb.0:
+; X86_BDVER3-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER3-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER3-NEXT:    addl %ecx, %ecx
+; X86_BDVER3-NEXT:    lwpins $-1985229329, %ecx, %eax # imm = 0x89ABCDEF
+; X86_BDVER3-NEXT:    setb %al
+; X86_BDVER3-NEXT:    retl
+;
+; X86_BDVER4-LABEL: test_lwpins32_rri:
+; X86_BDVER4:       # %bb.0:
+; X86_BDVER4-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER4-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER4-NEXT:    addl %ecx, %ecx
+; X86_BDVER4-NEXT:    lwpins $-1985229329, %ecx, %eax # imm = 0x89ABCDEF
+; X86_BDVER4-NEXT:    setb %al
+; X86_BDVER4-NEXT:    retl
 ;
 ; X64-LABEL: test_lwpins32_rri:
 ; X64:       # %bb.0:
@@ -80,13 +107,37 @@
 }
 
 define void @test_lwpval32_rri(i32 %a0, i32 %a1) nounwind {
-; X86-LABEL: test_lwpval32_rri:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    addl %ecx, %ecx
-; X86-NEXT:    lwpval $-19088744, %ecx, %eax # imm = 0xFEDCBA98
-; X86-NEXT:    retl
+; X86_BDVER1-LABEL: test_lwpval32_rri:
+; X86_BDVER1:       # %bb.0:
+; X86_BDVER1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER1-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER1-NEXT:    addl %ecx, %ecx
+; X86_BDVER1-NEXT:    lwpval $-19088744, %ecx, %eax # imm = 0xFEDCBA98
+; X86_BDVER1-NEXT:    retl
+;
+; X86_BDVER2-LABEL: test_lwpval32_rri:
+; X86_BDVER2:       # %bb.0:
+; X86_BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER2-NEXT:    addl %ecx, %ecx
+; X86_BDVER2-NEXT:    lwpval $-19088744, %ecx, %eax # imm = 0xFEDCBA98
+; X86_BDVER2-NEXT:    retl
+;
+; X86_BDVER3-LABEL: test_lwpval32_rri:
+; X86_BDVER3:       # %bb.0:
+; X86_BDVER3-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER3-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER3-NEXT:    addl %ecx, %ecx
+; X86_BDVER3-NEXT:    lwpval $-19088744, %ecx, %eax # imm = 0xFEDCBA98
+; X86_BDVER3-NEXT:    retl
+;
+; X86_BDVER4-LABEL: test_lwpval32_rri:
+; X86_BDVER4:       # %bb.0:
+; X86_BDVER4-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86_BDVER4-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86_BDVER4-NEXT:    addl %ecx, %ecx
+; X86_BDVER4-NEXT:    lwpval $-19088744, %ecx, %eax # imm = 0xFEDCBA98
+; X86_BDVER4-NEXT:    retl
 ;
 ; X64-LABEL: test_lwpval32_rri:
 ; X64:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/lwp-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/lwp-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/lwp-schedule.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=x86-64 -mattr=+lwp | FileCheck %s --check-prefix=GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=x86-64 -mattr=+lwp | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER1
-; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=x86-64 -mattr=+lwp | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=bdver1 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER1
+; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=bdver3 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER3
 ; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule -mcpu=bdver4 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER4
 
@@ -13,8 +13,8 @@
 ;
 ; BDVER12-LABEL: test_llwpcb:
 ; BDVER12:       # %bb.0:
-; BDVER12-NEXT:    llwpcb %rdi # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    llwpcb %rdi # sched: [100:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_llwpcb:
 ; BDVER3:       # %bb.0:
@@ -37,8 +37,8 @@
 ;
 ; BDVER12-LABEL: test_slwpcb:
 ; BDVER12:       # %bb.0:
-; BDVER12-NEXT:    slwpcb %rax # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    slwpcb %rax # sched: [100:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_slwpcb:
 ; BDVER3:       # %bb.0:
@@ -64,11 +64,11 @@
 ;
 ; BDVER12-LABEL: test_lwpins32_rri:
 ; BDVER12:       # %bb.0:
-; BDVER12-NEXT:    addl %esi, %esi # sched: [1:0.33]
+; BDVER12-NEXT:    addl %esi, %esi # sched: [1:0.50]
 ; BDVER12-NEXT:    lwpins $-1985229329, %esi, %edi # imm = 0x89ABCDEF
-; BDVER12-NEXT:    # sched: [100:0.33]
+; BDVER12-NEXT:    # sched: [100:0.50]
 ; BDVER12-NEXT:    setb %al # sched: [1:0.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpins32_rri:
 ; BDVER3:       # %bb.0:
@@ -99,9 +99,9 @@
 ; BDVER12-LABEL: test_lwpins32_rmi:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    lwpins $1985229328, (%rsi), %edi # imm = 0x76543210
-; BDVER12-NEXT:    # sched: [100:0.33]
+; BDVER12-NEXT:    # sched: [100:0.50]
 ; BDVER12-NEXT:    setb %al # sched: [1:0.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpins32_rmi:
 ; BDVER3:       # %bb.0:
@@ -130,9 +130,9 @@
 ; BDVER12-LABEL: test_lwpins64_rri:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    lwpins $-1985229329, %esi, %rdi # imm = 0x89ABCDEF
-; BDVER12-NEXT:    # sched: [100:0.33]
+; BDVER12-NEXT:    # sched: [100:0.50]
 ; BDVER12-NEXT:    setb %al # sched: [1:0.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpins64_rri:
 ; BDVER3:       # %bb.0:
@@ -160,9 +160,9 @@
 ; BDVER12-LABEL: test_lwpins64_rmi:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    lwpins $1985229328, (%rsi), %rdi # imm = 0x76543210
-; BDVER12-NEXT:    # sched: [100:0.33]
+; BDVER12-NEXT:    # sched: [100:0.50]
 ; BDVER12-NEXT:    setb %al # sched: [1:0.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpins64_rmi:
 ; BDVER3:       # %bb.0:
@@ -190,10 +190,10 @@
 ;
 ; BDVER12-LABEL: test_lwpval32_rri:
 ; BDVER12:       # %bb.0:
-; BDVER12-NEXT:    addl %esi, %esi # sched: [1:0.33]
+; BDVER12-NEXT:    addl %esi, %esi # sched: [1:0.50]
 ; BDVER12-NEXT:    lwpval $-19088744, %esi, %edi # imm = 0xFEDCBA98
-; BDVER12-NEXT:    # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    # sched: [100:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpval32_rri:
 ; BDVER3:       # %bb.0:
@@ -221,8 +221,8 @@
 ; BDVER12-LABEL: test_lwpval32_rmi:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    lwpval $305419896, (%rsi), %edi # imm = 0x12345678
-; BDVER12-NEXT:    # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    # sched: [100:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpval32_rmi:
 ; BDVER3:       # %bb.0:
@@ -248,8 +248,8 @@
 ; BDVER12-LABEL: test_lwpval64_rri:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    lwpval $-19088744, %esi, %rdi # imm = 0xFEDCBA98
-; BDVER12-NEXT:    # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    # sched: [100:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpval64_rri:
 ; BDVER3:       # %bb.0:
@@ -274,8 +274,8 @@
 ; BDVER12-LABEL: test_lwpval64_rmi:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    lwpval $305419896, (%rsi), %rdi # imm = 0x12345678
-; BDVER12-NEXT:    # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    # sched: [100:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_lwpval64_rmi:
 ; BDVER3:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/lzcnt-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/lzcnt-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/lzcnt-schedule.ll
@@ -4,7 +4,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake   | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=knl       | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+lzcnt    | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2    | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2    | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1    | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -43,11 +43,11 @@
 ;
 ; BDVER2-LABEL: test_ctlz_i16:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    lzcntw (%rsi), %cx # sched: [8:1.00]
-; BDVER2-NEXT:    lzcntw %di, %ax # sched: [3:1.00]
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
+; BDVER2-NEXT:    lzcntw (%rsi), %cx # sched: [6:0.50]
+; BDVER2-NEXT:    lzcntw %di, %ax # sched: [2:0.50]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BDVER2-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ctlz_i16:
 ; BTVER2:       # %bb.0:
@@ -103,10 +103,10 @@
 ;
 ; BDVER2-LABEL: test_ctlz_i32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    lzcntl (%rsi), %ecx # sched: [8:1.00]
-; BDVER2-NEXT:    lzcntl %edi, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    lzcntl (%rsi), %ecx # sched: [6:0.50]
+; BDVER2-NEXT:    lzcntl %edi, %eax # sched: [2:0.50]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ctlz_i32:
 ; BTVER2:       # %bb.0:
@@ -160,10 +160,10 @@
 ;
 ; BDVER2-LABEL: test_ctlz_i64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    lzcntq (%rsi), %rcx # sched: [8:1.00]
-; BDVER2-NEXT:    lzcntq %rdi, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    lzcntq (%rsi), %rcx # sched: [6:0.50]
+; BDVER2-NEXT:    lzcntq %rdi, %rax # sched: [2:0.50]
+; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ctlz_i64:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/memset.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/memset.ll
+++ llvm/trunk/test/CodeGen/X86/memset.ll
@@ -22,7 +22,6 @@
 ; X86-NEXT:    calll _foo
 ; X86-NEXT:    addl $44, %esp
 ; X86-NEXT:    retl
-; X86-NEXT:    ## -- End function
 ;
 ; XMM-LABEL: t:
 ; XMM:       ## %bb.0: ## %entry
@@ -35,7 +34,6 @@
 ; XMM-NEXT:    calll _foo
 ; XMM-NEXT:    addl $60, %esp
 ; XMM-NEXT:    retl
-; XMM-NEXT:    ## -- End function
 ;
 ; YMM-LABEL: t:
 ; YMM:       ## %bb.0: ## %entry
@@ -44,15 +42,14 @@
 ; YMM-NEXT:    andl $-32, %esp
 ; YMM-NEXT:    subl $96, %esp
 ; YMM-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; YMM-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%esp)
 ; YMM-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; YMM-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%esp)
 ; YMM-NEXT:    movl %eax, (%esp)
 ; YMM-NEXT:    vzeroupper
 ; YMM-NEXT:    calll _foo
 ; YMM-NEXT:    movl %ebp, %esp
 ; YMM-NEXT:    popl %ebp
 ; YMM-NEXT:    retl
-; YMM-NEXT:    ## -- End function
 entry:
 	%up_mvd = alloca [8 x %struct.x]		; <[8 x %struct.x]*> [#uses=2]
 	%up_mvd116 = getelementptr [8 x %struct.x], [8 x %struct.x]* %up_mvd, i32 0, i32 0		; <%struct.x*> [#uses=1]
Index: llvm/trunk/test/CodeGen/X86/mmx-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/mmx-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/mmx-schedule.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=+ssse3 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -79,11 +79,11 @@
 ;
 ; BDVER2-LABEL: test_cvtpd2pi:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    cvtpd2pi (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    cvtpd2pi %xmm0, %mm1 # sched: [4:1.00]
-; BDVER2-NEXT:    por %mm1, %mm0 # sched: [1:0.33]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    cvtpd2pi (%rdi), %mm1 # sched: [13:1.00]
+; BDVER2-NEXT:    cvtpd2pi %xmm0, %mm0 # sched: [6:1.00]
+; BDVER2-NEXT:    por %mm0, %mm1 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm1, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtpd2pi:
 ; BTVER2:       # %bb.0:
@@ -168,10 +168,10 @@
 ;
 ; BDVER2-LABEL: test_cvtpi2pd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    cvtpi2pd %mm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    cvtpi2pd (%rdi), %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    cvtpi2pd (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    cvtpi2pd %mm0, %xmm0 # sched: [6:1.00]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtpi2pd:
 ; BTVER2:       # %bb.0:
@@ -253,10 +253,10 @@
 ;
 ; BDVER2-LABEL: test_cvtpi2ps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    cvtpi2ps %mm0, %xmm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    cvtpi2ps (%rdi), %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    cvtpi2ps %mm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtpi2ps:
 ; BTVER2:       # %bb.0:
@@ -346,11 +346,11 @@
 ;
 ; BDVER2-LABEL: test_cvtps2pi:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    cvtps2pi %xmm0, %mm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    cvtps2pi (%rdi), %mm1 # sched: [9:1.00]
-; BDVER2-NEXT:    por %mm0, %mm1 # sched: [1:0.33]
-; BDVER2-NEXT:    movq %mm1, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    cvtps2pi %xmm0, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    por %mm0, %mm1 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm1, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvtps2pi:
 ; BTVER2:       # %bb.0:
@@ -443,11 +443,11 @@
 ;
 ; BDVER2-LABEL: test_cvttpd2pi:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    cvttpd2pi (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    cvttpd2pi %xmm0, %mm1 # sched: [4:1.00]
-; BDVER2-NEXT:    por %mm1, %mm0 # sched: [1:0.33]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    cvttpd2pi (%rdi), %mm1 # sched: [13:1.00]
+; BDVER2-NEXT:    cvttpd2pi %xmm0, %mm0 # sched: [6:1.00]
+; BDVER2-NEXT:    por %mm0, %mm1 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm1, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvttpd2pi:
 ; BTVER2:       # %bb.0:
@@ -540,11 +540,11 @@
 ;
 ; BDVER2-LABEL: test_cvttps2pi:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    cvttps2pi %xmm0, %mm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    cvttps2pi (%rdi), %mm1 # sched: [9:1.00]
-; BDVER2-NEXT:    por %mm0, %mm1 # sched: [1:0.33]
-; BDVER2-NEXT:    movq %mm1, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    cvttps2pi %xmm0, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    por %mm0, %mm1 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm1, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cvttps2pi:
 ; BTVER2:       # %bb.0:
@@ -613,8 +613,8 @@
 ;
 ; BDVER2-LABEL: test_emms:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    emms # sched: [31:10.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    emms # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_emms:
 ; BTVER2:       # %bb.0:
@@ -673,8 +673,8 @@
 ;
 ; BDVER2-LABEL: test_maskmovq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    maskmovq %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    maskmovq %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_maskmovq:
 ; BTVER2:       # %bb.0:
@@ -781,14 +781,14 @@
 ;
 ; BDVER2-LABEL: test_movd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movd %edi, %mm1 # sched: [1:1.00]
+; BDVER2-NEXT:    movd %edi, %mm1 # sched: [10:0.50]
 ; BDVER2-NEXT:    movd (%rsi), %mm2 # sched: [5:0.50]
-; BDVER2-NEXT:    paddd %mm1, %mm2 # sched: [3:1.00]
-; BDVER2-NEXT:    paddd %mm2, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    movd %mm2, %ecx # sched: [2:1.00]
-; BDVER2-NEXT:    movd %mm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    movl %ecx, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddd %mm1, %mm2 # sched: [2:0.50]
+; BDVER2-NEXT:    paddd %mm2, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movd %mm2, %ecx # sched: [10:1.00]
+; BDVER2-NEXT:    movd %mm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    movl %ecx, (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movd:
 ; BTVER2:       # %bb.0:
@@ -885,10 +885,10 @@
 ;
 ; BDVER2-LABEL: test_movdq2q:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movdq2q %xmm0, %mm0 # sched: [2:1.00]
-; BDVER2-NEXT:    paddd %mm0, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movdq2q %xmm0, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddd %mm0, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movdq2q:
 ; BTVER2:       # %bb.0:
@@ -953,8 +953,8 @@
 ;
 ; BDVER2-LABEL: test_movntq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movntq %mm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movntq %mm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movntq:
 ; BTVER2:       # %bb.0:
@@ -1032,9 +1032,9 @@
 ; BDVER2-LABEL: test_movq:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movq (%rdi), %mm0 # sched: [5:0.50]
-; BDVER2-NEXT:    paddd %mm0, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    movq %mm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddd %mm0, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movq:
 ; BTVER2:       # %bb.0:
@@ -1100,8 +1100,8 @@
 ;
 ; BDVER2-LABEL: test_movq2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movq2dq %mm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movq2dq %mm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movq2dq:
 ; BTVER2:       # %bb.0:
@@ -1176,10 +1176,10 @@
 ;
 ; BDVER2-LABEL: test_pabsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pabsb (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    pabsb %mm0, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pabsb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    pabsb %mm0, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pabsb:
 ; BTVER2:       # %bb.0:
@@ -1261,10 +1261,10 @@
 ;
 ; BDVER2-LABEL: test_pabsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pabsd (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    pabsd %mm0, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pabsd (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    pabsd %mm0, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pabsd:
 ; BTVER2:       # %bb.0:
@@ -1346,10 +1346,10 @@
 ;
 ; BDVER2-LABEL: test_pabsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pabsw (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    pabsw %mm0, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pabsw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    pabsw %mm0, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pabsw:
 ; BTVER2:       # %bb.0:
@@ -1431,10 +1431,10 @@
 ;
 ; BDVER2-LABEL: test_packssdw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    packssdw %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    packssdw (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    packssdw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    packssdw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_packssdw:
 ; BTVER2:       # %bb.0:
@@ -1516,10 +1516,10 @@
 ;
 ; BDVER2-LABEL: test_packsswb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    packsswb %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    packsswb (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    packsswb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    packsswb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_packsswb:
 ; BTVER2:       # %bb.0:
@@ -1601,10 +1601,10 @@
 ;
 ; BDVER2-LABEL: test_packuswb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    packuswb %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    packuswb (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    packuswb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    packuswb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_packuswb:
 ; BTVER2:       # %bb.0:
@@ -1686,10 +1686,10 @@
 ;
 ; BDVER2-LABEL: test_paddb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddb:
 ; BTVER2:       # %bb.0:
@@ -1771,10 +1771,10 @@
 ;
 ; BDVER2-LABEL: test_paddd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddd %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddd (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddd %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddd (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddd:
 ; BTVER2:       # %bb.0:
@@ -1856,10 +1856,10 @@
 ;
 ; BDVER2-LABEL: test_paddq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddq %mm1, %mm0 # sched: [1:0.50]
+; BDVER2-NEXT:    paddq %mm1, %mm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    paddq (%rdi), %mm0 # sched: [7:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddq:
 ; BTVER2:       # %bb.0:
@@ -1941,10 +1941,10 @@
 ;
 ; BDVER2-LABEL: test_paddsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddsb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddsb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddsb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddsb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddsb:
 ; BTVER2:       # %bb.0:
@@ -2026,10 +2026,10 @@
 ;
 ; BDVER2-LABEL: test_paddsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddsw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddsw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddsw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddsw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddsw:
 ; BTVER2:       # %bb.0:
@@ -2111,10 +2111,10 @@
 ;
 ; BDVER2-LABEL: test_paddusb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddusb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddusb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddusb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddusb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddusb:
 ; BTVER2:       # %bb.0:
@@ -2196,10 +2196,10 @@
 ;
 ; BDVER2-LABEL: test_paddusw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddusw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddusw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddusw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddusw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddusw:
 ; BTVER2:       # %bb.0:
@@ -2281,10 +2281,10 @@
 ;
 ; BDVER2-LABEL: test_paddw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    paddw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    paddw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    paddw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    paddw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_paddw:
 ; BTVER2:       # %bb.0:
@@ -2366,10 +2366,10 @@
 ;
 ; BDVER2-LABEL: test_palignr:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    palignr $1, %mm1, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    palignr $1, (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    palignr $1, %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    palignr $1, (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_palignr:
 ; BTVER2:       # %bb.0:
@@ -2451,10 +2451,10 @@
 ;
 ; BDVER2-LABEL: test_pand:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pand %mm1, %mm0 # sched: [1:0.33]
-; BDVER2-NEXT:    pand (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pand %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pand (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pand:
 ; BTVER2:       # %bb.0:
@@ -2536,10 +2536,10 @@
 ;
 ; BDVER2-LABEL: test_pandn:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pandn %mm1, %mm0 # sched: [1:0.33]
-; BDVER2-NEXT:    pandn (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pandn %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pandn (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pandn:
 ; BTVER2:       # %bb.0:
@@ -2621,10 +2621,10 @@
 ;
 ; BDVER2-LABEL: test_pavgb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pavgb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pavgb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pavgb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pavgb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pavgb:
 ; BTVER2:       # %bb.0:
@@ -2706,10 +2706,10 @@
 ;
 ; BDVER2-LABEL: test_pavgw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pavgw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pavgw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pavgw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pavgw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pavgw:
 ; BTVER2:       # %bb.0:
@@ -2791,10 +2791,10 @@
 ;
 ; BDVER2-LABEL: test_pcmpeqb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pcmpeqb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pcmpeqb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pcmpeqb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pcmpeqb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pcmpeqb:
 ; BTVER2:       # %bb.0:
@@ -2876,10 +2876,10 @@
 ;
 ; BDVER2-LABEL: test_pcmpeqd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pcmpeqd %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pcmpeqd (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pcmpeqd %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pcmpeqd (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pcmpeqd:
 ; BTVER2:       # %bb.0:
@@ -2961,10 +2961,10 @@
 ;
 ; BDVER2-LABEL: test_pcmpeqw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pcmpeqw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pcmpeqw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pcmpeqw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pcmpeqw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pcmpeqw:
 ; BTVER2:       # %bb.0:
@@ -3046,10 +3046,10 @@
 ;
 ; BDVER2-LABEL: test_pcmpgtb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pcmpgtb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pcmpgtb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pcmpgtb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pcmpgtb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pcmpgtb:
 ; BTVER2:       # %bb.0:
@@ -3131,10 +3131,10 @@
 ;
 ; BDVER2-LABEL: test_pcmpgtd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pcmpgtd %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pcmpgtd (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pcmpgtd %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pcmpgtd (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pcmpgtd:
 ; BTVER2:       # %bb.0:
@@ -3216,10 +3216,10 @@
 ;
 ; BDVER2-LABEL: test_pcmpgtw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pcmpgtw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pcmpgtw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pcmpgtw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pcmpgtw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pcmpgtw:
 ; BTVER2:       # %bb.0:
@@ -3285,8 +3285,8 @@
 ;
 ; BDVER2-LABEL: test_pextrw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pextrw $0, %mm0, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pextrw $0, %mm0, %eax # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pextrw:
 ; BTVER2:       # %bb.0:
@@ -3361,10 +3361,10 @@
 ;
 ; BDVER2-LABEL: test_phaddd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    phaddd %mm1, %mm0 # sched: [3:1.50]
-; BDVER2-NEXT:    phaddd (%rdi), %mm0 # sched: [8:1.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    phaddd %mm1, %mm0 # sched: [5:0.50]
+; BDVER2-NEXT:    phaddd (%rdi), %mm0 # sched: [10:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_phaddd:
 ; BTVER2:       # %bb.0:
@@ -3446,10 +3446,10 @@
 ;
 ; BDVER2-LABEL: test_phaddsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    phaddsw %mm1, %mm0 # sched: [3:1.50]
-; BDVER2-NEXT:    phaddsw (%rdi), %mm0 # sched: [8:1.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    phaddsw %mm1, %mm0 # sched: [5:0.50]
+; BDVER2-NEXT:    phaddsw (%rdi), %mm0 # sched: [10:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_phaddsw:
 ; BTVER2:       # %bb.0:
@@ -3531,10 +3531,10 @@
 ;
 ; BDVER2-LABEL: test_phaddw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    phaddw %mm1, %mm0 # sched: [3:1.50]
-; BDVER2-NEXT:    phaddw (%rdi), %mm0 # sched: [8:1.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    phaddw %mm1, %mm0 # sched: [5:0.50]
+; BDVER2-NEXT:    phaddw (%rdi), %mm0 # sched: [10:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_phaddw:
 ; BTVER2:       # %bb.0:
@@ -3616,10 +3616,10 @@
 ;
 ; BDVER2-LABEL: test_phsubd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    phsubd %mm1, %mm0 # sched: [3:1.50]
-; BDVER2-NEXT:    phsubd (%rdi), %mm0 # sched: [8:1.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    phsubd %mm1, %mm0 # sched: [5:0.50]
+; BDVER2-NEXT:    phsubd (%rdi), %mm0 # sched: [10:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_phsubd:
 ; BTVER2:       # %bb.0:
@@ -3701,10 +3701,10 @@
 ;
 ; BDVER2-LABEL: test_phsubsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    phsubsw %mm1, %mm0 # sched: [3:1.50]
-; BDVER2-NEXT:    phsubsw (%rdi), %mm0 # sched: [8:1.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    phsubsw %mm1, %mm0 # sched: [5:0.50]
+; BDVER2-NEXT:    phsubsw (%rdi), %mm0 # sched: [10:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_phsubsw:
 ; BTVER2:       # %bb.0:
@@ -3786,10 +3786,10 @@
 ;
 ; BDVER2-LABEL: test_phsubw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    phsubw %mm1, %mm0 # sched: [3:1.50]
-; BDVER2-NEXT:    phsubw (%rdi), %mm0 # sched: [8:1.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    phsubw %mm1, %mm0 # sched: [5:0.50]
+; BDVER2-NEXT:    phsubw (%rdi), %mm0 # sched: [10:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_phsubw:
 ; BTVER2:       # %bb.0:
@@ -3879,11 +3879,11 @@
 ;
 ; BDVER2-LABEL: test_pinsrw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pinsrw $0, %edi, %mm0 # sched: [2:1.00]
 ; BDVER2-NEXT:    movswl (%rsi), %eax # sched: [5:0.50]
-; BDVER2-NEXT:    pinsrw $1, %eax, %mm0 # sched: [2:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pinsrw $0, %edi, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pinsrw $1, %eax, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pinsrw:
 ; BTVER2:       # %bb.0:
@@ -3968,10 +3968,10 @@
 ;
 ; BDVER2-LABEL: test_pmaddwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmaddwd %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmaddwd (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmaddwd %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmaddwd (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmaddwd:
 ; BTVER2:       # %bb.0:
@@ -4053,10 +4053,10 @@
 ;
 ; BDVER2-LABEL: test_pmaddubsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmaddubsw %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmaddubsw (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmaddubsw %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmaddubsw (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmaddubsw:
 ; BTVER2:       # %bb.0:
@@ -4138,10 +4138,10 @@
 ;
 ; BDVER2-LABEL: test_pmaxsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmaxsw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pmaxsw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmaxsw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pmaxsw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmaxsw:
 ; BTVER2:       # %bb.0:
@@ -4223,10 +4223,10 @@
 ;
 ; BDVER2-LABEL: test_pmaxub:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmaxub %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pmaxub (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmaxub %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pmaxub (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmaxub:
 ; BTVER2:       # %bb.0:
@@ -4308,10 +4308,10 @@
 ;
 ; BDVER2-LABEL: test_pminsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pminsw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pminsw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pminsw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pminsw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pminsw:
 ; BTVER2:       # %bb.0:
@@ -4393,10 +4393,10 @@
 ;
 ; BDVER2-LABEL: test_pminub:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pminub %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    pminub (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pminub %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pminub (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pminub:
 ; BTVER2:       # %bb.0:
@@ -4462,8 +4462,8 @@
 ;
 ; BDVER2-LABEL: test_pmovmskb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmovmskb %mm0, %eax # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmovmskb %mm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmovmskb:
 ; BTVER2:       # %bb.0:
@@ -4538,10 +4538,10 @@
 ;
 ; BDVER2-LABEL: test_pmulhrsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmulhrsw %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmulhrsw (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmulhrsw %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmulhrsw (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmulhrsw:
 ; BTVER2:       # %bb.0:
@@ -4623,10 +4623,10 @@
 ;
 ; BDVER2-LABEL: test_pmulhw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmulhw %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmulhw (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmulhw %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmulhw (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmulhw:
 ; BTVER2:       # %bb.0:
@@ -4708,10 +4708,10 @@
 ;
 ; BDVER2-LABEL: test_pmulhuw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmulhuw %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmulhuw (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmulhuw %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmulhuw (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmulhuw:
 ; BTVER2:       # %bb.0:
@@ -4793,10 +4793,10 @@
 ;
 ; BDVER2-LABEL: test_pmullw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmullw %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmullw (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmullw %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmullw (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmullw:
 ; BTVER2:       # %bb.0:
@@ -4878,10 +4878,10 @@
 ;
 ; BDVER2-LABEL: test_pmuludq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pmuludq %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    pmuludq (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pmuludq %mm1, %mm0 # sched: [4:1.00]
+; BDVER2-NEXT:    pmuludq (%rdi), %mm0 # sched: [9:1.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pmuludq:
 ; BTVER2:       # %bb.0:
@@ -4963,10 +4963,10 @@
 ;
 ; BDVER2-LABEL: test_por:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    por %mm1, %mm0 # sched: [1:0.33]
-; BDVER2-NEXT:    por (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    por %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    por (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_por:
 ; BTVER2:       # %bb.0:
@@ -5048,10 +5048,10 @@
 ;
 ; BDVER2-LABEL: test_psadbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psadbw %mm1, %mm0 # sched: [5:1.00]
-; BDVER2-NEXT:    psadbw (%rdi), %mm0 # sched: [10:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psadbw %mm1, %mm0 # sched: [4:0.50]
+; BDVER2-NEXT:    psadbw (%rdi), %mm0 # sched: [9:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psadbw:
 ; BTVER2:       # %bb.0:
@@ -5133,10 +5133,10 @@
 ;
 ; BDVER2-LABEL: test_pshufb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pshufb %mm1, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    pshufb (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pshufb %mm1, %mm0 # sched: [3:2.00]
+; BDVER2-NEXT:    pshufb (%rdi), %mm0 # sched: [8:2.00]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pshufb:
 ; BTVER2:       # %bb.0:
@@ -5218,10 +5218,10 @@
 ;
 ; BDVER2-LABEL: test_pshufw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pshufw $0, (%rdi), %mm0 # mm0 = mem[0,0,0,0] sched: [6:1.00]
-; BDVER2-NEXT:    pshufw $0, %mm0, %mm0 # mm0 = mm0[0,0,0,0] sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pshufw $0, (%rdi), %mm0 # mm0 = mem[0,0,0,0] sched: [7:0.50]
+; BDVER2-NEXT:    pshufw $0, %mm0, %mm0 # mm0 = mm0[0,0,0,0] sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pshufw:
 ; BTVER2:       # %bb.0:
@@ -5303,10 +5303,10 @@
 ;
 ; BDVER2-LABEL: test_psignb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psignb %mm1, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    psignb (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psignb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psignb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psignb:
 ; BTVER2:       # %bb.0:
@@ -5388,10 +5388,10 @@
 ;
 ; BDVER2-LABEL: test_psignd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psignd %mm1, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    psignd (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psignd %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psignd (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psignd:
 ; BTVER2:       # %bb.0:
@@ -5473,10 +5473,10 @@
 ;
 ; BDVER2-LABEL: test_psignw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psignw %mm1, %mm0 # sched: [1:0.50]
-; BDVER2-NEXT:    psignw (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psignw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psignw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psignw:
 ; BTVER2:       # %bb.0:
@@ -5566,11 +5566,11 @@
 ;
 ; BDVER2-LABEL: test_pslld:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pslld %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    pslld (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    pslld $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pslld %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    pslld (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    pslld $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pslld:
 ; BTVER2:       # %bb.0:
@@ -5664,11 +5664,11 @@
 ;
 ; BDVER2-LABEL: test_psllq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psllq %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psllq (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psllq $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psllq %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psllq (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psllq $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psllq:
 ; BTVER2:       # %bb.0:
@@ -5762,11 +5762,11 @@
 ;
 ; BDVER2-LABEL: test_psllw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psllw %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psllw (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psllw $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psllw %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psllw (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psllw $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psllw:
 ; BTVER2:       # %bb.0:
@@ -5860,11 +5860,11 @@
 ;
 ; BDVER2-LABEL: test_psrad:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psrad %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psrad (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psrad $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psrad %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psrad (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psrad $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psrad:
 ; BTVER2:       # %bb.0:
@@ -5958,11 +5958,11 @@
 ;
 ; BDVER2-LABEL: test_psraw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psraw %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psraw (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psraw $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psraw %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psraw (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psraw $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psraw:
 ; BTVER2:       # %bb.0:
@@ -6056,11 +6056,11 @@
 ;
 ; BDVER2-LABEL: test_psrld:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psrld %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psrld (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psrld $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psrld %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psrld (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psrld $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psrld:
 ; BTVER2:       # %bb.0:
@@ -6154,11 +6154,11 @@
 ;
 ; BDVER2-LABEL: test_psrlq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psrlq %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psrlq (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psrlq $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psrlq %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psrlq (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psrlq $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psrlq:
 ; BTVER2:       # %bb.0:
@@ -6252,11 +6252,11 @@
 ;
 ; BDVER2-LABEL: test_psrlw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psrlw %mm1, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    psrlw (%rdi), %mm0 # sched: [6:1.00]
-; BDVER2-NEXT:    psrlw $7, %mm0 # sched: [1:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psrlw %mm1, %mm0 # sched: [3:0.50]
+; BDVER2-NEXT:    psrlw (%rdi), %mm0 # sched: [8:0.50]
+; BDVER2-NEXT:    psrlw $7, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psrlw:
 ; BTVER2:       # %bb.0:
@@ -6342,10 +6342,10 @@
 ;
 ; BDVER2-LABEL: test_psubb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubb:
 ; BTVER2:       # %bb.0:
@@ -6427,10 +6427,10 @@
 ;
 ; BDVER2-LABEL: test_psubd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubd %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubd (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubd %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubd (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubd:
 ; BTVER2:       # %bb.0:
@@ -6512,10 +6512,10 @@
 ;
 ; BDVER2-LABEL: test_psubq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubq %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubq (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubq %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubq (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubq:
 ; BTVER2:       # %bb.0:
@@ -6597,10 +6597,10 @@
 ;
 ; BDVER2-LABEL: test_psubsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubsb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubsb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubsb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubsb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubsb:
 ; BTVER2:       # %bb.0:
@@ -6682,10 +6682,10 @@
 ;
 ; BDVER2-LABEL: test_psubsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubsw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubsw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubsw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubsw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubsw:
 ; BTVER2:       # %bb.0:
@@ -6767,10 +6767,10 @@
 ;
 ; BDVER2-LABEL: test_psubusb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubusb %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubusb (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubusb %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubusb (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubusb:
 ; BTVER2:       # %bb.0:
@@ -6852,10 +6852,10 @@
 ;
 ; BDVER2-LABEL: test_psubusw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubusw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubusw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubusw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubusw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubusw:
 ; BTVER2:       # %bb.0:
@@ -6937,10 +6937,10 @@
 ;
 ; BDVER2-LABEL: test_psubw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    psubw %mm1, %mm0 # sched: [3:1.00]
-; BDVER2-NEXT:    psubw (%rdi), %mm0 # sched: [8:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    psubw %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    psubw (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_psubw:
 ; BTVER2:       # %bb.0:
@@ -7022,10 +7022,10 @@
 ;
 ; BDVER2-LABEL: test_punpckhbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    punpckhbw %mm1, %mm0 # mm0 = mm0[4],mm1[4],mm0[5],mm1[5],mm0[6],mm1[6],mm0[7],mm1[7] sched: [1:1.00]
-; BDVER2-NEXT:    punpckhbw (%rdi), %mm0 # mm0 = mm0[4],mem[4],mm0[5],mem[5],mm0[6],mem[6],mm0[7],mem[7] sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    punpckhbw %mm1, %mm0 # mm0 = mm0[4],mm1[4],mm0[5],mm1[5],mm0[6],mm1[6],mm0[7],mm1[7] sched: [2:0.50]
+; BDVER2-NEXT:    punpckhbw (%rdi), %mm0 # mm0 = mm0[4],mem[4],mm0[5],mem[5],mm0[6],mem[6],mm0[7],mem[7] sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_punpckhbw:
 ; BTVER2:       # %bb.0:
@@ -7107,10 +7107,10 @@
 ;
 ; BDVER2-LABEL: test_punpckhdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    punpckhdq %mm1, %mm0 # mm0 = mm0[1],mm1[1] sched: [1:1.00]
-; BDVER2-NEXT:    punpckhdq (%rdi), %mm0 # mm0 = mm0[1],mem[1] sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    punpckhdq %mm1, %mm0 # mm0 = mm0[1],mm1[1] sched: [2:0.50]
+; BDVER2-NEXT:    punpckhdq (%rdi), %mm0 # mm0 = mm0[1],mem[1] sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_punpckhdq:
 ; BTVER2:       # %bb.0:
@@ -7192,10 +7192,10 @@
 ;
 ; BDVER2-LABEL: test_punpckhwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    punpckhwd %mm1, %mm0 # mm0 = mm0[2],mm1[2],mm0[3],mm1[3] sched: [1:1.00]
-; BDVER2-NEXT:    punpckhwd (%rdi), %mm0 # mm0 = mm0[2],mem[2],mm0[3],mem[3] sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    punpckhwd %mm1, %mm0 # mm0 = mm0[2],mm1[2],mm0[3],mm1[3] sched: [2:0.50]
+; BDVER2-NEXT:    punpckhwd (%rdi), %mm0 # mm0 = mm0[2],mem[2],mm0[3],mem[3] sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_punpckhwd:
 ; BTVER2:       # %bb.0:
@@ -7277,10 +7277,10 @@
 ;
 ; BDVER2-LABEL: test_punpcklbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    punpcklbw %mm1, %mm0 # mm0 = mm0[0],mm1[0],mm0[1],mm1[1],mm0[2],mm1[2],mm0[3],mm1[3] sched: [1:1.00]
-; BDVER2-NEXT:    punpcklbw (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1],mm0[2],mem[2],mm0[3],mem[3] sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    punpcklbw %mm1, %mm0 # mm0 = mm0[0],mm1[0],mm0[1],mm1[1],mm0[2],mm1[2],mm0[3],mm1[3] sched: [2:0.50]
+; BDVER2-NEXT:    punpcklbw (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1],mm0[2],mem[2],mm0[3],mem[3] sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_punpcklbw:
 ; BTVER2:       # %bb.0:
@@ -7362,10 +7362,10 @@
 ;
 ; BDVER2-LABEL: test_punpckldq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    punpckldq %mm1, %mm0 # mm0 = mm0[0],mm1[0] sched: [1:1.00]
-; BDVER2-NEXT:    punpckldq (%rdi), %mm0 # mm0 = mm0[0],mem[0] sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    punpckldq %mm1, %mm0 # mm0 = mm0[0],mm1[0] sched: [2:0.50]
+; BDVER2-NEXT:    punpckldq (%rdi), %mm0 # mm0 = mm0[0],mem[0] sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_punpckldq:
 ; BTVER2:       # %bb.0:
@@ -7447,10 +7447,10 @@
 ;
 ; BDVER2-LABEL: test_punpcklwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    punpcklwd %mm1, %mm0 # mm0 = mm0[0],mm1[0],mm0[1],mm1[1] sched: [1:1.00]
-; BDVER2-NEXT:    punpcklwd (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1] sched: [6:1.00]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    punpcklwd %mm1, %mm0 # mm0 = mm0[0],mm1[0],mm0[1],mm1[1] sched: [2:0.50]
+; BDVER2-NEXT:    punpcklwd (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1] sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_punpcklwd:
 ; BTVER2:       # %bb.0:
@@ -7532,10 +7532,10 @@
 ;
 ; BDVER2-LABEL: test_pxor:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pxor %mm1, %mm0 # sched: [1:0.33]
-; BDVER2-NEXT:    pxor (%rdi), %mm0 # sched: [6:0.50]
-; BDVER2-NEXT:    movq %mm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    pxor %mm1, %mm0 # sched: [2:0.50]
+; BDVER2-NEXT:    pxor (%rdi), %mm0 # sched: [7:0.50]
+; BDVER2-NEXT:    movq %mm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pxor:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/popcnt-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/popcnt-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/popcnt-schedule.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell   | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake     | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=knl         | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+popcnt      | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2      | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2      | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1      | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -63,11 +63,11 @@
 ;
 ; BDVER2-LABEL: test_ctpop_i16:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    popcntw (%rsi), %cx # sched: [9:1.00]
-; BDVER2-NEXT:    popcntw %di, %ax # sched: [3:1.00]
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
+; BDVER2-NEXT:    popcntw (%rsi), %cx # sched: [8:0.50]
+; BDVER2-NEXT:    popcntw %di, %ax # sched: [4:0.50]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BDVER2-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ctpop_i16:
 ; BTVER2:       # %bb.0:
@@ -137,10 +137,10 @@
 ;
 ; BDVER2-LABEL: test_ctpop_i32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    popcntl (%rsi), %ecx # sched: [9:1.00]
-; BDVER2-NEXT:    popcntl %edi, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    popcntl (%rsi), %ecx # sched: [8:0.50]
+; BDVER2-NEXT:    popcntl %edi, %eax # sched: [4:0.50]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ctpop_i32:
 ; BTVER2:       # %bb.0:
@@ -208,10 +208,10 @@
 ;
 ; BDVER2-LABEL: test_ctpop_i64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    popcntq (%rsi), %rcx # sched: [9:1.00]
-; BDVER2-NEXT:    popcntq %rdi, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    popcntq (%rsi), %rcx # sched: [8:0.50]
+; BDVER2-NEXT:    popcntq %rdi, %rax # sched: [4:0.50]
+; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ctpop_i64:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/recip-fastmath.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/recip-fastmath.ll
+++ llvm/trunk/test/CodeGen/X86/recip-fastmath.ll
@@ -2,7 +2,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2     | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE-RECIP
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX-RECIP
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=FMA-RECIP
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+fma4 -mattr=+avx -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=btver2 -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -print-schedule | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=SANDY
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=haswell -print-schedule     | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=HASWELL
@@ -40,9 +40,9 @@
 ;
 ; BDVER2-LABEL: f32_no_estimate:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-NEXT:    vdivss %xmm0, %xmm1, %xmm0 # sched: [14:14.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-NEXT:    vdivss %xmm0, %xmm1, %xmm0 # sched: [9:9.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_no_estimate:
 ; BTVER2:       # %bb.0:
@@ -117,7 +117,7 @@
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
 ; BDVER2-NEXT:    vfnmaddss {{.*}}(%rip), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_one_step:
 ; BTVER2:       # %bb.0:
@@ -219,12 +219,12 @@
 ; BDVER2-LABEL: f32_two_step:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm3 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm3, %xmm1, %xmm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_two_step:
 ; BTVER2:       # %bb.0:
@@ -326,9 +326,9 @@
 ;
 ; BDVER2-LABEL: v4f32_no_estimate:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovaps {{.*#+}} xmm1 = [1,1,1,1] sched: [6:0.50]
-; BDVER2-NEXT:    vdivps %xmm0, %xmm1, %xmm0 # sched: [14:14.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} xmm1 = [1,1,1,1] sched: [5:0.50]
+; BDVER2-NEXT:    vdivps %xmm0, %xmm1, %xmm0 # sched: [9:9.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v4f32_no_estimate:
 ; BTVER2:       # %bb.0:
@@ -403,7 +403,7 @@
 ; BDVER2-NEXT:    vrcpps %xmm0, %xmm1 # sched: [5:1.00]
 ; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v4f32_one_step:
 ; BTVER2:       # %bb.0:
@@ -507,12 +507,12 @@
 ; BDVER2-LABEL: v4f32_two_step:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vrcpps %xmm0, %xmm1 # sched: [5:1.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} xmm2 = [1,1,1,1] sched: [6:0.50]
+; BDVER2-NEXT:    vmovaps {{.*#+}} xmm2 = [1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %xmm2, %xmm1, %xmm0, %xmm3 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm3, %xmm1, %xmm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v4f32_two_step:
 ; BTVER2:       # %bb.0:
@@ -617,9 +617,9 @@
 ;
 ; BDVER2-LABEL: v8f32_no_estimate:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
-; BDVER2-NEXT:    vdivps %ymm0, %ymm1, %ymm0 # sched: [29:28.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
+; BDVER2-NEXT:    vdivps %ymm0, %ymm1, %ymm0 # sched: [9:19.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_no_estimate:
 ; BTVER2:       # %bb.0:
@@ -698,10 +698,10 @@
 ;
 ; BDVER2-LABEL: v8f32_one_step:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm0, %ymm1, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_one_step:
 ; BTVER2:       # %bb.0:
@@ -817,13 +817,13 @@
 ;
 ; BDVER2-LABEL: v8f32_two_step:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm3 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm3, %ymm1, %ymm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm0, %ymm1, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_two_step:
 ; BTVER2:       # %bb.0:
@@ -936,10 +936,10 @@
 ;
 ; BDVER2-LABEL: v16f32_no_estimate:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
-; BDVER2-NEXT:    vdivps %ymm0, %ymm2, %ymm0 # sched: [29:28.00]
-; BDVER2-NEXT:    vdivps %ymm1, %ymm2, %ymm1 # sched: [29:28.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
+; BDVER2-NEXT:    vdivps %ymm0, %ymm2, %ymm0 # sched: [9:19.00]
+; BDVER2-NEXT:    vdivps %ymm1, %ymm2, %ymm1 # sched: [9:19.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_no_estimate:
 ; BTVER2:       # %bb.0:
@@ -1045,14 +1045,14 @@
 ;
 ; BDVER2-LABEL: v16f32_one_step:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm4 # sched: [7:2.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm4 # sched: [5:2.00]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm0, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vfmaddps %ymm2, %ymm0, %ymm2, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm4, %ymm1, %ymm1 # sched: [5:0.50]
+; BDVER2-NEXT:    vfmaddps %ymm2, %ymm0, %ymm2, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm4, %ymm1, %ymm4, %ymm1 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_one_step:
 ; BTVER2:       # %bb.0:
@@ -1226,18 +1226,18 @@
 ;
 ; BDVER2-LABEL: v16f32_two_step:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm0, %ymm4 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm4, %ymm2, %ymm2 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm0, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm0, %ymm2, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [7:2.00]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [5:2.00]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm1, %ymm4 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm4, %ymm2, %ymm2 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm1, %ymm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm1, %ymm2, %ymm1 # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_two_step:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/recip-fastmath2.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/recip-fastmath2.ll
+++ llvm/trunk/test/CodeGen/X86/recip-fastmath2.ll
@@ -2,7 +2,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE-RECIP
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -print-schedule       | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX-RECIP
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+fma -print-schedule  | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=FMA-RECIP
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+fma4 -mattr=+avx -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=btver2 -print-schedule      | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -print-schedule | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=SANDY
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=haswell -print-schedule     | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=HASWELL
@@ -34,8 +34,8 @@
 ; BDVER2-LABEL: f32_no_step_2:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_no_step_2:
 ; BTVER2:       # %bb.0:
@@ -113,8 +113,8 @@
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
 ; BDVER2-NEXT:    vfnmaddss {{.*}}(%rip), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_one_step_2:
 ; BTVER2:       # %bb.0:
@@ -216,9 +216,9 @@
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
 ; BDVER2-NEXT:    vfnmaddss {{.*}}(%rip), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm1 # sched: [11:1.00]
+; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm1 # sched: [10:1.00]
 ; BDVER2-NEXT:    vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_one_step_2_divs:
 ; BTVER2:       # %bb.0:
@@ -336,13 +336,13 @@
 ; BDVER2-LABEL: f32_two_step_2:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm3 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm3, %xmm1, %xmm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddss %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: f32_two_step_2:
 ; BTVER2:       # %bb.0:
@@ -465,8 +465,8 @@
 ; BDVER2-NEXT:    vrcpps %xmm0, %xmm1 # sched: [5:1.00]
 ; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v4f32_one_step2:
 ; BTVER2:       # %bb.0:
@@ -570,9 +570,9 @@
 ; BDVER2-NEXT:    vrcpps %xmm0, %xmm1 # sched: [5:1.00]
 ; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %xmm1, %xmm0, %xmm0 # sched: [10:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm1 # sched: [11:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm1 # sched: [10:1.00]
 ; BDVER2-NEXT:    vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v4f32_one_step_2_divs:
 ; BTVER2:       # %bb.0:
@@ -692,13 +692,13 @@
 ; BDVER2-LABEL: v4f32_two_step2:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vrcpps %xmm0, %xmm1 # sched: [5:1.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} xmm2 = [1,1,1,1] sched: [6:0.50]
+; BDVER2-NEXT:    vmovaps {{.*#+}} xmm2 = [1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %xmm2, %xmm1, %xmm0, %xmm3 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm3, %xmm1, %xmm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %xmm1, %xmm0, %xmm1, %xmm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v4f32_two_step2:
 ; BTVER2:       # %bb.0:
@@ -826,11 +826,11 @@
 ;
 ; BDVER2-LABEL: v8f32_one_step2:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm0, %ymm1, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_one_step2:
 ; BTVER2:       # %bb.0:
@@ -940,12 +940,12 @@
 ;
 ; BDVER2-LABEL: v8f32_one_step_2_divs:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    vfnmaddps {{.*}}(%rip), %ymm1, %ymm0, %ymm0 # sched: [10:1.00]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm0, %ymm1, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm1 # sched: [12:1.00]
-; BDVER2-NEXT:    vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm1 # sched: [10:2.00]
+; BDVER2-NEXT:    vmulps %ymm0, %ymm1, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_one_step_2_divs:
 ; BTVER2:       # %bb.0:
@@ -1078,14 +1078,14 @@
 ;
 ; BDVER2-LABEL: v8f32_two_step2:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm3 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm3, %ymm1, %ymm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm1, %ymm0, %ymm1, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_two_step2:
 ; BTVER2:       # %bb.0:
@@ -1190,8 +1190,8 @@
 ;
 ; BDVER2-LABEL: v8f32_no_step:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [7:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_no_step:
 ; BTVER2:       # %bb.0:
@@ -1249,9 +1249,9 @@
 ;
 ; BDVER2-LABEL: v8f32_no_step2:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [7:2.00]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v8f32_no_step2:
 ; BTVER2:       # %bb.0:
@@ -1361,16 +1361,16 @@
 ;
 ; BDVER2-LABEL: v16f32_one_step2:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm4 # sched: [7:2.00]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm4 # sched: [5:2.00]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm1, %ymm1 # sched: [5:0.50]
-; BDVER2-NEXT:    vfmaddps %ymm2, %ymm1, %ymm2, %ymm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm4, %ymm0, %ymm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vfmaddps %ymm2, %ymm1, %ymm2, %ymm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm4, %ymm0, %ymm4, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm1 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm1 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_one_step2:
 ; BTVER2:       # %bb.0:
@@ -1532,18 +1532,18 @@
 ;
 ; BDVER2-LABEL: v16f32_one_step_2_divs:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm0, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm0, %ymm2, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [7:2.00]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [5:2.00]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm1, %ymm1 # sched: [5:0.50]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm3 # sched: [10:2.00]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm1, %ymm2, %ymm1 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm2 # sched: [12:1.00]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm3 # sched: [12:1.00]
-; BDVER2-NEXT:    vmulps %ymm0, %ymm3, %ymm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulps %ymm1, %ymm2, %ymm1 # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm2 # sched: [10:2.00]
+; BDVER2-NEXT:    vmulps %ymm0, %ymm3, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmulps %ymm1, %ymm2, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_one_step_2_divs:
 ; BTVER2:       # %bb.0:
@@ -1745,20 +1745,20 @@
 ;
 ; BDVER2-LABEL: v16f32_two_step2:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [7:2.00]
-; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [7:0.50]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm2 # sched: [5:2.00]
+; BDVER2-NEXT:    vmovaps {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1] sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm1, %ymm4 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm4, %ymm2, %ymm2 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm1, %ymm1 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm1, %ymm2, %ymm1 # sched: [5:0.50]
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [7:2.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm2 # sched: [5:2.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm1 # sched: [10:2.00]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm0, %ymm4 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm4, %ymm2, %ymm2 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfnmaddps %ymm3, %ymm2, %ymm0, %ymm0 # sched: [5:0.50]
 ; BDVER2-NEXT:    vfmaddps %ymm2, %ymm0, %ymm2, %ymm0 # sched: [5:0.50]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm1 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_two_step2:
 ; BTVER2:       # %bb.0:
@@ -1904,9 +1904,9 @@
 ;
 ; BDVER2-LABEL: v16f32_no_step:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [7:2.00]
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_no_step:
 ; BTVER2:       # %bb.0:
@@ -1976,11 +1976,11 @@
 ;
 ; BDVER2-LABEL: v16f32_no_step2:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vrcpps %ymm1, %ymm1 # sched: [7:2.00]
-; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [7:2.00]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
-; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm1 # sched: [12:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vrcpps %ymm1, %ymm1 # sched: [5:2.00]
+; BDVER2-NEXT:    vrcpps %ymm0, %ymm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [10:2.00]
+; BDVER2-NEXT:    vmulps {{.*}}(%rip), %ymm1, %ymm1 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: v16f32_no_step2:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/schedule-x86-64-shld.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/schedule-x86-64-shld.ll
+++ llvm/trunk/test/CodeGen/X86/schedule-x86-64-shld.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=+slow-shld | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER12 --check-prefix=BDVER1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=+slow-shld | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER12 --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver1 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER12 --check-prefix=BDVER1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER12 --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 
 
@@ -19,9 +19,9 @@
 ;
 ; BDVER12-LABEL: lshift10_optsize:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    shldq $10, %rsi, %rax # sched: [2:0.67]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    shldq $10, %rsi, %rax # sched: [4:3.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift10_optsize:
 ; BTVER2:       # %bb.0: # %entry
@@ -47,7 +47,7 @@
 ; BDVER12-NEXT:    shlq $10, %rdi # sched: [1:0.50]
 ; BDVER12-NEXT:    shrq $54, %rsi # sched: [1:0.50]
 ; BDVER12-NEXT:    leaq (%rsi,%rdi), %rax # sched: [1:0.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift10:
 ; BTVER2:       # %bb.0: # %entry
@@ -77,9 +77,9 @@
 ;
 ; BDVER12-LABEL: rshift10_optsize:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    shrdq $62, %rsi, %rax # sched: [2:0.67]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    shrdq $62, %rsi, %rax # sched: [4:3.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: rshift10_optsize:
 ; BTVER2:       # %bb.0: # %entry
@@ -105,7 +105,7 @@
 ; BDVER12:       # %bb.0: # %entry
 ; BDVER12-NEXT:    shrq $62, %rdi # sched: [1:0.50]
 ; BDVER12-NEXT:    leaq (%rdi,%rsi,4), %rax # sched: [1:0.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: rshift10:
 ; BTVER2:       # %bb.0: # %entry
@@ -135,11 +135,11 @@
 ;
 ; BDVER12-LABEL: lshift_cl_optsize:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.33]
+; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.50]
 ; BDVER12-NEXT:    # kill: def $cl killed $cl killed $rcx
-; BDVER12-NEXT:    shldq %cl, %rsi, %rax # sched: [4:1.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shldq %cl, %rsi, %rax # sched: [4:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_cl_optsize:
 ; BTVER2:       # %bb.0: # %entry
@@ -167,14 +167,14 @@
 ;
 ; BDVER12-LABEL: lshift_cl:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rsi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    shlq %cl, %rdi # sched: [3:1.50]
-; BDVER12-NEXT:    negl %ecx # sched: [1:0.33]
+; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rsi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    shlq %cl, %rdi # sched: [1:0.50]
+; BDVER12-NEXT:    negl %ecx # sched: [1:0.50]
 ; BDVER12-NEXT:    # kill: def $cl killed $cl killed $rcx
-; BDVER12-NEXT:    shrq %cl, %rax # sched: [3:1.50]
-; BDVER12-NEXT:    orq %rdi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shrq %cl, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    orq %rdi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_cl:
 ; BTVER2:       # %bb.0: # %entry
@@ -211,11 +211,11 @@
 ;
 ; BDVER12-LABEL: rshift_cl_optsize:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.33]
+; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rdi, %rax # sched: [1:0.50]
 ; BDVER12-NEXT:    # kill: def $cl killed $cl killed $rcx
-; BDVER12-NEXT:    shrdq %cl, %rsi, %rax # sched: [4:1.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shrdq %cl, %rsi, %rax # sched: [4:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: rshift_cl_optsize:
 ; BTVER2:       # %bb.0: # %entry
@@ -243,14 +243,14 @@
 ;
 ; BDVER12-LABEL: rshift_cl:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rsi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    shrq %cl, %rdi # sched: [3:1.50]
-; BDVER12-NEXT:    negl %ecx # sched: [1:0.33]
+; BDVER12-NEXT:    movq %rdx, %rcx # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rsi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    shrq %cl, %rdi # sched: [1:0.50]
+; BDVER12-NEXT:    negl %ecx # sched: [1:0.50]
 ; BDVER12-NEXT:    # kill: def $cl killed $cl killed $rcx
-; BDVER12-NEXT:    shlq %cl, %rax # sched: [3:1.50]
-; BDVER12-NEXT:    orq %rdi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shlq %cl, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    orq %rdi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: rshift_cl:
 ; BTVER2:       # %bb.0: # %entry
@@ -287,10 +287,10 @@
 ;
 ; BDVER12-LABEL: lshift_mem_cl_optsize:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rsi, %rcx # sched: [1:0.33]
+; BDVER12-NEXT:    movq %rsi, %rcx # sched: [1:0.50]
 ; BDVER12-NEXT:    # kill: def $cl killed $cl killed $rcx
-; BDVER12-NEXT:    shldq %cl, %rdi, {{.*}}(%rip) # sched: [10:1.50]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shldq %cl, %rdi, {{.*}}(%rip) # sched: [4:11.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_mem_cl_optsize:
 ; BTVER2:       # %bb.0: # %entry
@@ -318,15 +318,15 @@
 ;
 ; BDVER12-LABEL: lshift_mem_cl:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    movq %rsi, %rcx # sched: [1:0.33]
 ; BDVER12-NEXT:    movq {{.*}}(%rip), %rax # sched: [5:0.50]
-; BDVER12-NEXT:    shlq %cl, %rax # sched: [3:1.50]
-; BDVER12-NEXT:    negl %ecx # sched: [1:0.33]
+; BDVER12-NEXT:    movq %rsi, %rcx # sched: [1:0.50]
+; BDVER12-NEXT:    shlq %cl, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    negl %ecx # sched: [1:0.50]
 ; BDVER12-NEXT:    # kill: def $cl killed $cl killed $rcx
-; BDVER12-NEXT:    shrq %cl, %rdi # sched: [3:1.50]
-; BDVER12-NEXT:    orq %rax, %rdi # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rdi, {{.*}}(%rip) # sched: [1:1.00]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shrq %cl, %rdi # sched: [1:0.50]
+; BDVER12-NEXT:    orq %rax, %rdi # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rdi, {{.*}}(%rip) # sched: [1:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_mem_cl:
 ; BTVER2:       # %bb.0: # %entry
@@ -358,11 +358,11 @@
 ; BDVER12-LABEL: lshift_mem:
 ; BDVER12:       # %bb.0: # %entry
 ; BDVER12-NEXT:    movq {{.*}}(%rip), %rax # sched: [5:0.50]
-; BDVER12-NEXT:    shlq $10, %rax # sched: [1:0.50]
 ; BDVER12-NEXT:    shrq $54, %rdi # sched: [1:0.50]
-; BDVER12-NEXT:    orq %rax, %rdi # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rdi, {{.*}}(%rip) # sched: [1:1.00]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shlq $10, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    orq %rax, %rdi # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rdi, {{.*}}(%rip) # sched: [1:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_mem:
 ; BTVER2:       # %bb.0: # %entry
@@ -389,8 +389,8 @@
 ;
 ; BDVER12-LABEL: lshift_mem_optsize:
 ; BDVER12:       # %bb.0: # %entry
-; BDVER12-NEXT:    shldq $10, %rdi, {{.*}}(%rip) # sched: [8:1.00]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shldq $10, %rdi, {{.*}}(%rip) # sched: [4:11.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_mem_optsize:
 ; BTVER2:       # %bb.0: # %entry
@@ -418,9 +418,9 @@
 ; BDVER12-NEXT:    movq {{.*}}(%rip), %rax # sched: [5:0.50]
 ; BDVER12-NEXT:    shlq $10, %rdi # sched: [1:0.50]
 ; BDVER12-NEXT:    shrq $54, %rax # sched: [1:0.50]
-; BDVER12-NEXT:    orq %rdi, %rax # sched: [1:0.33]
-; BDVER12-NEXT:    movq %rax, {{.*}}(%rip) # sched: [1:1.00]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    orq %rdi, %rax # sched: [1:0.50]
+; BDVER12-NEXT:    movq %rax, {{.*}}(%rip) # sched: [1:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_mem_b:
 ; BTVER2:       # %bb.0: # %entry
@@ -450,9 +450,9 @@
 ; BDVER12-LABEL: lshift_mem_b_optsize:
 ; BDVER12:       # %bb.0: # %entry
 ; BDVER12-NEXT:    movq {{.*}}(%rip), %rax # sched: [5:0.50]
-; BDVER12-NEXT:    shrdq $54, %rdi, %rax # sched: [2:0.67]
-; BDVER12-NEXT:    movq %rax, {{.*}}(%rip) # sched: [1:1.00]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    shrdq $54, %rdi, %rax # sched: [4:3.00]
+; BDVER12-NEXT:    movq %rax, {{.*}}(%rip) # sched: [1:0.50]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: lshift_mem_b_optsize:
 ; BTVER2:       # %bb.0: # %entry
Index: llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll
+++ llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
-; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -81,9 +81,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    aaa # sched: [100:0.33]
+; BDVER2-NEXT:    aaa # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_aaa:
 ; BTVER2:       # %bb.0:
@@ -181,10 +181,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    aad # sched: [100:0.33]
-; BDVER2-NEXT:    aad $16 # sched: [100:0.33]
+; BDVER2-NEXT:    aad # sched: [100:0.50]
+; BDVER2-NEXT:    aad $16 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_aad:
 ; BTVER2:       # %bb.0:
@@ -284,10 +284,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    aam # sched: [100:0.33]
-; BDVER2-NEXT:    aam $16 # sched: [100:0.33]
+; BDVER2-NEXT:    aam # sched: [100:0.50]
+; BDVER2-NEXT:    aam $16 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_aam:
 ; BTVER2:       # %bb.0:
@@ -379,9 +379,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    aas # sched: [100:0.33]
+; BDVER2-NEXT:    aas # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_aas:
 ; BTVER2:       # %bb.0:
@@ -480,9 +480,9 @@
 ; BDVER2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    arpl %ax, (%ecx) # sched: [100:0.33]
+; BDVER2-NEXT:    arpl %ax, (%ecx) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_arpl:
 ; BTVER2:       # %bb.0:
@@ -644,7 +644,7 @@
 ;
 ; BDVER2-LABEL: test_bound:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    pushl %esi # sched: [5:1.00]
+; BDVER2-NEXT:    pushl %esi # sched: [1:0.50]
 ; BDVER2-NEXT:    .cfi_def_cfa_offset 8
 ; BDVER2-NEXT:    .cfi_offset %esi, -8
 ; BDVER2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
@@ -652,12 +652,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %edx # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %esi # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bound %ax, (%esi) # sched: [100:0.33]
-; BDVER2-NEXT:    bound %ecx, (%edx) # sched: [100:0.33]
+; BDVER2-NEXT:    bound %ax, (%esi) # sched: [100:0.50]
+; BDVER2-NEXT:    bound %ecx, (%edx) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    popl %esi # sched: [6:0.50]
+; BDVER2-NEXT:    popl %esi # sched: [5:0.50]
 ; BDVER2-NEXT:    .cfi_def_cfa_offset 4
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bound:
 ; BTVER2:       # %bb.0:
@@ -767,9 +767,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    daa # sched: [100:0.33]
+; BDVER2-NEXT:    daa # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_daa:
 ; BTVER2:       # %bb.0:
@@ -859,9 +859,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    das # sched: [100:0.33]
+; BDVER2-NEXT:    das # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_das:
 ; BTVER2:       # %bb.0:
@@ -968,10 +968,10 @@
 ; BDVER2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    decw %ax # sched: [1:0.33]
-; BDVER2-NEXT:    decw (%ecx) # sched: [7:1.00]
+; BDVER2-NEXT:    decw %ax # sched: [1:0.50]
+; BDVER2-NEXT:    decw (%ecx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dec16:
 ; BTVER2:       # %bb.0:
@@ -1081,10 +1081,10 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    decl %eax # sched: [1:0.33]
-; BDVER2-NEXT:    decl (%ecx) # sched: [7:1.00]
+; BDVER2-NEXT:    decl %eax # sched: [1:0.50]
+; BDVER2-NEXT:    decl (%ecx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dec32:
 ; BTVER2:       # %bb.0:
@@ -1195,10 +1195,10 @@
 ; BDVER2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    incw %ax # sched: [1:0.33]
-; BDVER2-NEXT:    incw (%ecx) # sched: [7:1.00]
+; BDVER2-NEXT:    incw %ax # sched: [1:0.50]
+; BDVER2-NEXT:    incw (%ecx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_inc16:
 ; BTVER2:       # %bb.0:
@@ -1308,10 +1308,10 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    incl %eax # sched: [1:0.33]
-; BDVER2-NEXT:    incl (%ecx) # sched: [7:1.00]
+; BDVER2-NEXT:    incl %eax # sched: [1:0.50]
+; BDVER2-NEXT:    incl (%ecx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_inc32:
 ; BTVER2:       # %bb.0:
@@ -1396,9 +1396,9 @@
 ; BDVER2-LABEL: test_into:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    into # sched: [100:0.33]
+; BDVER2-NEXT:    into # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_into:
 ; BTVER2:       # %bb.0:
@@ -1496,10 +1496,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:  JXTGT:
-; BDVER2-NEXT:    jcxz JXTGT # sched: [2:1.00]
-; BDVER2-NEXT:    jecxz JXTGT # sched: [2:1.00]
+; BDVER2-NEXT:    jcxz JXTGT # sched: [1:1.00]
+; BDVER2-NEXT:    jecxz JXTGT # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_jcxz_jecxz:
 ; BTVER2:       # %bb.0:
@@ -1584,9 +1584,9 @@
 ; BDVER2-LABEL: test_leave:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    leave # sched: [7:0.67]
+; BDVER2-NEXT:    leave # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_leave:
 ; BTVER2:       # %bb.0:
@@ -1747,19 +1747,19 @@
 ; BDVER2-LABEL: test_pop_push:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    popl %ds # sched: [100:0.33]
-; BDVER2-NEXT:    popl %es # sched: [100:0.33]
-; BDVER2-NEXT:    popl %ss # sched: [100:0.33]
-; BDVER2-NEXT:    popl %fs # sched: [100:0.33]
-; BDVER2-NEXT:    popl %gs # sched: [100:0.33]
-; BDVER2-NEXT:    pushl %cs # sched: [100:0.33]
-; BDVER2-NEXT:    pushl %ds # sched: [100:0.33]
-; BDVER2-NEXT:    pushl %es # sched: [100:0.33]
-; BDVER2-NEXT:    pushl %ss # sched: [100:0.33]
-; BDVER2-NEXT:    pushl %fs # sched: [100:0.33]
-; BDVER2-NEXT:    pushl %gs # sched: [100:0.33]
+; BDVER2-NEXT:    popl %ds # sched: [100:0.50]
+; BDVER2-NEXT:    popl %es # sched: [100:0.50]
+; BDVER2-NEXT:    popl %ss # sched: [100:0.50]
+; BDVER2-NEXT:    popl %fs # sched: [100:0.50]
+; BDVER2-NEXT:    popl %gs # sched: [100:0.50]
+; BDVER2-NEXT:    pushl %cs # sched: [100:0.50]
+; BDVER2-NEXT:    pushl %ds # sched: [100:0.50]
+; BDVER2-NEXT:    pushl %es # sched: [100:0.50]
+; BDVER2-NEXT:    pushl %ss # sched: [100:0.50]
+; BDVER2-NEXT:    pushl %fs # sched: [100:0.50]
+; BDVER2-NEXT:    pushl %gs # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pop_push:
 ; BTVER2:       # %bb.0:
@@ -1922,15 +1922,15 @@
 ; BDVER2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    popw %ax # sched: [6:0.50]
-; BDVER2-NEXT:    popw (%ecx) # sched: [6:0.50]
-; BDVER2-NEXT:    pushw %ax # sched: [5:1.00]
-; BDVER2-NEXT:    pushw (%ecx) # sched: [5:1.00]
+; BDVER2-NEXT:    popw %ax # sched: [5:0.50]
+; BDVER2-NEXT:    popw (%ecx) # sched: [6:1.00]
+; BDVER2-NEXT:    pushw %ax # sched: [1:0.50]
+; BDVER2-NEXT:    pushw (%ecx) # sched: [6:1.00]
 ; BDVER2-NEXT:    pushw $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [1:1.00]
-; BDVER2-NEXT:    pushw $7 # sched: [1:1.00]
+; BDVER2-NEXT:    # sched: [1:0.50]
+; BDVER2-NEXT:    pushw $7 # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pop_push_16:
 ; BTVER2:       # %bb.0:
@@ -2089,15 +2089,15 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    popl %eax # sched: [6:0.50]
-; BDVER2-NEXT:    popl (%ecx) # sched: [6:0.50]
-; BDVER2-NEXT:    pushl %eax # sched: [5:1.00]
-; BDVER2-NEXT:    pushl (%ecx) # sched: [5:1.00]
+; BDVER2-NEXT:    popl %eax # sched: [5:0.50]
+; BDVER2-NEXT:    popl (%ecx) # sched: [6:1.00]
+; BDVER2-NEXT:    pushl %eax # sched: [1:0.50]
+; BDVER2-NEXT:    pushl (%ecx) # sched: [6:1.00]
 ; BDVER2-NEXT:    pushl $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [1:1.00]
-; BDVER2-NEXT:    pushl $7 # sched: [1:1.00]
+; BDVER2-NEXT:    # sched: [1:0.50]
+; BDVER2-NEXT:    pushl $7 # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pop_push_32:
 ; BTVER2:       # %bb.0:
@@ -2218,10 +2218,10 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    popal # sched: [5:0.50]
 ; BDVER2-NEXT:    popfl # sched: [5:0.50]
-; BDVER2-NEXT:    pushal # sched: [1:1.00]
-; BDVER2-NEXT:    pushfl # sched: [1:1.00]
+; BDVER2-NEXT:    pushal # sched: [1:0.50]
+; BDVER2-NEXT:    pushfl # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_popa_popf_pusha_pushf:
 ; BTVER2:       # %bb.0:
@@ -2344,14 +2344,14 @@
 ; BDVER2-LABEL: test_ret:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ; BDVER2-NEXT:    retl $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [6:1.00]
-; BDVER2-NEXT:    lretl # sched: [6:1.00]
+; BDVER2-NEXT:    # sched: [5:1.00]
+; BDVER2-NEXT:    lretl # sched: [5:1.00]
 ; BDVER2-NEXT:    lretl $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ret:
 ; BTVER2:       # %bb.0:
@@ -2440,9 +2440,9 @@
 ; BDVER2-LABEL: test_salc:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    salc # sched: [1:0.33]
+; BDVER2-NEXT:    salc # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_salc:
 ; BTVER2:       # %bb.0:
@@ -2567,11 +2567,11 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %edx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xchgl %eax, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    xchgl %ecx, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    xchgl %eax, (%edx) # sched: [6:1.00]
+; BDVER2-NEXT:    xchgl %eax, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    xchgl %ecx, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    xchgl %eax, (%edx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xchg_32:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll
+++ llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -112,14 +112,14 @@
 ; BDVER2-LABEL: test_adc_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    adcb $7, %al # sched: [2:0.67]
-; BDVER2-NEXT:    adcb $7, %dil # sched: [2:0.67]
-; BDVER2-NEXT:    adcb $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcb %dl, %dil # sched: [2:0.67]
-; BDVER2-NEXT:    adcb %dil, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcb (%rsi), %dil # sched: [7:0.67]
+; BDVER2-NEXT:    adcb $7, %al # sched: [1:1.00]
+; BDVER2-NEXT:    adcb $7, %dil # sched: [1:1.00]
+; BDVER2-NEXT:    adcb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcb %dl, %dil # sched: [1:1.00]
+; BDVER2-NEXT:    adcb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcb (%rsi), %dil # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_adc_8:
 ; BTVER2:       # %bb.0:
@@ -288,18 +288,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    adcw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    adcw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    adcw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [9:1.00]
-; BDVER2-NEXT:    adcw $7, %di # sched: [2:0.67]
-; BDVER2-NEXT:    adcw $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcw %dx, %di # sched: [2:0.67]
-; BDVER2-NEXT:    adcw %di, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcw (%rsi), %di # sched: [7:0.67]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    adcw $7, %di # sched: [1:1.00]
+; BDVER2-NEXT:    adcw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcw %dx, %di # sched: [1:1.00]
+; BDVER2-NEXT:    adcw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcw (%rsi), %di # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_adc_16:
 ; BTVER2:       # %bb.0:
@@ -478,18 +478,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    adcl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    adcl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    adcl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [9:1.00]
-; BDVER2-NEXT:    adcl $7, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    adcl $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcl %edx, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    adcl %edi, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcl (%rsi), %edi # sched: [7:0.67]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    adcl $7, %edi # sched: [1:1.00]
+; BDVER2-NEXT:    adcl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcl %edx, %edi # sched: [1:1.00]
+; BDVER2-NEXT:    adcl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcl (%rsi), %edi # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_adc_32:
 ; BTVER2:       # %bb.0:
@@ -668,18 +668,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    adcq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    adcq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    adcq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [9:1.00]
-; BDVER2-NEXT:    adcq $7, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    adcq $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcq %rdx, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    adcq %rdi, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    adcq (%rsi), %rdi # sched: [7:0.67]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    adcq $7, %rdi # sched: [1:1.00]
+; BDVER2-NEXT:    adcq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcq %rdx, %rdi # sched: [1:1.00]
+; BDVER2-NEXT:    adcq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    adcq (%rsi), %rdi # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_adc_64:
 ; BTVER2:       # %bb.0:
@@ -818,14 +818,14 @@
 ; BDVER2-LABEL: test_add_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    addb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    addb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    addb $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addb %dl, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    addb %dil, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addb (%rsi), %dil # sched: [6:0.50]
+; BDVER2-NEXT:    addb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    addb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    addb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addb %dl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    addb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addb (%rsi), %dil # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_add_8:
 ; BTVER2:       # %bb.0:
@@ -994,18 +994,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    addw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    addw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    addw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    addw $7, %di # sched: [1:0.33]
-; BDVER2-NEXT:    addw $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addw %dx, %di # sched: [1:0.33]
-; BDVER2-NEXT:    addw %di, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addw (%rsi), %di # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    addw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    addw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addw %dx, %di # sched: [1:0.50]
+; BDVER2-NEXT:    addw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addw (%rsi), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_add_16:
 ; BTVER2:       # %bb.0:
@@ -1184,18 +1184,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    addl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    addl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    addl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    addl $7, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    addl $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addl %edx, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    addl %edi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addl (%rsi), %edi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    addl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    addl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addl %edx, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    addl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addl (%rsi), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_add_32:
 ; BTVER2:       # %bb.0:
@@ -1374,18 +1374,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    addq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    addq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    addq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    addq $7, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    addq $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addq %rdx, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    addq %rdi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    addq (%rsi), %rdi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    addq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    addq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addq %rdx, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    addq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    addq (%rsi), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_add_64:
 ; BTVER2:       # %bb.0:
@@ -1524,14 +1524,14 @@
 ; BDVER2-LABEL: test_and_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    andb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    andb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    andb $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andb %dl, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    andb %dil, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andb (%rsi), %dil # sched: [6:0.50]
+; BDVER2-NEXT:    andb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    andb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    andb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andb %dl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    andb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andb (%rsi), %dil # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_and_8:
 ; BTVER2:       # %bb.0:
@@ -1700,18 +1700,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    andw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    andw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    andw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    andw $7, %di # sched: [1:0.33]
-; BDVER2-NEXT:    andw $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andw %dx, %di # sched: [1:0.33]
-; BDVER2-NEXT:    andw %di, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andw (%rsi), %di # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    andw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    andw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andw %dx, %di # sched: [1:0.50]
+; BDVER2-NEXT:    andw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andw (%rsi), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_and_16:
 ; BTVER2:       # %bb.0:
@@ -1890,18 +1890,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    andl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    andl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    andl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    andl $7, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    andl $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andl %edx, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    andl %edi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andl (%rsi), %edi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    andl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    andl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andl %edx, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    andl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andl (%rsi), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_and_32:
 ; BTVER2:       # %bb.0:
@@ -2080,18 +2080,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    andq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    andq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    andq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    andq $7, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    andq $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andq %rdx, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    andq %rdi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    andq (%rsi), %rdi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    andq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    andq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andq %rdx, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    andq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    andq (%rsi), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_and_64:
 ; BTVER2:       # %bb.0:
@@ -2214,12 +2214,12 @@
 ; BDVER2-LABEL: test_bsf16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bsfw %di, %ax # sched: [3:1.00]
-; BDVER2-NEXT:    bsfw (%rsi), %cx # sched: [8:1.00]
+; BDVER2-NEXT:    bsfw %di, %ax # sched: [3:2.00]
+; BDVER2-NEXT:    bsfw (%rsi), %cx # sched: [7:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BDVER2-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bsf16:
 ; BTVER2:       # %bb.0:
@@ -2322,11 +2322,11 @@
 ; BDVER2-LABEL: test_bsf32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bsfl %edi, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    bsfl (%rsi), %ecx # sched: [8:1.00]
+; BDVER2-NEXT:    bsfl %edi, %eax # sched: [3:2.00]
+; BDVER2-NEXT:    bsfl (%rsi), %ecx # sched: [7:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bsf32:
 ; BTVER2:       # %bb.0:
@@ -2427,11 +2427,11 @@
 ; BDVER2-LABEL: test_bsf64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bsfq %rdi, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    bsfq (%rsi), %rcx # sched: [8:1.00]
+; BDVER2-NEXT:    bsfq %rdi, %rax # sched: [3:2.00]
+; BDVER2-NEXT:    bsfq (%rsi), %rcx # sched: [7:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bsf64:
 ; BTVER2:       # %bb.0:
@@ -2541,12 +2541,12 @@
 ; BDVER2-LABEL: test_bsr16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bsrw %di, %ax # sched: [3:1.00]
-; BDVER2-NEXT:    bsrw (%rsi), %cx # sched: [8:1.00]
+; BDVER2-NEXT:    bsrw %di, %ax # sched: [4:2.00]
+; BDVER2-NEXT:    bsrw (%rsi), %cx # sched: [8:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
 ; BDVER2-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bsr16:
 ; BTVER2:       # %bb.0:
@@ -2649,11 +2649,11 @@
 ; BDVER2-LABEL: test_bsr32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bsrl %edi, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    bsrl (%rsi), %ecx # sched: [8:1.00]
+; BDVER2-NEXT:    bsrl %edi, %eax # sched: [4:2.00]
+; BDVER2-NEXT:    bsrl (%rsi), %ecx # sched: [8:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    orl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bsr32:
 ; BTVER2:       # %bb.0:
@@ -2754,11 +2754,11 @@
 ; BDVER2-LABEL: test_bsr64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    bsrq %rdi, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    bsrq (%rsi), %rcx # sched: [8:1.00]
+; BDVER2-NEXT:    bsrq %rdi, %rax # sched: [4:2.00]
+; BDVER2-NEXT:    bsrq (%rsi), %rcx # sched: [8:2.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bsr64:
 ; BTVER2:       # %bb.0:
@@ -2835,9 +2835,9 @@
 ;
 ; BDVER2-LABEL: test_bswap32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.33]
+; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.50]
 ; BDVER2-NEXT:    bswapl %eax # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bswap32:
 ; BTVER2:       # %bb.0:
@@ -2904,9 +2904,9 @@
 ;
 ; BDVER2-LABEL: test_bswap64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    bswapq %rax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    bswapq %rax # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bswap64:
 ; BTVER2:       # %bb.0:
@@ -3104,23 +3104,23 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    btw %si, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btcw %si, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btrw %si, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btsw %si, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btw %si, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btcw %si, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btrw %si, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btsw %si, (%rdx) # sched: [9:1.00]
+; BDVER2-NEXT:    btcw %si, %di # sched: [2:0.50]
+; BDVER2-NEXT:    btrw %si, %di # sched: [2:0.50]
+; BDVER2-NEXT:    btsw %si, %di # sched: [2:0.50]
+; BDVER2-NEXT:    btw %si, (%rdx) # sched: [5:0.50]
+; BDVER2-NEXT:    btcw %si, (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    btrw %si, (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    btsw %si, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btw $7, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btcw $7, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btrw $7, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btsw $7, %di # sched: [1:0.50]
-; BDVER2-NEXT:    btw $7, (%rdx) # sched: [6:0.50]
+; BDVER2-NEXT:    btcw $7, %di # sched: [2:0.50]
+; BDVER2-NEXT:    btrw $7, %di # sched: [2:0.50]
+; BDVER2-NEXT:    btsw $7, %di # sched: [2:0.50]
+; BDVER2-NEXT:    btw $7, (%rdx) # sched: [5:0.50]
 ; BDVER2-NEXT:    btcw $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btrw $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btsw $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bt_btc_btr_bts_16:
 ; BTVER2:       # %bb.0:
@@ -3349,23 +3349,23 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    btl %esi, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btcl %esi, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btrl %esi, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btsl %esi, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btl %esi, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btcl %esi, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btrl %esi, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btsl %esi, (%rdx) # sched: [9:1.00]
+; BDVER2-NEXT:    btcl %esi, %edi # sched: [2:0.50]
+; BDVER2-NEXT:    btrl %esi, %edi # sched: [2:0.50]
+; BDVER2-NEXT:    btsl %esi, %edi # sched: [2:0.50]
+; BDVER2-NEXT:    btl %esi, (%rdx) # sched: [5:0.50]
+; BDVER2-NEXT:    btcl %esi, (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    btrl %esi, (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    btsl %esi, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btl $7, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btcl $7, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btrl $7, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btsl $7, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    btl $7, (%rdx) # sched: [6:0.50]
+; BDVER2-NEXT:    btcl $7, %edi # sched: [2:0.50]
+; BDVER2-NEXT:    btrl $7, %edi # sched: [2:0.50]
+; BDVER2-NEXT:    btsl $7, %edi # sched: [2:0.50]
+; BDVER2-NEXT:    btl $7, (%rdx) # sched: [5:0.50]
 ; BDVER2-NEXT:    btcl $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btrl $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btsl $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bt_btc_btr_bts_32:
 ; BTVER2:       # %bb.0:
@@ -3594,23 +3594,23 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    btq %rsi, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btcq %rsi, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btrq %rsi, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btsq %rsi, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btq %rsi, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btcq %rsi, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btrq %rsi, (%rdx) # sched: [9:1.00]
-; BDVER2-NEXT:    btsq %rsi, (%rdx) # sched: [9:1.00]
+; BDVER2-NEXT:    btcq %rsi, %rdi # sched: [2:0.50]
+; BDVER2-NEXT:    btrq %rsi, %rdi # sched: [2:0.50]
+; BDVER2-NEXT:    btsq %rsi, %rdi # sched: [2:0.50]
+; BDVER2-NEXT:    btq %rsi, (%rdx) # sched: [5:0.50]
+; BDVER2-NEXT:    btcq %rsi, (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    btrq %rsi, (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    btsq %rsi, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btq $7, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btcq $7, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btrq $7, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btsq $7, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    btq $7, (%rdx) # sched: [6:0.50]
+; BDVER2-NEXT:    btcq $7, %rdi # sched: [2:0.50]
+; BDVER2-NEXT:    btrq $7, %rdi # sched: [2:0.50]
+; BDVER2-NEXT:    btsq $7, %rdi # sched: [2:0.50]
+; BDVER2-NEXT:    btq $7, (%rdx) # sched: [5:0.50]
 ; BDVER2-NEXT:    btcq $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btrq $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    btsq $7, (%rdx) # sched: [7:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_bt_btc_btr_bts_64:
 ; BTVER2:       # %bb.0:
@@ -3761,14 +3761,14 @@
 ; BDVER2-LABEL: test_cbw_cdq_cdqe_cqo_cwd_cwde:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cbtw # sched: [1:0.33]
+; BDVER2-NEXT:    cbtw # sched: [1:0.50]
 ; BDVER2-NEXT:    cltd # sched: [1:0.50]
-; BDVER2-NEXT:    cltq # sched: [1:0.33]
+; BDVER2-NEXT:    cltq # sched: [1:0.50]
 ; BDVER2-NEXT:    cqto # sched: [1:0.50]
-; BDVER2-NEXT:    cwtd # sched: [2:1.00]
-; BDVER2-NEXT:    cwtl # sched: [1:0.33]
+; BDVER2-NEXT:    cwtd # sched: [1:0.50]
+; BDVER2-NEXT:    cwtl # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cbw_cdq_cdqe_cqo_cwd_cwde:
 ; BTVER2:       # %bb.0:
@@ -3873,11 +3873,11 @@
 ; BDVER2-LABEL: test_clc_cld_cmc:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    clc # sched: [1:0.25]
-; BDVER2-NEXT:    cld # sched: [1:0.33]
-; BDVER2-NEXT:    cmc # sched: [1:0.33]
+; BDVER2-NEXT:    clc # sched: [1:0.50]
+; BDVER2-NEXT:    cld # sched: [1:0.50]
+; BDVER2-NEXT:    cmc # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_clc_cld_cmc:
 ; BTVER2:       # %bb.0:
@@ -4000,14 +4000,14 @@
 ; BDVER2-LABEL: test_cmp_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    cmpb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    cmpb $7, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpb %dil, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    cmpb %dil, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpb (%rsi), %dil # sched: [6:0.50]
+; BDVER2-NEXT:    cmpb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    cmpb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    cmpb $7, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpb %dil, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    cmpb %dil, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpb (%rsi), %dil # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmp_8:
 ; BTVER2:       # %bb.0:
@@ -4176,18 +4176,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    cmpw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    cmpw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    cmpw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [6:0.50]
-; BDVER2-NEXT:    cmpw $7, %di # sched: [1:0.33]
-; BDVER2-NEXT:    cmpw $7, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpw %di, %di # sched: [1:0.33]
-; BDVER2-NEXT:    cmpw %di, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpw (%rsi), %di # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [5:0.50]
+; BDVER2-NEXT:    cmpw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmpw $7, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpw %di, %di # sched: [1:0.50]
+; BDVER2-NEXT:    cmpw %di, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpw (%rsi), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmp_16:
 ; BTVER2:       # %bb.0:
@@ -4366,18 +4366,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    cmpl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    cmpl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    cmpl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [6:0.50]
-; BDVER2-NEXT:    cmpl $7, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    cmpl $7, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpl %edi, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    cmpl %edi, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpl (%rsi), %edi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [5:0.50]
+; BDVER2-NEXT:    cmpl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmpl $7, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpl %edi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    cmpl %edi, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpl (%rsi), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmp_32:
 ; BTVER2:       # %bb.0:
@@ -4556,18 +4556,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    cmpq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    cmpq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    cmpq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [6:0.50]
-; BDVER2-NEXT:    cmpq $7, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    cmpq $7, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpq %rdi, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    cmpq %rdi, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    cmpq (%rsi), %rdi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [5:0.50]
+; BDVER2-NEXT:    cmpq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmpq $7, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpq %rdi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    cmpq %rdi, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    cmpq (%rsi), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmp_64:
 ; BTVER2:       # %bb.0:
@@ -4690,12 +4690,12 @@
 ; BDVER2-LABEL: test_cmps:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpsb %es:(%rdi), (%rsi) # sched: [8:1.00]
-; BDVER2-NEXT:    cmpsw %es:(%rdi), (%rsi) # sched: [8:1.00]
-; BDVER2-NEXT:    cmpsl %es:(%rdi), (%rsi) # sched: [8:1.00]
-; BDVER2-NEXT:    cmpsq %es:(%rdi), (%rsi) # sched: [8:1.00]
+; BDVER2-NEXT:    cmpsb %es:(%rdi), (%rsi) # sched: [100:0.50]
+; BDVER2-NEXT:    cmpsw %es:(%rdi), (%rsi) # sched: [100:0.50]
+; BDVER2-NEXT:    cmpsl %es:(%rdi), (%rsi) # sched: [100:0.50]
+; BDVER2-NEXT:    cmpsq %es:(%rdi), (%rsi) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmps:
 ; BTVER2:       # %bb.0:
@@ -4788,10 +4788,10 @@
 ; BDVER2-LABEL: test_cmpxchg_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpxchgb %dil, %sil # sched: [5:1.33]
-; BDVER2-NEXT:    cmpxchgb %dil, (%rdx) # sched: [8:2.00]
+; BDVER2-NEXT:    cmpxchgb %dil, %sil # sched: [3:1.00]
+; BDVER2-NEXT:    cmpxchgb %dil, (%rdx) # sched: [3:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmpxchg_8:
 ; BTVER2:       # %bb.0:
@@ -4879,10 +4879,10 @@
 ; BDVER2-LABEL: test_cmpxchg_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpxchgw %di, %si # sched: [5:1.33]
-; BDVER2-NEXT:    cmpxchgw %di, (%rdx) # sched: [8:2.00]
+; BDVER2-NEXT:    cmpxchgw %di, %si # sched: [3:1.00]
+; BDVER2-NEXT:    cmpxchgw %di, (%rdx) # sched: [3:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmpxchg_16:
 ; BTVER2:       # %bb.0:
@@ -4970,10 +4970,10 @@
 ; BDVER2-LABEL: test_cmpxchg_32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpxchgl %edi, %esi # sched: [5:1.33]
-; BDVER2-NEXT:    cmpxchgl %edi, (%rdx) # sched: [8:2.00]
+; BDVER2-NEXT:    cmpxchgl %edi, %esi # sched: [3:1.00]
+; BDVER2-NEXT:    cmpxchgl %edi, (%rdx) # sched: [3:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmpxchg_32:
 ; BTVER2:       # %bb.0:
@@ -5061,10 +5061,10 @@
 ; BDVER2-LABEL: test_cmpxchg_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpxchgq %rdi, %rsi # sched: [5:1.33]
-; BDVER2-NEXT:    cmpxchgq %rdi, (%rdx) # sched: [8:2.00]
+; BDVER2-NEXT:    cmpxchgq %rdi, %rsi # sched: [3:1.00]
+; BDVER2-NEXT:    cmpxchgq %rdi, (%rdx) # sched: [3:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmpxchg_64:
 ; BTVER2:       # %bb.0:
@@ -5152,10 +5152,10 @@
 ; BDVER2-LABEL: test_cmpxchg8b_cmpxchg16b:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cmpxchg8b (%rdi) # sched: [6:1.00]
-; BDVER2-NEXT:    cmpxchg16b (%rdi) # sched: [6:1.00]
+; BDVER2-NEXT:    cmpxchg8b (%rdi) # sched: [3:1.00]
+; BDVER2-NEXT:    cmpxchg16b (%rdi) # sched: [3:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cmpxchg8b_cmpxchg16b:
 ; BTVER2:       # %bb.0:
@@ -5236,9 +5236,9 @@
 ; BDVER2-LABEL: test_cpuid:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    cpuid # sched: [100:0.33]
+; BDVER2-NEXT:    cpuid # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_cpuid:
 ; BTVER2:       # %bb.0:
@@ -5325,10 +5325,10 @@
 ; BDVER2-LABEL: test_dec8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    decb %dil # sched: [1:0.33]
-; BDVER2-NEXT:    decb (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    decb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    decb (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dec8:
 ; BTVER2:       # %bb.0:
@@ -5416,10 +5416,10 @@
 ; BDVER2-LABEL: test_dec16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    decw %di # sched: [1:0.33]
-; BDVER2-NEXT:    decw (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    decw %di # sched: [1:0.50]
+; BDVER2-NEXT:    decw (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dec16:
 ; BTVER2:       # %bb.0:
@@ -5507,10 +5507,10 @@
 ; BDVER2-LABEL: test_dec32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    decl %edi # sched: [1:0.33]
-; BDVER2-NEXT:    decl (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    decl %edi # sched: [1:0.50]
+; BDVER2-NEXT:    decl (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dec32:
 ; BTVER2:       # %bb.0:
@@ -5598,10 +5598,10 @@
 ; BDVER2-LABEL: test_dec64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    decq %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    decq (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    decq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    decq (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_dec64:
 ; BTVER2:       # %bb.0:
@@ -5756,16 +5756,16 @@
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # sched: [5:0.50]
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    divb %dil # sched: [25:10.00]
-; BDVER2-NEXT:    divb (%r8) # sched: [30:10.00]
-; BDVER2-NEXT:    divw %si # sched: [25:10.00]
-; BDVER2-NEXT:    divw (%r9) # sched: [30:10.00]
-; BDVER2-NEXT:    divl %edx # sched: [25:10.00]
-; BDVER2-NEXT:    divl (%rax) # sched: [30:10.00]
-; BDVER2-NEXT:    divq %rcx # sched: [25:10.00]
-; BDVER2-NEXT:    divq (%r10) # sched: [30:10.00]
+; BDVER2-NEXT:    divb %dil # sched: [12:12.00]
+; BDVER2-NEXT:    divb (%r8) # sched: [16:12.00]
+; BDVER2-NEXT:    divw %si # sched: [15:15.00]
+; BDVER2-NEXT:    divw (%r9) # sched: [19:15.00]
+; BDVER2-NEXT:    divl %edx # sched: [14:14.00]
+; BDVER2-NEXT:    divl (%rax) # sched: [18:14.00]
+; BDVER2-NEXT:    divq %rcx # sched: [14:14.00]
+; BDVER2-NEXT:    divq (%r10) # sched: [18:14.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_div:
 ; BTVER2:       # %bb.0:
@@ -5871,9 +5871,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    enter $7, $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [100:0.33]
+; BDVER2-NEXT:    # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_enter:
 ; BTVER2:       # %bb.0:
@@ -6028,16 +6028,16 @@
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # sched: [5:0.50]
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    idivb %dil # sched: [25:10.00]
-; BDVER2-NEXT:    idivb (%r8) # sched: [30:10.00]
-; BDVER2-NEXT:    idivw %si # sched: [25:10.00]
-; BDVER2-NEXT:    idivw (%r9) # sched: [30:10.00]
-; BDVER2-NEXT:    idivl %edx # sched: [25:10.00]
-; BDVER2-NEXT:    idivl (%rax) # sched: [30:10.00]
-; BDVER2-NEXT:    idivq %rcx # sched: [25:10.00]
-; BDVER2-NEXT:    idivq (%r10) # sched: [30:10.00]
+; BDVER2-NEXT:    idivb %dil # sched: [12:12.00]
+; BDVER2-NEXT:    idivb (%r8) # sched: [16:12.00]
+; BDVER2-NEXT:    idivw %si # sched: [15:17.00]
+; BDVER2-NEXT:    idivw (%r9) # sched: [19:17.00]
+; BDVER2-NEXT:    idivl %edx # sched: [14:25.00]
+; BDVER2-NEXT:    idivl (%rax) # sched: [18:25.00]
+; BDVER2-NEXT:    idivq %rcx # sched: [14:14.00]
+; BDVER2-NEXT:    idivq (%r10) # sched: [18:14.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_idiv:
 ; BTVER2:       # %bb.0:
@@ -6142,10 +6142,10 @@
 ; BDVER2-LABEL: test_imul_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    imulb %dil # sched: [3:1.00]
+; BDVER2-NEXT:    imulb %dil # sched: [4:1.00]
 ; BDVER2-NEXT:    imulb (%rsi) # sched: [8:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_imul_8:
 ; BTVER2:       # %bb.0:
@@ -6297,18 +6297,18 @@
 ; BDVER2-LABEL: test_imul_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    imulw %di # sched: [4:1.33]
-; BDVER2-NEXT:    imulw (%rsi) # sched: [9:1.33]
-; BDVER2-NEXT:    imulw %dx, %di # sched: [3:1.00]
+; BDVER2-NEXT:    imulw %di # sched: [4:1.00]
+; BDVER2-NEXT:    imulw (%rsi) # sched: [8:1.00]
+; BDVER2-NEXT:    imulw %dx, %di # sched: [4:1.00]
 ; BDVER2-NEXT:    imulw (%rsi), %di # sched: [8:1.00]
 ; BDVER2-NEXT:    imulw $511, %di, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [4:1.00]
+; BDVER2-NEXT:    # sched: [5:1.00]
 ; BDVER2-NEXT:    imulw $511, (%rsi), %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [8:1.00]
-; BDVER2-NEXT:    imulw $7, %di, %di # sched: [4:1.00]
-; BDVER2-NEXT:    imulw $7, (%rsi), %di # sched: [8:1.00]
+; BDVER2-NEXT:    # sched: [9:1.00]
+; BDVER2-NEXT:    imulw $7, %di, %di # sched: [5:1.00]
+; BDVER2-NEXT:    imulw $7, (%rsi), %di # sched: [9:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_imul_16:
 ; BTVER2:       # %bb.0:
@@ -6477,17 +6477,17 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    imull %edi # sched: [4:1.00]
-; BDVER2-NEXT:    imull (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    imull %edx, %edi # sched: [3:1.00]
+; BDVER2-NEXT:    imull (%rsi) # sched: [8:1.00]
+; BDVER2-NEXT:    imull %edx, %edi # sched: [4:1.00]
 ; BDVER2-NEXT:    imull (%rsi), %edi # sched: [8:1.00]
 ; BDVER2-NEXT:    imull $665536, %edi, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [3:1.00]
+; BDVER2-NEXT:    # sched: [4:1.00]
 ; BDVER2-NEXT:    imull $665536, (%rsi), %edi # imm = 0xA27C0
 ; BDVER2-NEXT:    # sched: [8:1.00]
-; BDVER2-NEXT:    imull $7, %edi, %edi # sched: [3:1.00]
+; BDVER2-NEXT:    imull $7, %edi, %edi # sched: [4:1.00]
 ; BDVER2-NEXT:    imull $7, (%rsi), %edi # sched: [8:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_imul_32:
 ; BTVER2:       # %bb.0:
@@ -6655,18 +6655,18 @@
 ; BDVER2-LABEL: test_imul_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    imulq %rdi # sched: [4:1.00]
-; BDVER2-NEXT:    imulq (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    imulq %rdx, %rdi # sched: [3:1.00]
-; BDVER2-NEXT:    imulq (%rsi), %rdi # sched: [8:1.00]
+; BDVER2-NEXT:    imulq %rdi # sched: [6:4.00]
+; BDVER2-NEXT:    imulq (%rsi) # sched: [10:4.00]
+; BDVER2-NEXT:    imulq %rdx, %rdi # sched: [6:4.00]
+; BDVER2-NEXT:    imulq (%rsi), %rdi # sched: [10:4.00]
 ; BDVER2-NEXT:    imulq $665536, %rdi, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [3:1.00]
+; BDVER2-NEXT:    # sched: [6:4.00]
 ; BDVER2-NEXT:    imulq $665536, (%rsi), %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [8:1.00]
-; BDVER2-NEXT:    imulq $7, %rdi, %rdi # sched: [3:1.00]
-; BDVER2-NEXT:    imulq $7, (%rsi), %rdi # sched: [8:1.00]
+; BDVER2-NEXT:    # sched: [10:4.00]
+; BDVER2-NEXT:    imulq $7, %rdi, %rdi # sched: [6:4.00]
+; BDVER2-NEXT:    imulq $7, (%rsi), %rdi # sched: [10:4.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_imul_64:
 ; BTVER2:       # %bb.0:
@@ -6803,14 +6803,14 @@
 ; BDVER2-LABEL: test_in:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    inb $7, %al # sched: [100:0.33]
-; BDVER2-NEXT:    inw $7, %ax # sched: [100:0.33]
-; BDVER2-NEXT:    inl $7, %eax # sched: [100:0.33]
-; BDVER2-NEXT:    inb %dx, %al # sched: [100:0.33]
-; BDVER2-NEXT:    inw %dx, %ax # sched: [100:0.33]
-; BDVER2-NEXT:    inl %dx, %eax # sched: [100:0.33]
+; BDVER2-NEXT:    inb $7, %al # sched: [100:0.50]
+; BDVER2-NEXT:    inw $7, %ax # sched: [100:0.50]
+; BDVER2-NEXT:    inl $7, %eax # sched: [100:0.50]
+; BDVER2-NEXT:    inb %dx, %al # sched: [100:0.50]
+; BDVER2-NEXT:    inw %dx, %ax # sched: [100:0.50]
+; BDVER2-NEXT:    inl %dx, %eax # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_in:
 ; BTVER2:       # %bb.0:
@@ -6907,10 +6907,10 @@
 ; BDVER2-LABEL: test_inc8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    incb %dil # sched: [1:0.33]
-; BDVER2-NEXT:    incb (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    incb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    incb (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_inc8:
 ; BTVER2:       # %bb.0:
@@ -6998,10 +6998,10 @@
 ; BDVER2-LABEL: test_inc16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    incw %di # sched: [1:0.33]
-; BDVER2-NEXT:    incw (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    incw %di # sched: [1:0.50]
+; BDVER2-NEXT:    incw (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_inc16:
 ; BTVER2:       # %bb.0:
@@ -7089,10 +7089,10 @@
 ; BDVER2-LABEL: test_inc32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    incl %edi # sched: [1:0.33]
-; BDVER2-NEXT:    incl (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    incl %edi # sched: [1:0.50]
+; BDVER2-NEXT:    incl (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_inc32:
 ; BTVER2:       # %bb.0:
@@ -7180,10 +7180,10 @@
 ; BDVER2-LABEL: test_inc64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    incq %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    incq (%rsi) # sched: [7:1.00]
+; BDVER2-NEXT:    incq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    incq (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_inc64:
 ; BTVER2:       # %bb.0:
@@ -7280,11 +7280,11 @@
 ; BDVER2-LABEL: test_ins:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    insb %dx, %es:(%rdi) # sched: [100:0.33]
-; BDVER2-NEXT:    insw %dx, %es:(%rdi) # sched: [100:0.33]
-; BDVER2-NEXT:    insl %dx, %es:(%rdi) # sched: [100:0.33]
+; BDVER2-NEXT:    insb %dx, %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    insw %dx, %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    insl %dx, %es:(%rdi) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ins:
 ; BTVER2:       # %bb.0:
@@ -7367,9 +7367,9 @@
 ; BDVER2-LABEL: test_int:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    int $7 # sched: [100:0.33]
+; BDVER2-NEXT:    int $7 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_int:
 ; BTVER2:       # %bb.0:
@@ -7456,10 +7456,10 @@
 ; BDVER2-LABEL: test_invlpg_invlpga:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    invlpg (%rdi) # sched: [100:0.33]
-; BDVER2-NEXT:    invlpga %rax, %ecx # sched: [100:0.33]
+; BDVER2-NEXT:    invlpg (%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    invlpga %rax, %ecx # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_invlpg_invlpga:
 ; BTVER2:       # %bb.0:
@@ -7812,7 +7812,7 @@
 ; BDVER2-NEXT:    jg JCCTGT # sched: [1:1.00]
 ; BDVER2-NEXT:    jg JCCTGT # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_jcc:
 ; BTVER2:       # %bb.0:
@@ -7968,10 +7968,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:  JXTGT:
-; BDVER2-NEXT:    jecxz JXTGT # sched: [2:1.00]
-; BDVER2-NEXT:    jrcxz JXTGT # sched: [2:1.00]
+; BDVER2-NEXT:    jecxz JXTGT # sched: [1:1.00]
+; BDVER2-NEXT:    jrcxz JXTGT # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_jecxz_jrcxz:
 ; BTVER2:       # %bb.0:
@@ -8064,10 +8064,10 @@
 ; BDVER2-LABEL: test_lahf_sahf:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    lahf # sched: [1:0.50]
-; BDVER2-NEXT:    sahf # sched: [1:0.50]
+; BDVER2-NEXT:    lahf # sched: [2:0.50]
+; BDVER2-NEXT:    sahf # sched: [2:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lahf_sahf:
 ; BTVER2:       # %bb.0:
@@ -8156,9 +8156,9 @@
 ; BDVER2-LABEL: test_leave:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    leave # sched: [7:0.67]
+; BDVER2-NEXT:    leave # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_leave:
 ; BTVER2:       # %bb.0:
@@ -8261,12 +8261,12 @@
 ; BDVER2-LABEL: test_lods:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    lodsb (%rsi), %al # sched: [7:0.67]
-; BDVER2-NEXT:    lodsw (%rsi), %ax # sched: [7:0.67]
-; BDVER2-NEXT:    lodsl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    lodsq (%rsi), %rax # sched: [6:0.50]
+; BDVER2-NEXT:    lodsb (%rsi), %al # sched: [100:0.50]
+; BDVER2-NEXT:    lodsw (%rsi), %ax # sched: [100:0.50]
+; BDVER2-NEXT:    lodsl (%rsi), %eax # sched: [100:0.50]
+; BDVER2-NEXT:    lodsq (%rsi), %rax # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_lods:
 ; BTVER2:       # %bb.0:
@@ -8380,7 +8380,7 @@
 ; BDVER2-NEXT:    loope LTGT # sched: [1:1.00]
 ; BDVER2-NEXT:    loopne LTGT # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_loop:
 ; BTVER2:       # %bb.0:
@@ -8475,10 +8475,10 @@
 ; BDVER2-LABEL: test_movnti:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    movntil %edi, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    movntiq %rdx, (%rcx) # sched: [1:1.00]
+; BDVER2-NEXT:    movntil %edi, (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    movntiq %rdx, (%rcx) # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movnti:
 ; BTVER2:       # %bb.0:
@@ -8583,12 +8583,12 @@
 ; BDVER2-LABEL: test_movs:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    movsb (%rsi), %es:(%rdi) # sched: [8:1.00]
-; BDVER2-NEXT:    movsw (%rsi), %es:(%rdi) # sched: [8:1.00]
-; BDVER2-NEXT:    movsl (%rsi), %es:(%rdi) # sched: [8:1.00]
-; BDVER2-NEXT:    movsq (%rsi), %es:(%rdi) # sched: [8:1.00]
+; BDVER2-NEXT:    movsb (%rsi), %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    movsw (%rsi), %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    movsl (%rsi), %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    movsq (%rsi), %es:(%rdi) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movs:
 ; BTVER2:       # %bb.0:
@@ -8692,11 +8692,11 @@
 ; BDVER2-LABEL: test_movslq:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    movslq %edi, %rax # sched: [1:0.33]
+; BDVER2-NEXT:    movslq %edi, %rax # sched: [1:0.50]
 ; BDVER2-NEXT:    movslq (%rsi), %rcx # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    orq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movslq:
 ; BTVER2:       # %bb.0:
@@ -8856,16 +8856,16 @@
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # sched: [5:0.50]
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    mulb %dil # sched: [3:1.00]
+; BDVER2-NEXT:    mulb %dil # sched: [4:1.00]
 ; BDVER2-NEXT:    mulb (%r8) # sched: [8:1.00]
-; BDVER2-NEXT:    mulw %si # sched: [4:1.33]
-; BDVER2-NEXT:    mulw (%r9) # sched: [9:1.33]
+; BDVER2-NEXT:    mulw %si # sched: [4:1.00]
+; BDVER2-NEXT:    mulw (%r9) # sched: [8:1.00]
 ; BDVER2-NEXT:    mull %edx # sched: [4:1.00]
-; BDVER2-NEXT:    mull (%rax) # sched: [9:1.00]
-; BDVER2-NEXT:    mulq %rcx # sched: [4:1.00]
-; BDVER2-NEXT:    mulq (%r10) # sched: [9:1.00]
+; BDVER2-NEXT:    mull (%rax) # sched: [8:1.00]
+; BDVER2-NEXT:    mulq %rcx # sched: [6:4.00]
+; BDVER2-NEXT:    mulq (%r10) # sched: [10:4.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_mul:
 ; BTVER2:       # %bb.0:
@@ -9036,16 +9036,16 @@
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # sched: [5:0.50]
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    negb %dil # sched: [1:0.33]
-; BDVER2-NEXT:    negb (%r8) # sched: [7:1.00]
-; BDVER2-NEXT:    negw %si # sched: [1:0.33]
-; BDVER2-NEXT:    negw (%r9) # sched: [7:1.00]
-; BDVER2-NEXT:    negl %edx # sched: [1:0.33]
-; BDVER2-NEXT:    negl (%rax) # sched: [7:1.00]
-; BDVER2-NEXT:    negq %rcx # sched: [1:0.33]
-; BDVER2-NEXT:    negq (%r10) # sched: [7:1.00]
+; BDVER2-NEXT:    negb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    negb (%r8) # sched: [6:1.00]
+; BDVER2-NEXT:    negw %si # sched: [1:0.50]
+; BDVER2-NEXT:    negw (%r9) # sched: [6:1.00]
+; BDVER2-NEXT:    negl %edx # sched: [1:0.50]
+; BDVER2-NEXT:    negl (%rax) # sched: [6:1.00]
+; BDVER2-NEXT:    negq %rcx # sched: [1:0.50]
+; BDVER2-NEXT:    negq (%r10) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_neg:
 ; BTVER2:       # %bb.0:
@@ -9190,15 +9190,15 @@
 ; BDVER2-LABEL: test_nop:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    nop # sched: [1:0.25]
-; BDVER2-NEXT:    nopw %di # sched: [1:0.25]
-; BDVER2-NEXT:    nopw (%rcx) # sched: [1:0.25]
-; BDVER2-NEXT:    nopl %esi # sched: [1:0.25]
-; BDVER2-NEXT:    nopl (%r8) # sched: [1:0.25]
-; BDVER2-NEXT:    nopq %rdx # sched: [1:0.25]
-; BDVER2-NEXT:    nopq (%r9) # sched: [1:0.25]
+; BDVER2-NEXT:    nop # sched: [1:0.50]
+; BDVER2-NEXT:    nopw %di # sched: [1:0.50]
+; BDVER2-NEXT:    nopw (%rcx) # sched: [1:0.50]
+; BDVER2-NEXT:    nopl %esi # sched: [1:0.50]
+; BDVER2-NEXT:    nopl (%r8) # sched: [1:0.50]
+; BDVER2-NEXT:    nopq %rdx # sched: [1:0.50]
+; BDVER2-NEXT:    nopq (%r9) # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_nop:
 ; BTVER2:       # %bb.0:
@@ -9363,16 +9363,16 @@
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # sched: [5:0.50]
 ; BDVER2-NEXT:    movq {{[0-9]+}}(%rsp), %rax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    notb %dil # sched: [1:0.33]
-; BDVER2-NEXT:    notb (%r8) # sched: [7:1.00]
-; BDVER2-NEXT:    notw %si # sched: [1:0.33]
-; BDVER2-NEXT:    notw (%r9) # sched: [7:1.00]
-; BDVER2-NEXT:    notl %edx # sched: [1:0.33]
-; BDVER2-NEXT:    notl (%rax) # sched: [7:1.00]
-; BDVER2-NEXT:    notq %rcx # sched: [1:0.33]
-; BDVER2-NEXT:    notq (%r10) # sched: [7:1.00]
+; BDVER2-NEXT:    notb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    notb (%r8) # sched: [6:1.00]
+; BDVER2-NEXT:    notw %si # sched: [1:0.50]
+; BDVER2-NEXT:    notw (%r9) # sched: [6:1.00]
+; BDVER2-NEXT:    notl %edx # sched: [1:0.50]
+; BDVER2-NEXT:    notl (%rax) # sched: [6:1.00]
+; BDVER2-NEXT:    notq %rcx # sched: [1:0.50]
+; BDVER2-NEXT:    notq (%r10) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_not:
 ; BTVER2:       # %bb.0:
@@ -9509,14 +9509,14 @@
 ; BDVER2-LABEL: test_or_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    orb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    orb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    orb $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orb %dl, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    orb %dil, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orb (%rsi), %dil # sched: [6:0.50]
+; BDVER2-NEXT:    orb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    orb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    orb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orb %dl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    orb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orb (%rsi), %dil # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_or_8:
 ; BTVER2:       # %bb.0:
@@ -9685,18 +9685,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    orw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    orw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    orw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    orw $7, %di # sched: [1:0.33]
-; BDVER2-NEXT:    orw $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orw %dx, %di # sched: [1:0.33]
-; BDVER2-NEXT:    orw %di, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orw (%rsi), %di # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    orw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    orw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orw %dx, %di # sched: [1:0.50]
+; BDVER2-NEXT:    orw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orw (%rsi), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_or_16:
 ; BTVER2:       # %bb.0:
@@ -9875,18 +9875,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    orl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    orl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    orl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    orl $7, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    orl $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orl %edx, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    orl %edi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orl (%rsi), %edi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    orl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    orl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orl %edx, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    orl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orl (%rsi), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_or_32:
 ; BTVER2:       # %bb.0:
@@ -10065,18 +10065,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    orq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    orq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    orq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    orq $7, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    orq $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orq %rdx, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    orq %rdi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    orq (%rsi), %rdi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    orq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    orq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orq %rdx, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    orq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    orq (%rsi), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_or_64:
 ; BTVER2:       # %bb.0:
@@ -10215,14 +10215,14 @@
 ; BDVER2-LABEL: test_out:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    outb %al, $7 # sched: [100:0.33]
-; BDVER2-NEXT:    outw %ax, $7 # sched: [100:0.33]
-; BDVER2-NEXT:    outl %eax, $7 # sched: [100:0.33]
-; BDVER2-NEXT:    outb %al, %dx # sched: [100:0.33]
-; BDVER2-NEXT:    outw %ax, %dx # sched: [100:0.33]
-; BDVER2-NEXT:    outl %eax, %dx # sched: [100:0.33]
+; BDVER2-NEXT:    outb %al, $7 # sched: [100:0.50]
+; BDVER2-NEXT:    outw %ax, $7 # sched: [100:0.50]
+; BDVER2-NEXT:    outl %eax, $7 # sched: [100:0.50]
+; BDVER2-NEXT:    outb %al, %dx # sched: [100:0.50]
+; BDVER2-NEXT:    outw %ax, %dx # sched: [100:0.50]
+; BDVER2-NEXT:    outl %eax, %dx # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_out:
 ; BTVER2:       # %bb.0:
@@ -10327,11 +10327,11 @@
 ; BDVER2-LABEL: test_outs:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    outsb (%rsi), %dx # sched: [100:0.33]
-; BDVER2-NEXT:    outsw (%rsi), %dx # sched: [100:0.33]
-; BDVER2-NEXT:    outsl (%rsi), %dx # sched: [100:0.33]
+; BDVER2-NEXT:    outsb (%rsi), %dx # sched: [100:0.50]
+; BDVER2-NEXT:    outsw (%rsi), %dx # sched: [100:0.50]
+; BDVER2-NEXT:    outsl (%rsi), %dx # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_outs:
 ; BTVER2:       # %bb.0:
@@ -10414,9 +10414,9 @@
 ; BDVER2-LABEL: test_pause:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    pause # sched: [4:1.33]
+; BDVER2-NEXT:    pause # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pause:
 ; BTVER2:       # %bb.0:
@@ -10519,12 +10519,12 @@
 ; BDVER2-LABEL: test_pop_push:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    popq %fs # sched: [100:0.33]
-; BDVER2-NEXT:    popq %gs # sched: [100:0.33]
-; BDVER2-NEXT:    pushq %fs # sched: [3:1.00]
-; BDVER2-NEXT:    pushq %gs # sched: [5:1.00]
+; BDVER2-NEXT:    popq %fs # sched: [100:0.50]
+; BDVER2-NEXT:    popq %gs # sched: [100:0.50]
+; BDVER2-NEXT:    pushq %fs # sched: [100:0.50]
+; BDVER2-NEXT:    pushq %gs # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pop_push:
 ; BTVER2:       # %bb.0:
@@ -10656,15 +10656,15 @@
 ; BDVER2-LABEL: test_pop_push_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    popw %ax # sched: [6:0.50]
-; BDVER2-NEXT:    popw (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    pushw %di # sched: [5:1.00]
-; BDVER2-NEXT:    pushw (%rsi) # sched: [5:1.00]
+; BDVER2-NEXT:    popw %ax # sched: [5:0.50]
+; BDVER2-NEXT:    popw (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    pushw %di # sched: [1:0.50]
+; BDVER2-NEXT:    pushw (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    pushw $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [1:1.00]
-; BDVER2-NEXT:    pushw $7 # sched: [1:1.00]
+; BDVER2-NEXT:    # sched: [1:0.50]
+; BDVER2-NEXT:    pushw $7 # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pop_push_16:
 ; BTVER2:       # %bb.0:
@@ -10802,15 +10802,15 @@
 ; BDVER2-LABEL: test_pop_push_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    popq %rax # sched: [6:0.50]
-; BDVER2-NEXT:    popq (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    pushq %rdi # sched: [5:1.00]
-; BDVER2-NEXT:    pushq (%rsi) # sched: [5:1.00]
+; BDVER2-NEXT:    popq %rax # sched: [5:0.50]
+; BDVER2-NEXT:    popq (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    pushq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    pushq (%rsi) # sched: [6:1.00]
 ; BDVER2-NEXT:    pushq $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [1:1.00]
-; BDVER2-NEXT:    pushq $7 # sched: [5:1.00]
+; BDVER2-NEXT:    # sched: [1:0.50]
+; BDVER2-NEXT:    pushq $7 # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_pop_push_64:
 ; BTVER2:       # %bb.0:
@@ -10910,9 +10910,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    popfq # sched: [5:0.50]
-; BDVER2-NEXT:    pushfq # sched: [5:1.00]
+; BDVER2-NEXT:    pushfq # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_popf_pushf:
 ; BTVER2:       # %bb.0:
@@ -11081,20 +11081,20 @@
 ; BDVER2-LABEL: test_rcl_rcr_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rclb %dil # sched: [2:1.50]
-; BDVER2-NEXT:    rcrb %dil # sched: [2:1.50]
-; BDVER2-NEXT:    rclb (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrb (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rclb $7, %dil # sched: [5:4.00]
-; BDVER2-NEXT:    rcrb $7, %dil # sched: [5:4.00]
-; BDVER2-NEXT:    rclb $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrb $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rclb %cl, %dil # sched: [5:4.00]
-; BDVER2-NEXT:    rcrb %cl, %dil # sched: [5:4.00]
-; BDVER2-NEXT:    rclb %cl, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrb %cl, (%rdx) # sched: [11:3.50]
+; BDVER2-NEXT:    rclb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rcrb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rclb (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrb (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rclb $7, %dil # sched: [13:0.50]
+; BDVER2-NEXT:    rcrb $7, %dil # sched: [12:0.50]
+; BDVER2-NEXT:    rclb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rclb %cl, %dil # sched: [12:0.50]
+; BDVER2-NEXT:    rcrb %cl, %dil # sched: [11:0.50]
+; BDVER2-NEXT:    rclb %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrb %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rcl_rcr_8:
 ; BTVER2:       # %bb.0:
@@ -11282,20 +11282,20 @@
 ; BDVER2-LABEL: test_rcl_rcr_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rclw %di # sched: [2:1.50]
-; BDVER2-NEXT:    rcrw %di # sched: [2:1.50]
-; BDVER2-NEXT:    rclw (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrw (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rclw $7, %di # sched: [5:4.00]
-; BDVER2-NEXT:    rcrw $7, %di # sched: [5:4.00]
-; BDVER2-NEXT:    rclw $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrw $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rclw %cl, %di # sched: [5:4.00]
-; BDVER2-NEXT:    rcrw %cl, %di # sched: [5:4.00]
-; BDVER2-NEXT:    rclw %cl, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrw %cl, (%rdx) # sched: [11:3.50]
+; BDVER2-NEXT:    rclw %di # sched: [1:0.50]
+; BDVER2-NEXT:    rcrw %di # sched: [1:0.50]
+; BDVER2-NEXT:    rclw (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrw (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rclw $7, %di # sched: [11:0.50]
+; BDVER2-NEXT:    rcrw $7, %di # sched: [10:0.50]
+; BDVER2-NEXT:    rclw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rclw %cl, %di # sched: [10:0.50]
+; BDVER2-NEXT:    rcrw %cl, %di # sched: [9:0.50]
+; BDVER2-NEXT:    rclw %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrw %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rcl_rcr_16:
 ; BTVER2:       # %bb.0:
@@ -11483,20 +11483,20 @@
 ; BDVER2-LABEL: test_rcl_rcr_32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rcll %edi # sched: [2:1.50]
-; BDVER2-NEXT:    rcrl %edi # sched: [2:1.50]
-; BDVER2-NEXT:    rcll (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrl (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcll $7, %edi # sched: [5:4.00]
-; BDVER2-NEXT:    rcrl $7, %edi # sched: [5:4.00]
-; BDVER2-NEXT:    rcll $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrl $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcll %cl, %edi # sched: [5:4.00]
-; BDVER2-NEXT:    rcrl %cl, %edi # sched: [5:4.00]
-; BDVER2-NEXT:    rcll %cl, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrl %cl, (%rdx) # sched: [11:3.50]
+; BDVER2-NEXT:    rcll %edi # sched: [1:0.50]
+; BDVER2-NEXT:    rcrl %edi # sched: [1:0.50]
+; BDVER2-NEXT:    rcll (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrl (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcll $7, %edi # sched: [8:0.50]
+; BDVER2-NEXT:    rcrl $7, %edi # sched: [7:0.50]
+; BDVER2-NEXT:    rcll $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrl $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcll %cl, %edi # sched: [7:0.50]
+; BDVER2-NEXT:    rcrl %cl, %edi # sched: [7:0.50]
+; BDVER2-NEXT:    rcll %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrl %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rcl_rcr_32:
 ; BTVER2:       # %bb.0:
@@ -11684,20 +11684,20 @@
 ; BDVER2-LABEL: test_rcl_rcr_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rclq %rdi # sched: [2:1.50]
-; BDVER2-NEXT:    rcrq %rdi # sched: [2:1.50]
-; BDVER2-NEXT:    rclq (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrq (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rclq $7, %rdi # sched: [5:4.00]
-; BDVER2-NEXT:    rcrq $7, %rdi # sched: [5:4.00]
-; BDVER2-NEXT:    rclq $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrq $7, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rclq %cl, %rdi # sched: [5:4.00]
-; BDVER2-NEXT:    rcrq %cl, %rdi # sched: [5:4.00]
-; BDVER2-NEXT:    rclq %cl, (%rdx) # sched: [11:3.50]
-; BDVER2-NEXT:    rcrq %cl, (%rdx) # sched: [11:3.50]
+; BDVER2-NEXT:    rclq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rcrq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rclq (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrq (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rclq $7, %rdi # sched: [8:0.50]
+; BDVER2-NEXT:    rcrq $7, %rdi # sched: [7:0.50]
+; BDVER2-NEXT:    rclq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rclq %cl, %rdi # sched: [7:0.50]
+; BDVER2-NEXT:    rcrq %cl, %rdi # sched: [7:0.50]
+; BDVER2-NEXT:    rclq %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rcrq %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rcl_rcr_64:
 ; BTVER2:       # %bb.0:
@@ -11806,10 +11806,10 @@
 ; BDVER2-LABEL: test_rdmsr_wrmsr:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rdmsr # sched: [100:0.33]
-; BDVER2-NEXT:    wrmsr # sched: [100:0.33]
+; BDVER2-NEXT:    rdmsr # sched: [100:0.50]
+; BDVER2-NEXT:    wrmsr # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rdmsr_wrmsr:
 ; BTVER2:       # %bb.0:
@@ -11890,9 +11890,9 @@
 ; BDVER2-LABEL: test_rdpmc:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rdpmc # sched: [100:0.33]
+; BDVER2-NEXT:    rdpmc # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rdpmc:
 ; BTVER2:       # %bb.0:
@@ -11979,10 +11979,10 @@
 ; BDVER2-LABEL: test_rdtsc_rdtscp:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rdtsc # sched: [100:0.33]
-; BDVER2-NEXT:    rdtscp # sched: [100:0.33]
+; BDVER2-NEXT:    rdtsc # sched: [100:0.50]
+; BDVER2-NEXT:    rdtscp # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rdtsc_rdtscp:
 ; BTVER2:       # %bb.0:
@@ -12103,14 +12103,14 @@
 ; BDVER2-LABEL: test_ret:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ; BDVER2-NEXT:    retq $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [6:1.00]
-; BDVER2-NEXT:    lretl # sched: [6:1.00]
+; BDVER2-NEXT:    # sched: [5:1.00]
+; BDVER2-NEXT:    lretl # sched: [5:1.00]
 ; BDVER2-NEXT:    lretl $4095 # imm = 0xFFF
-; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ret:
 ; BTVER2:       # %bb.0:
@@ -12287,20 +12287,20 @@
 ; BDVER2-LABEL: test_rol_ror_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rolb %dil # sched: [2:1.00]
-; BDVER2-NEXT:    rorb %dil # sched: [2:1.00]
-; BDVER2-NEXT:    rolb (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorb (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rolb $7, %dil # sched: [2:1.00]
-; BDVER2-NEXT:    rorb $7, %dil # sched: [2:1.00]
-; BDVER2-NEXT:    rolb $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorb $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rolb %cl, %dil # sched: [3:1.50]
-; BDVER2-NEXT:    rorb %cl, %dil # sched: [3:1.50]
-; BDVER2-NEXT:    rolb %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    rorb %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    rolb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rorb %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rolb (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorb (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rolb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rorb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rolb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rolb %cl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rorb %cl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    rolb %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorb %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rol_ror_8:
 ; BTVER2:       # %bb.0:
@@ -12488,20 +12488,20 @@
 ; BDVER2-LABEL: test_rol_ror_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rolw %di # sched: [2:1.00]
-; BDVER2-NEXT:    rorw %di # sched: [2:1.00]
-; BDVER2-NEXT:    rolw (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorw (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rolw $7, %di # sched: [2:1.00]
-; BDVER2-NEXT:    rorw $7, %di # sched: [2:1.00]
-; BDVER2-NEXT:    rolw $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorw $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rolw %cl, %di # sched: [3:1.50]
-; BDVER2-NEXT:    rorw %cl, %di # sched: [3:1.50]
-; BDVER2-NEXT:    rolw %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    rorw %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    rolw %di # sched: [1:0.50]
+; BDVER2-NEXT:    rorw %di # sched: [1:0.50]
+; BDVER2-NEXT:    rolw (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorw (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rolw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    rorw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    rolw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rolw %cl, %di # sched: [1:0.50]
+; BDVER2-NEXT:    rorw %cl, %di # sched: [1:0.50]
+; BDVER2-NEXT:    rolw %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorw %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rol_ror_16:
 ; BTVER2:       # %bb.0:
@@ -12689,20 +12689,20 @@
 ; BDVER2-LABEL: test_rol_ror_32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    roll %edi # sched: [2:1.00]
-; BDVER2-NEXT:    rorl %edi # sched: [2:1.00]
-; BDVER2-NEXT:    roll (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorl (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    roll $7, %edi # sched: [2:1.00]
-; BDVER2-NEXT:    rorl $7, %edi # sched: [2:1.00]
-; BDVER2-NEXT:    roll $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorl $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    roll %cl, %edi # sched: [3:1.50]
-; BDVER2-NEXT:    rorl %cl, %edi # sched: [3:1.50]
-; BDVER2-NEXT:    roll %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    rorl %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    roll %edi # sched: [1:0.50]
+; BDVER2-NEXT:    rorl %edi # sched: [1:0.50]
+; BDVER2-NEXT:    roll (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorl (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    roll $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    rorl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    roll $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorl $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    roll %cl, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    rorl %cl, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    roll %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorl %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rol_ror_32:
 ; BTVER2:       # %bb.0:
@@ -12890,20 +12890,20 @@
 ; BDVER2-LABEL: test_rol_ror_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    rolq %rdi # sched: [2:1.00]
-; BDVER2-NEXT:    rorq %rdi # sched: [2:1.00]
-; BDVER2-NEXT:    rolq (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorq (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rolq $7, %rdi # sched: [2:1.00]
-; BDVER2-NEXT:    rorq $7, %rdi # sched: [2:1.00]
-; BDVER2-NEXT:    rolq $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rorq $7, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    rolq %cl, %rdi # sched: [3:1.50]
-; BDVER2-NEXT:    rorq %cl, %rdi # sched: [3:1.50]
-; BDVER2-NEXT:    rolq %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    rorq %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    rolq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rorq %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rolq (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorq (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rolq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rorq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rolq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rolq %cl, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rorq %cl, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    rolq %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    rorq %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_rol_ror_64:
 ; BTVER2:       # %bb.0:
@@ -13143,23 +13143,23 @@
 ; BDVER2-NEXT:    sarb %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    shlb %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    shrb %dil # sched: [1:0.50]
-; BDVER2-NEXT:    sarb (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shlb (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrb (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    sarb (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlb (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrb (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    sarb $7, %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    shlb $7, %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    shrb $7, %dil # sched: [1:0.50]
-; BDVER2-NEXT:    sarb $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shlb $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrb $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    sarb %cl, %dil # sched: [3:1.50]
-; BDVER2-NEXT:    shlb %cl, %dil # sched: [3:1.50]
-; BDVER2-NEXT:    shrb %cl, %dil # sched: [3:1.50]
-; BDVER2-NEXT:    sarb %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shlb %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shrb %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    sarb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrb $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    sarb %cl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    shlb %cl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    shrb %cl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    sarb %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlb %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrb %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sar_shl_shr_8:
 ; BTVER2:       # %bb.0:
@@ -13410,23 +13410,23 @@
 ; BDVER2-NEXT:    sarw %di # sched: [1:0.50]
 ; BDVER2-NEXT:    shlw %di # sched: [1:0.50]
 ; BDVER2-NEXT:    shrw %di # sched: [1:0.50]
-; BDVER2-NEXT:    sarw (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shlw (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrw (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    sarw (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlw (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrw (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    sarw $7, %di # sched: [1:0.50]
 ; BDVER2-NEXT:    shlw $7, %di # sched: [1:0.50]
 ; BDVER2-NEXT:    shrw $7, %di # sched: [1:0.50]
-; BDVER2-NEXT:    sarw $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shlw $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrw $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    sarw %cl, %di # sched: [3:1.50]
-; BDVER2-NEXT:    shlw %cl, %di # sched: [3:1.50]
-; BDVER2-NEXT:    shrw %cl, %di # sched: [3:1.50]
-; BDVER2-NEXT:    sarw %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shlw %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shrw %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    sarw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrw $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    sarw %cl, %di # sched: [1:0.50]
+; BDVER2-NEXT:    shlw %cl, %di # sched: [1:0.50]
+; BDVER2-NEXT:    shrw %cl, %di # sched: [1:0.50]
+; BDVER2-NEXT:    sarw %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlw %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrw %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sar_shl_shr_16:
 ; BTVER2:       # %bb.0:
@@ -13677,23 +13677,23 @@
 ; BDVER2-NEXT:    sarl %edi # sched: [1:0.50]
 ; BDVER2-NEXT:    shll %edi # sched: [1:0.50]
 ; BDVER2-NEXT:    shrl %edi # sched: [1:0.50]
-; BDVER2-NEXT:    sarl (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shll (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrl (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    sarl (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shll (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrl (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    sarl $7, %edi # sched: [1:0.50]
 ; BDVER2-NEXT:    shll $7, %edi # sched: [1:0.50]
 ; BDVER2-NEXT:    shrl $7, %edi # sched: [1:0.50]
-; BDVER2-NEXT:    sarl $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shll $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrl $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    sarl %cl, %edi # sched: [3:1.50]
-; BDVER2-NEXT:    shll %cl, %edi # sched: [3:1.50]
-; BDVER2-NEXT:    shrl %cl, %edi # sched: [3:1.50]
-; BDVER2-NEXT:    sarl %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shll %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shrl %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    sarl $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shll $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrl $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    sarl %cl, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    shll %cl, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    shrl %cl, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    sarl %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shll %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrl %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sar_shl_shr_32:
 ; BTVER2:       # %bb.0:
@@ -13944,23 +13944,23 @@
 ; BDVER2-NEXT:    sarq %rdi # sched: [1:0.50]
 ; BDVER2-NEXT:    shlq %rdi # sched: [1:0.50]
 ; BDVER2-NEXT:    shrq %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    sarq (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shlq (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrq (%rdx) # sched: [7:1.00]
+; BDVER2-NEXT:    sarq (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlq (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrq (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    sarq $7, %rdi # sched: [1:0.50]
 ; BDVER2-NEXT:    shlq $7, %rdi # sched: [1:0.50]
 ; BDVER2-NEXT:    shrq $7, %rdi # sched: [1:0.50]
-; BDVER2-NEXT:    sarq $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shlq $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    shrq $7, (%rdx) # sched: [7:1.00]
-; BDVER2-NEXT:    sarq %cl, %rdi # sched: [3:1.50]
-; BDVER2-NEXT:    shlq %cl, %rdi # sched: [3:1.50]
-; BDVER2-NEXT:    shrq %cl, %rdi # sched: [3:1.50]
-; BDVER2-NEXT:    sarq %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shlq %cl, (%rdx) # sched: [9:1.50]
-; BDVER2-NEXT:    shrq %cl, (%rdx) # sched: [9:1.50]
+; BDVER2-NEXT:    sarq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrq $7, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    sarq %cl, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    shlq %cl, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    shrq %cl, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    sarq %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shlq %cl, (%rdx) # sched: [5:1.00]
+; BDVER2-NEXT:    shrq %cl, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sar_shl_shr_64:
 ; BTVER2:       # %bb.0:
@@ -14113,14 +14113,14 @@
 ; BDVER2-LABEL: test_sbb_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    sbbb $7, %al # sched: [2:0.67]
-; BDVER2-NEXT:    sbbb $7, %dil # sched: [2:0.67]
-; BDVER2-NEXT:    sbbb $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbb %dl, %dil # sched: [2:0.67]
-; BDVER2-NEXT:    sbbb %dil, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbb (%rsi), %dil # sched: [7:0.67]
+; BDVER2-NEXT:    sbbb $7, %al # sched: [1:1.00]
+; BDVER2-NEXT:    sbbb $7, %dil # sched: [1:1.00]
+; BDVER2-NEXT:    sbbb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbb %dl, %dil # sched: [1:1.00]
+; BDVER2-NEXT:    sbbb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbb (%rsi), %dil # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sbb_8:
 ; BTVER2:       # %bb.0:
@@ -14289,18 +14289,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    sbbw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    sbbw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    sbbw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [9:1.00]
-; BDVER2-NEXT:    sbbw $7, %di # sched: [2:0.67]
-; BDVER2-NEXT:    sbbw $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbw %dx, %di # sched: [2:0.67]
-; BDVER2-NEXT:    sbbw %di, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbw (%rsi), %di # sched: [7:0.67]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    sbbw $7, %di # sched: [1:1.00]
+; BDVER2-NEXT:    sbbw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbw %dx, %di # sched: [1:1.00]
+; BDVER2-NEXT:    sbbw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbw (%rsi), %di # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sbb_16:
 ; BTVER2:       # %bb.0:
@@ -14479,18 +14479,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    sbbl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    sbbl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    sbbl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [9:1.00]
-; BDVER2-NEXT:    sbbl $7, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    sbbl $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbl %edx, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    sbbl %edi, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbl (%rsi), %edi # sched: [7:0.67]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    sbbl $7, %edi # sched: [1:1.00]
+; BDVER2-NEXT:    sbbl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbl %edx, %edi # sched: [1:1.00]
+; BDVER2-NEXT:    sbbl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbl (%rsi), %edi # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sbb_32:
 ; BTVER2:       # %bb.0:
@@ -14669,18 +14669,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    sbbq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    sbbq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [2:0.67]
+; BDVER2-NEXT:    # sched: [1:1.00]
 ; BDVER2-NEXT:    sbbq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [9:1.00]
-; BDVER2-NEXT:    sbbq $7, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    sbbq $7, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbq %rdx, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    sbbq %rdi, (%rsi) # sched: [9:1.00]
-; BDVER2-NEXT:    sbbq (%rsi), %rdi # sched: [7:0.67]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    sbbq $7, %rdi # sched: [1:1.00]
+; BDVER2-NEXT:    sbbq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbq %rdx, %rdi # sched: [1:1.00]
+; BDVER2-NEXT:    sbbq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    sbbq (%rsi), %rdi # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sbb_64:
 ; BTVER2:       # %bb.0:
@@ -14803,12 +14803,12 @@
 ; BDVER2-LABEL: test_scas:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    scasb %es:(%rdi), %al # sched: [2:0.67]
-; BDVER2-NEXT:    scasw %es:(%rdi), %ax # sched: [2:0.67]
-; BDVER2-NEXT:    scasl %es:(%rdi), %eax # sched: [2:0.67]
-; BDVER2-NEXT:    scasq %es:(%rdi), %rax # sched: [2:0.67]
+; BDVER2-NEXT:    scasb %es:(%rdi), %al # sched: [100:0.50]
+; BDVER2-NEXT:    scasw %es:(%rdi), %ax # sched: [100:0.50]
+; BDVER2-NEXT:    scasl %es:(%rdi), %eax # sched: [100:0.50]
+; BDVER2-NEXT:    scasq %es:(%rdi), %rax # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_scas:
 ; BTVER2:       # %bb.0:
@@ -15147,8 +15147,8 @@
 ; BDVER2-NEXT:    setae %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    setne %dil # sched: [1:0.50]
-; BDVER2-NEXT:    setbe %dil # sched: [2:1.00]
-; BDVER2-NEXT:    seta %dil # sched: [2:1.00]
+; BDVER2-NEXT:    setbe %dil # sched: [1:0.50]
+; BDVER2-NEXT:    seta %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    sets %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    setns %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    setp %dil # sched: [1:0.50]
@@ -15157,24 +15157,24 @@
 ; BDVER2-NEXT:    setge %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    setle %dil # sched: [1:0.50]
 ; BDVER2-NEXT:    setg %dil # sched: [1:0.50]
-; BDVER2-NEXT:    seto (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setno (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setb (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setae (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    sete (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setne (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setbe (%rsi) # sched: [3:1.00]
-; BDVER2-NEXT:    seta (%rsi) # sched: [3:1.00]
-; BDVER2-NEXT:    sets (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setns (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setp (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setnp (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setl (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setge (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setle (%rsi) # sched: [2:1.00]
-; BDVER2-NEXT:    setg (%rsi) # sched: [2:1.00]
+; BDVER2-NEXT:    seto (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setno (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setb (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setae (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    sete (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setne (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setbe (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    seta (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    sets (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setns (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setp (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setnp (%rsi) # sched: [1:0.50]
+; BDVER2-NEXT:    setl (%rsi) # sched: [1:1.00]
+; BDVER2-NEXT:    setge (%rsi) # sched: [1:1.00]
+; BDVER2-NEXT:    setle (%rsi) # sched: [1:1.00]
+; BDVER2-NEXT:    setg (%rsi) # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_setcc:
 ; BTVER2:       # %bb.0:
@@ -15373,16 +15373,16 @@
 ; BDVER2-LABEL: test_shld_shrd_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    shldw %cl, %si, %di # sched: [4:1.50]
-; BDVER2-NEXT:    shrdw %cl, %si, %di # sched: [4:1.50]
-; BDVER2-NEXT:    shldw %cl, %si, (%rdx) # sched: [10:1.50]
-; BDVER2-NEXT:    shrdw %cl, %si, (%rdx) # sched: [10:1.50]
-; BDVER2-NEXT:    shldw $7, %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    shrdw $7, %si, %di # sched: [2:0.67]
-; BDVER2-NEXT:    shldw $7, %si, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    shrdw $7, %si, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    shldw %cl, %si, %di # sched: [4:4.00]
+; BDVER2-NEXT:    shrdw %cl, %si, %di # sched: [4:4.00]
+; BDVER2-NEXT:    shldw %cl, %si, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shrdw %cl, %si, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shldw $7, %si, %di # sched: [4:3.00]
+; BDVER2-NEXT:    shrdw $7, %si, %di # sched: [3:3.00]
+; BDVER2-NEXT:    shldw $7, %si, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shrdw $7, %si, (%rdx) # sched: [4:11.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_shld_shrd_16:
 ; BTVER2:       # %bb.0:
@@ -15530,16 +15530,16 @@
 ; BDVER2-LABEL: test_shld_shrd_32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    shldl %cl, %esi, %edi # sched: [4:1.50]
-; BDVER2-NEXT:    shrdl %cl, %esi, %edi # sched: [4:1.50]
-; BDVER2-NEXT:    shldl %cl, %esi, (%rdx) # sched: [10:1.50]
-; BDVER2-NEXT:    shrdl %cl, %esi, (%rdx) # sched: [10:1.50]
-; BDVER2-NEXT:    shldl $7, %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    shrdl $7, %esi, %edi # sched: [2:0.67]
-; BDVER2-NEXT:    shldl $7, %esi, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    shrdl $7, %esi, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    shldl %cl, %esi, %edi # sched: [4:4.00]
+; BDVER2-NEXT:    shrdl %cl, %esi, %edi # sched: [4:4.00]
+; BDVER2-NEXT:    shldl %cl, %esi, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shrdl %cl, %esi, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shldl $7, %esi, %edi # sched: [3:3.00]
+; BDVER2-NEXT:    shrdl $7, %esi, %edi # sched: [4:3.00]
+; BDVER2-NEXT:    shldl $7, %esi, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shrdl $7, %esi, (%rdx) # sched: [4:11.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_shld_shrd_32:
 ; BTVER2:       # %bb.0:
@@ -15687,16 +15687,16 @@
 ; BDVER2-LABEL: test_shld_shrd_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    shldq %cl, %rsi, %rdi # sched: [4:1.50]
-; BDVER2-NEXT:    shrdq %cl, %rsi, %rdi # sched: [4:1.50]
-; BDVER2-NEXT:    shldq %cl, %rsi, (%rdx) # sched: [10:1.50]
-; BDVER2-NEXT:    shrdq %cl, %rsi, (%rdx) # sched: [10:1.50]
-; BDVER2-NEXT:    shldq $7, %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    shrdq $7, %rsi, %rdi # sched: [2:0.67]
-; BDVER2-NEXT:    shldq $7, %rsi, (%rdx) # sched: [8:1.00]
-; BDVER2-NEXT:    shrdq $7, %rsi, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    shldq %cl, %rsi, %rdi # sched: [4:4.00]
+; BDVER2-NEXT:    shrdq %cl, %rsi, %rdi # sched: [4:4.00]
+; BDVER2-NEXT:    shldq %cl, %rsi, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shrdq %cl, %rsi, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shldq $7, %rsi, %rdi # sched: [4:3.00]
+; BDVER2-NEXT:    shrdq $7, %rsi, %rdi # sched: [4:3.00]
+; BDVER2-NEXT:    shldq $7, %rsi, (%rdx) # sched: [4:11.00]
+; BDVER2-NEXT:    shrdq $7, %rsi, (%rdx) # sched: [4:11.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_shld_shrd_64:
 ; BTVER2:       # %bb.0:
@@ -15801,10 +15801,10 @@
 ; BDVER2-LABEL: test_stc_std:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    stc # sched: [1:0.33]
-; BDVER2-NEXT:    std # sched: [1:0.33]
+; BDVER2-NEXT:    stc # sched: [1:0.50]
+; BDVER2-NEXT:    std # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_stc_std:
 ; BTVER2:       # %bb.0:
@@ -15912,12 +15912,12 @@
 ; BDVER2-LABEL: test_stos:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    stosb %al, %es:(%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    stosw %ax, %es:(%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    stosl %eax, %es:(%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    stosq %rax, %es:(%rdi) # sched: [5:1.00]
+; BDVER2-NEXT:    stosb %al, %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    stosw %ax, %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    stosl %eax, %es:(%rdi) # sched: [100:0.50]
+; BDVER2-NEXT:    stosq %rax, %es:(%rdi) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_stos:
 ; BTVER2:       # %bb.0:
@@ -16044,14 +16044,14 @@
 ; BDVER2-LABEL: test_sub_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    subb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    subb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    subb $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subb %dl, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    subb %dil, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subb (%rsi), %dil # sched: [6:0.50]
+; BDVER2-NEXT:    subb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    subb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    subb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subb %dl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    subb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subb (%rsi), %dil # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sub_8:
 ; BTVER2:       # %bb.0:
@@ -16220,18 +16220,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    subw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    subw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    subw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    subw $7, %di # sched: [1:0.33]
-; BDVER2-NEXT:    subw $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subw %dx, %di # sched: [1:0.33]
-; BDVER2-NEXT:    subw %di, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subw (%rsi), %di # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    subw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    subw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subw %dx, %di # sched: [1:0.50]
+; BDVER2-NEXT:    subw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subw (%rsi), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sub_16:
 ; BTVER2:       # %bb.0:
@@ -16410,18 +16410,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    subl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    subl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    subl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    subl $7, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    subl $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subl %edx, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    subl %edi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subl (%rsi), %edi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    subl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    subl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subl %edx, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    subl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subl (%rsi), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sub_32:
 ; BTVER2:       # %bb.0:
@@ -16600,18 +16600,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    subq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    subq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    subq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    subq $7, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    subq $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subq %rdx, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    subq %rdi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    subq (%rsi), %rdi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    subq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    subq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subq %rdx, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    subq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    subq (%rsi), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_sub_64:
 ; BTVER2:       # %bb.0:
@@ -16748,13 +16748,13 @@
 ; BDVER2-LABEL: test_test_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    testb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    testb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    testb $7, (%rsi) # sched: [6:0.50]
-; BDVER2-NEXT:    testb %dil, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    testb %dil, (%rsi) # sched: [6:0.50]
+; BDVER2-NEXT:    testb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    testb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    testb $7, (%rsi) # sched: [5:0.50]
+; BDVER2-NEXT:    testb %dil, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    testb %dil, (%rsi) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_test_8:
 ; BTVER2:       # %bb.0:
@@ -16897,15 +16897,15 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    testw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    testw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    testw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [6:0.50]
-; BDVER2-NEXT:    testw %di, %di # sched: [1:0.33]
-; BDVER2-NEXT:    testw %di, (%rsi) # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [5:0.50]
+; BDVER2-NEXT:    testw %di, %di # sched: [1:0.50]
+; BDVER2-NEXT:    testw %di, (%rsi) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_test_16:
 ; BTVER2:       # %bb.0:
@@ -17054,15 +17054,15 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    testl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    testl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    testl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [6:0.50]
-; BDVER2-NEXT:    testl %edi, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    testl %edi, (%rsi) # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [5:0.50]
+; BDVER2-NEXT:    testl %edi, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    testl %edi, (%rsi) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_test_32:
 ; BTVER2:       # %bb.0:
@@ -17211,15 +17211,15 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    testq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    testq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    testq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [6:0.50]
-; BDVER2-NEXT:    testq %rdi, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    testq %rdi, (%rsi) # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [5:0.50]
+; BDVER2-NEXT:    testq %rdi, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    testq %rdi, (%rsi) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_test_64:
 ; BTVER2:       # %bb.0:
@@ -17313,9 +17313,9 @@
 ; BDVER2-LABEL: test_ud2:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    ud2 # sched: [100:0.33]
+; BDVER2-NEXT:    ud2 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ud2:
 ; BTVER2:       # %bb.0:
@@ -17411,9 +17411,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xaddb %dil, %sil # sched: [2:1.00]
-; BDVER2-NEXT:    xaddb %dil, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    xaddb %dil, (%rdx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xadd_8:
 ; BTVER2:       # %bb.0:
@@ -17502,9 +17502,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xaddw %di, %si # sched: [2:1.00]
-; BDVER2-NEXT:    xaddw %di, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    xaddw %di, (%rdx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xadd_16:
 ; BTVER2:       # %bb.0:
@@ -17593,9 +17593,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xaddl %edi, %esi # sched: [2:1.00]
-; BDVER2-NEXT:    xaddl %edi, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    xaddl %edi, (%rdx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xadd_32:
 ; BTVER2:       # %bb.0:
@@ -17684,9 +17684,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xaddq %rdi, %rsi # sched: [2:1.00]
-; BDVER2-NEXT:    xaddq %rdi, (%rdx) # sched: [8:1.00]
+; BDVER2-NEXT:    xaddq %rdi, (%rdx) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xadd_64:
 ; BTVER2:       # %bb.0:
@@ -17775,10 +17775,10 @@
 ; BDVER2-LABEL: test_xchg_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xchgb %sil, %dil # sched: [2:1.00]
-; BDVER2-NEXT:    xchgb %dil, (%rdx) # sched: [6:1.00]
+; BDVER2-NEXT:    xchgb %sil, %dil # sched: [1:1.00]
+; BDVER2-NEXT:    xchgb %dil, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xchg_8:
 ; BTVER2:       # %bb.0:
@@ -17874,11 +17874,11 @@
 ; BDVER2-LABEL: test_xchg_16:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xchgw %di, %ax # sched: [2:1.00]
+; BDVER2-NEXT:    xchgw %di, %ax # sched: [1:1.00]
 ; BDVER2-NEXT:    xchgw %si, %di # sched: [2:1.00]
-; BDVER2-NEXT:    xchgw %di, (%rdx) # sched: [6:1.00]
+; BDVER2-NEXT:    xchgw %di, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xchg_16:
 ; BTVER2:       # %bb.0:
@@ -17976,11 +17976,11 @@
 ; BDVER2-LABEL: test_xchg_32:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xchgl %edi, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    xchgl %esi, %edi # sched: [2:1.00]
-; BDVER2-NEXT:    xchgl %edi, (%rdx) # sched: [6:1.00]
+; BDVER2-NEXT:    xchgl %edi, %eax # sched: [1:1.00]
+; BDVER2-NEXT:    xchgl %esi, %edi # sched: [1:1.00]
+; BDVER2-NEXT:    xchgl %edi, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xchg_32:
 ; BTVER2:       # %bb.0:
@@ -18078,11 +18078,11 @@
 ; BDVER2-LABEL: test_xchg_64:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xchgq %rdi, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    xchgq %rsi, %rdi # sched: [2:1.00]
-; BDVER2-NEXT:    xchgq %rdi, (%rdx) # sched: [6:1.00]
+; BDVER2-NEXT:    xchgq %rdi, %rax # sched: [1:1.00]
+; BDVER2-NEXT:    xchgq %rsi, %rdi # sched: [1:1.00]
+; BDVER2-NEXT:    xchgq %rdi, (%rdx) # sched: [5:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xchg_64:
 ; BTVER2:       # %bb.0:
@@ -18165,9 +18165,9 @@
 ; BDVER2-LABEL: test_xlat:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xlatb # sched: [5:0.50]
+; BDVER2-NEXT:    xlatb # sched: [6:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xlat:
 ; BTVER2:       # %bb.0:
@@ -18286,14 +18286,14 @@
 ; BDVER2-LABEL: test_xor_8:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    xorb $7, %al # sched: [1:0.33]
-; BDVER2-NEXT:    xorb $7, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    xorb $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorb %dl, %dil # sched: [1:0.33]
-; BDVER2-NEXT:    xorb %dil, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorb (%rsi), %dil # sched: [6:0.50]
+; BDVER2-NEXT:    xorb $7, %al # sched: [1:0.50]
+; BDVER2-NEXT:    xorb $7, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    xorb $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorb %dl, %dil # sched: [1:0.50]
+; BDVER2-NEXT:    xorb %dil, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorb (%rsi), %dil # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xor_8:
 ; BTVER2:       # %bb.0:
@@ -18462,18 +18462,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xorw $511, %ax # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    xorw $511, %di # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    xorw $511, (%rsi) # imm = 0x1FF
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    xorw $7, %di # sched: [1:0.33]
-; BDVER2-NEXT:    xorw $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorw %dx, %di # sched: [1:0.33]
-; BDVER2-NEXT:    xorw %di, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorw (%rsi), %di # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    xorw $7, %di # sched: [1:0.50]
+; BDVER2-NEXT:    xorw $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorw %dx, %di # sched: [1:0.50]
+; BDVER2-NEXT:    xorw %di, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorw (%rsi), %di # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xor_16:
 ; BTVER2:       # %bb.0:
@@ -18652,18 +18652,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xorl $665536, %eax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    xorl $665536, %edi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    xorl $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    xorl $7, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    xorl $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorl %edx, %edi # sched: [1:0.33]
-; BDVER2-NEXT:    xorl %edi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorl (%rsi), %edi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    xorl $7, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    xorl $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorl %edx, %edi # sched: [1:0.50]
+; BDVER2-NEXT:    xorl %edi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorl (%rsi), %edi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xor_32:
 ; BTVER2:       # %bb.0:
@@ -18842,18 +18842,18 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    xorq $665536, %rax # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    xorq $665536, %rdi # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [1:0.33]
+; BDVER2-NEXT:    # sched: [1:0.50]
 ; BDVER2-NEXT:    xorq $665536, (%rsi) # imm = 0xA27C0
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    xorq $7, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    xorq $7, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorq %rdx, %rdi # sched: [1:0.33]
-; BDVER2-NEXT:    xorq %rdi, (%rsi) # sched: [7:1.00]
-; BDVER2-NEXT:    xorq (%rsi), %rdi # sched: [6:0.50]
+; BDVER2-NEXT:    # sched: [6:1.00]
+; BDVER2-NEXT:    xorq $7, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    xorq $7, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorq %rdx, %rdi # sched: [1:0.50]
+; BDVER2-NEXT:    xorq %rdi, (%rsi) # sched: [6:1.00]
+; BDVER2-NEXT:    xorq (%rsi), %rdi # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_xor_64:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/small-byval-memcpy.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/small-byval-memcpy.ll
+++ llvm/trunk/test/CodeGen/X86/small-byval-memcpy.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core2 | FileCheck %s --check-prefix=CORE2
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=nehalem | FileCheck %s --check-prefix=NEHALEM
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=x86-64 | FileCheck %s --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=bdver2 | FileCheck %s --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=btver2 | FileCheck %s --check-prefix=BTVER2
 
 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i1)
Index: llvm/trunk/test/CodeGen/X86/sse-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/sse-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/sse-schedule.ll
@@ -14,8 +14,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
@@ -104,15 +104,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_addps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    addps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    addps (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_addps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vaddps (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_addps:
 ; BTVER2-SSE:       # %bb.0:
@@ -224,15 +224,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_addss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    addss (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    addss (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_addss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vaddss (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_addss:
 ; BTVER2-SSE:       # %bb.0:
@@ -348,15 +348,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_andps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    andps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    andps (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andps %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    andps (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_andps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandps %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vandps (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_andps:
 ; BTVER2-SSE:       # %bb.0:
@@ -476,15 +476,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_andnotps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    andnps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    andnps (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andnps %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    andnps (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_andnotps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandnps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandnps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandnps %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vandnps (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_andnotps:
 ; BTVER2-SSE:       # %bb.0:
@@ -615,17 +615,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cmpps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cmpeqps %xmm0, %xmm1 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    cmpeqps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    orps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cmpeqps %xmm0, %xmm1 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    cmpeqps (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    orps %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cmpps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
-; BDVER2-NEXT:    vcmpeqps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    vorps %xmm0, %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm1 # sched: [2:1.00]
+; BDVER2-NEXT:    vcmpeqps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vorps %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cmpps:
 ; BTVER2-SSE:       # %bb.0:
@@ -745,15 +745,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_cmpss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cmpeqss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    cmpeqss (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cmpeqss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    cmpeqss (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cmpss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcmpeqss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vcmpeqss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcmpeqss %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vcmpeqss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cmpss:
 ; BTVER2-SSE:       # %bb.0:
@@ -974,31 +974,31 @@
 ;
 ; BDVER2-SSE-LABEL: test_comiss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    comiss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    comiss %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    comiss (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    comiss (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_comiss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcomiss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vcomiss %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-NEXT:    vcomiss (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-NEXT:    vcomiss (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_comiss:
 ; BTVER2-SSE:       # %bb.0:
@@ -1157,17 +1157,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtsi2ss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtsi2ssl %edi, %xmm1 # sched: [5:2.00]
-; BDVER2-SSE-NEXT:    cvtsi2ssl (%rsi), %xmm0 # sched: [10:1.00]
-; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtsi2ssl (%rsi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    cvtsi2ssl %edi, %xmm1 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsi2ss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtsi2ssl %edi, %xmm0, %xmm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vcvtsi2ssl (%rsi), %xmm1, %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtsi2ssl %edi, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vcvtsi2ssl (%rsi), %xmm1, %xmm1 # sched: [9:1.00]
+; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsi2ss:
 ; BTVER2-SSE:       # %bb.0:
@@ -1297,17 +1297,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtsi2ssq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtsi2ssq %rdi, %xmm1 # sched: [5:2.00]
-; BDVER2-SSE-NEXT:    cvtsi2ssq (%rsi), %xmm0 # sched: [10:1.00]
-; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtsi2ssq %rdi, %xmm1 # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    cvtsi2ssq (%rsi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsi2ssq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtsi2ssq %rdi, %xmm0, %xmm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vcvtsi2ssq (%rsi), %xmm1, %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtsi2ssq %rdi, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vcvtsi2ssq (%rsi), %xmm1, %xmm1 # sched: [9:1.00]
+; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsi2ssq:
 ; BTVER2-SSE:       # %bb.0:
@@ -1437,17 +1437,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtss2si:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtss2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvtss2si (%rdi), %eax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtss2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvtss2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtss2si:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtss2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvtss2si (%rdi), %eax # sched: [10:1.00]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtss2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvtss2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtss2si:
 ; BTVER2-SSE:       # %bb.0:
@@ -1580,17 +1580,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtss2siq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtss2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvtss2si (%rdi), %rax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtss2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvtss2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtss2siq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtss2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvtss2si (%rdi), %rax # sched: [10:1.00]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtss2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvtss2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtss2siq:
 ; BTVER2-SSE:       # %bb.0:
@@ -1723,17 +1723,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvttss2si:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvttss2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvttss2si (%rdi), %eax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvttss2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvttss2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvttss2si:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttss2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvttss2si (%rdi), %eax # sched: [10:1.00]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttss2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvttss2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvttss2si:
 ; BTVER2-SSE:       # %bb.0:
@@ -1863,17 +1863,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvttss2siq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvttss2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvttss2si (%rdi), %rax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvttss2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvttss2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvttss2siq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttss2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvttss2si (%rdi), %rax # sched: [10:1.00]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttss2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvttss2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvttss2siq:
 ; BTVER2-SSE:       # %bb.0:
@@ -1990,15 +1990,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_divps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    divps %xmm1, %xmm0 # sched: [14:14.00]
-; BDVER2-SSE-NEXT:    divps (%rdi), %xmm0 # sched: [20:14.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    divps %xmm1, %xmm0 # sched: [9:9.50]
+; BDVER2-SSE-NEXT:    divps (%rdi), %xmm0 # sched: [14:9.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_divps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdivps %xmm1, %xmm0, %xmm0 # sched: [14:14.00]
-; BDVER2-NEXT:    vdivps (%rdi), %xmm0, %xmm0 # sched: [20:14.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdivps %xmm1, %xmm0, %xmm0 # sched: [9:9.50]
+; BDVER2-NEXT:    vdivps (%rdi), %xmm0, %xmm0 # sched: [14:9.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_divps:
 ; BTVER2-SSE:       # %bb.0:
@@ -2110,15 +2110,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_divss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    divss %xmm1, %xmm0 # sched: [14:14.00]
-; BDVER2-SSE-NEXT:    divss (%rdi), %xmm0 # sched: [20:14.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    divss %xmm1, %xmm0 # sched: [9:9.50]
+; BDVER2-SSE-NEXT:    divss (%rdi), %xmm0 # sched: [14:9.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_divss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdivss %xmm1, %xmm0, %xmm0 # sched: [14:14.00]
-; BDVER2-NEXT:    vdivss (%rdi), %xmm0, %xmm0 # sched: [20:14.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdivss %xmm1, %xmm0, %xmm0 # sched: [9:9.50]
+; BDVER2-NEXT:    vdivss (%rdi), %xmm0, %xmm0 # sched: [14:9.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_divss:
 ; BTVER2-SSE:       # %bb.0:
@@ -2230,15 +2230,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_ldmxcsr:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl %edi, -{{[0-9]+}}(%rsp) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    ldmxcsr -{{[0-9]+}}(%rsp) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl %edi, -{{[0-9]+}}(%rsp) # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    ldmxcsr -{{[0-9]+}}(%rsp) # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_ldmxcsr:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %edi, -{{[0-9]+}}(%rsp) # sched: [1:1.00]
-; BDVER2-NEXT:    vldmxcsr -{{[0-9]+}}(%rsp) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl %edi, -{{[0-9]+}}(%rsp) # sched: [1:0.50]
+; BDVER2-NEXT:    vldmxcsr -{{[0-9]+}}(%rsp) # sched: [5:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_ldmxcsr:
 ; BTVER2-SSE:       # %bb.0:
@@ -2352,15 +2352,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_maxps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    maxps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    maxps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    maxps %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    maxps (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_maxps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaxps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmaxps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaxps %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vmaxps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_maxps:
 ; BTVER2-SSE:       # %bb.0:
@@ -2473,15 +2473,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_maxss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    maxss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    maxss (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    maxss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    maxss (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_maxss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaxss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmaxss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaxss %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vmaxss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_maxss:
 ; BTVER2-SSE:       # %bb.0:
@@ -2594,15 +2594,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_minps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    minps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    minps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    minps %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    minps (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_minps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vminps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vminps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vminps %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vminps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_minps:
 ; BTVER2-SSE:       # %bb.0:
@@ -2715,15 +2715,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_minss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    minss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    minss (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    minss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    minss (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_minss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vminss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vminss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vminss %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vminss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_minss:
 ; BTVER2-SSE:       # %bb.0:
@@ -2849,17 +2849,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movaps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movaps (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addps %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    movaps (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-SSE-NEXT:    movaps %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movaps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovaps (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    vaddps %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovaps (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddps %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-NEXT:    vmovaps %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movaps:
 ; BTVER2-SSE:       # %bb.0:
@@ -2970,13 +2970,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_movhlps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movhlps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm1[1],xmm0[1] sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm1[1],xmm0[1] sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movhlps:
 ; BTVER2-SSE:       # %bb.0:
@@ -3111,19 +3111,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_movhps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movhps %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movaps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movhps %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    movaps %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movhps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovhpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    vmovaps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovhpd %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    vmovaps %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movhps:
 ; BTVER2-SSE:       # %bb.0:
@@ -3249,15 +3249,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_movlhps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movlhps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:1.00]
-; BDVER2-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [2:0.50]
+; BDVER2-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movlhps:
 ; BTVER2-SSE:       # %bb.0:
@@ -3395,19 +3395,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_movlps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movlps %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movaps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movlps %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    movaps %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movlps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovlps %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    vmovaps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovlps %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    vmovaps %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movlps:
 ; BTVER2-SSE:       # %bb.0:
@@ -3520,13 +3520,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_movmskps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movmskps %xmm0, %eax # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movmskps %xmm0, %eax # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movmskps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovmskps %xmm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovmskps %xmm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movmskps:
 ; BTVER2-SSE:       # %bb.0:
@@ -3626,13 +3626,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_movntps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movntps %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movntps %xmm0, (%rdi) # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movntps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovntps %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovntps %xmm0, (%rdi) # sched: [3:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movntps:
 ; BTVER2-SSE:       # %bb.0:
@@ -3751,17 +3751,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movss_mem:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addss %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movss %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-SSE-NEXT:    addss %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movss %xmm0, (%rsi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movss_mem:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-NEXT:    vaddss %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovss %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-NEXT:    vaddss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovss %xmm0, (%rsi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movss_mem:
 ; BTVER2-SSE:       # %bb.0:
@@ -3870,13 +3870,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_movss_reg:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movss_reg:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movss_reg:
 ; BTVER2-SSE:       # %bb.0:
@@ -3995,17 +3995,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movups:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movups (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addps %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    movups (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-SSE-NEXT:    movups %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movups:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovups (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    vaddps %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovups (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddps %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-NEXT:    vmovups %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movups:
 ; BTVER2-SSE:       # %bb.0:
@@ -4122,14 +4122,14 @@
 ; BDVER2-SSE-LABEL: test_mulps:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    mulps %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    mulps (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    mulps (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mulps:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmulps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulps (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulps (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mulps:
 ; BTVER2-SSE:       # %bb.0:
@@ -4242,14 +4242,14 @@
 ; BDVER2-SSE-LABEL: test_mulss:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    mulss %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    mulss (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    mulss (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mulss:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmulss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulss (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulss (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mulss:
 ; BTVER2-SSE:       # %bb.0:
@@ -4365,15 +4365,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_orps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    orps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    orps (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    orps %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    orps (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_orps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vorps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vorps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vorps %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vorps (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_orps:
 ; BTVER2-SSE:       # %bb.0:
@@ -4547,7 +4547,7 @@
 ; BDVER2-SSE-NEXT:    prefetcht1 (%rdi) # sched: [5:0.50]
 ; BDVER2-SSE-NEXT:    prefetcht2 (%rdi) # sched: [5:0.50]
 ; BDVER2-SSE-NEXT:    #NO_APP
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_prefetch:
 ; BDVER2:       # %bb.0:
@@ -4557,7 +4557,7 @@
 ; BDVER2-NEXT:    prefetcht1 (%rdi) # sched: [5:0.50]
 ; BDVER2-NEXT:    prefetcht2 (%rdi) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_prefetch:
 ; BTVER2-SSE:       # %bb.0:
@@ -4699,16 +4699,16 @@
 ; BDVER2-SSE-LABEL: test_rcpps:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    rcpps %xmm0, %xmm1 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    rcpps (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    rcpps (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_rcpps:
 ; BDVER2:       # %bb.0:
+; BDVER2-NEXT:    vrcpps (%rdi), %xmm1 # sched: [10:1.00]
 ; BDVER2-NEXT:    vrcpps %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vrcpps (%rdi), %xmm1 # sched: [11:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_rcpps:
 ; BTVER2-SSE:       # %bb.0:
@@ -4854,19 +4854,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_rcpss:
 ; BDVER2-SSE:       # %bb.0:
+; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
 ; BDVER2-SSE-NEXT:    rcpss %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
 ; BDVER2-SSE-NEXT:    rcpss %xmm1, %xmm1 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_rcpss:
 ; BDVER2:       # %bb.0:
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
 ; BDVER2-NEXT:    vrcpss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
 ; BDVER2-NEXT:    vrcpss %xmm1, %xmm1, %xmm1 # sched: [5:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_rcpss:
 ; BTVER2-SSE:       # %bb.0:
@@ -5006,16 +5006,16 @@
 ; BDVER2-SSE-LABEL: test_rsqrtps:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    rsqrtps %xmm0, %xmm1 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    rsqrtps (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    rsqrtps (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_rsqrtps:
 ; BDVER2:       # %bb.0:
+; BDVER2-NEXT:    vrsqrtps (%rdi), %xmm1 # sched: [10:1.00]
 ; BDVER2-NEXT:    vrsqrtps %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vrsqrtps (%rdi), %xmm1 # sched: [11:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_rsqrtps:
 ; BTVER2-SSE:       # %bb.0:
@@ -5161,19 +5161,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_rsqrtss:
 ; BDVER2-SSE:       # %bb.0:
+; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
 ; BDVER2-SSE-NEXT:    rsqrtss %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
 ; BDVER2-SSE-NEXT:    rsqrtss %xmm1, %xmm1 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_rsqrtss:
 ; BDVER2:       # %bb.0:
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
 ; BDVER2-NEXT:    vrsqrtss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
 ; BDVER2-NEXT:    vrsqrtss %xmm1, %xmm1, %xmm1 # sched: [5:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_rsqrtss:
 ; BTVER2-SSE:       # %bb.0:
@@ -5290,13 +5290,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_sfence:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    sfence # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    sfence # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_sfence:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    sfence # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    sfence # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_sfence:
 ; BTVER2-SSE:       # %bb.0:
@@ -5416,17 +5416,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_shufps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],mem[0,0] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],mem[0,0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_shufps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0] sched: [1:1.00]
-; BDVER2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,3],mem[0,0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0] sched: [2:0.50]
+; BDVER2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,3],mem[0,0] sched: [7:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_shufps:
 ; BTVER2-SSE:       # %bb.0:
@@ -5557,17 +5557,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_sqrtps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    sqrtps %xmm0, %xmm1 # sched: [14:14.00]
-; BDVER2-SSE-NEXT:    sqrtps (%rdi), %xmm0 # sched: [20:14.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    sqrtps %xmm0, %xmm1 # sched: [9:10.50]
+; BDVER2-SSE-NEXT:    sqrtps (%rdi), %xmm0 # sched: [14:10.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_sqrtps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsqrtps %xmm0, %xmm0 # sched: [14:14.00]
-; BDVER2-NEXT:    vsqrtps (%rdi), %xmm1 # sched: [20:14.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsqrtps (%rdi), %xmm1 # sched: [14:10.50]
+; BDVER2-NEXT:    vsqrtps %xmm0, %xmm0 # sched: [9:10.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_sqrtps:
 ; BTVER2-SSE:       # %bb.0:
@@ -5713,19 +5713,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_sqrtss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    sqrtss %xmm0, %xmm0 # sched: [14:14.00]
-; BDVER2-SSE-NEXT:    movaps (%rdi), %xmm1 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    sqrtss %xmm1, %xmm1 # sched: [14:14.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movaps (%rdi), %xmm1 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    sqrtss %xmm0, %xmm0 # sched: [9:10.50]
+; BDVER2-SSE-NEXT:    sqrtss %xmm1, %xmm1 # sched: [9:10.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_sqrtss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsqrtss %xmm0, %xmm0, %xmm0 # sched: [14:14.00]
-; BDVER2-NEXT:    vmovaps (%rdi), %xmm1 # sched: [6:0.50]
-; BDVER2-NEXT:    vsqrtss %xmm1, %xmm1, %xmm1 # sched: [14:14.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovaps (%rdi), %xmm1 # sched: [5:0.50]
+; BDVER2-NEXT:    vsqrtss %xmm0, %xmm0, %xmm0 # sched: [9:10.50]
+; BDVER2-NEXT:    vsqrtss %xmm1, %xmm1, %xmm1 # sched: [9:10.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_sqrtss:
 ; BTVER2-SSE:       # %bb.0:
@@ -5847,15 +5847,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_stmxcsr:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    stmxcsr -{{[0-9]+}}(%rsp) # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    stmxcsr -{{[0-9]+}}(%rsp) # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    movl -{{[0-9]+}}(%rsp), %eax # sched: [5:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_stmxcsr:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vstmxcsr -{{[0-9]+}}(%rsp) # sched: [5:1.00]
+; BDVER2-NEXT:    vstmxcsr -{{[0-9]+}}(%rsp) # sched: [1:0.50]
 ; BDVER2-NEXT:    movl -{{[0-9]+}}(%rsp), %eax # sched: [5:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_stmxcsr:
 ; BTVER2-SSE:       # %bb.0:
@@ -5969,15 +5969,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_subps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    subps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    subps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    subps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    subps (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_subps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsubps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vsubps (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_subps:
 ; BTVER2-SSE:       # %bb.0:
@@ -6089,15 +6089,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_subss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    subss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    subss (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    subss %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    subss (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_subss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsubss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vsubss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsubss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vsubss (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_subss:
 ; BTVER2-SSE:       # %bb.0:
@@ -6313,31 +6313,31 @@
 ;
 ; BDVER2-SSE-LABEL: test_ucomiss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    ucomiss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    ucomiss %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    ucomiss (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    ucomiss (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_ucomiss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vucomiss %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vucomiss %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-NEXT:    vucomiss (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-NEXT:    vucomiss (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_ucomiss:
 ; BTVER2-SSE:       # %bb.0:
@@ -6496,17 +6496,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_unpckhps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    unpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    unpckhps {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    unpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    unpckhps {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_unpckhps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:1.00]
-; BDVER2-NEXT:    vunpckhps {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [2:0.50]
+; BDVER2-NEXT:    vunpckhps {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_unpckhps:
 ; BTVER2-SSE:       # %bb.0:
@@ -6636,17 +6636,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_unpcklps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_unpcklps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:1.00]
-; BDVER2-NEXT:    vunpcklps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [2:0.50]
+; BDVER2-NEXT:    vunpcklps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_unpcklps:
 ; BTVER2-SSE:       # %bb.0:
@@ -6767,15 +6767,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_xorps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    xorps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    xorps (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    xorps %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    xorps (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_xorps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vxorps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vxorps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vxorps %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vxorps (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_xorps:
 ; BTVER2-SSE:       # %bb.0:
@@ -6923,19 +6923,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_fnop:
 ; BDVER2-SSE:       # %bb.0:
+; BDVER2-SSE-NEXT:    xorps %xmm0, %xmm0 # sched: [0:0.25]
 ; BDVER2-SSE-NEXT:    #APP
-; BDVER2-SSE-NEXT:    nop # sched: [1:0.25]
+; BDVER2-SSE-NEXT:    nop # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    #NO_APP
-; BDVER2-SSE-NEXT:    xorps %xmm0, %xmm0 # sched: [0:0.25]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_fnop:
 ; BDVER2:       # %bb.0:
+; BDVER2-NEXT:    vxorps %xmm0, %xmm0, %xmm0 # sched: [0:0.25]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    nop # sched: [1:0.25]
+; BDVER2-NEXT:    nop # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    vxorps %xmm0, %xmm0, %xmm0 # sched: [0:0.25]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_fnop:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/sse2-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/sse2-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/sse2-schedule.ll
@@ -14,8 +14,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx,+xop -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-sse3 | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
@@ -102,15 +102,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_addpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    addpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    addpd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_addpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vaddpd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_addpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -222,15 +222,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_addsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    addsd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    addsd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_addsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddsd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vaddsd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_addsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -355,17 +355,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_andpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    andpd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    andpd (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andpd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    andpd (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_andpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandpd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vandpd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_andpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -499,17 +499,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_andnotpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    andnpd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    andnpd (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andnpd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    andnpd (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_andnotpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vandnpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vandnpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vandnpd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vandnpd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_andnotpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -625,13 +625,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_clflush:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    clflush (%rdi) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    clflush (%rdi) # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_clflush:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    clflush (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    clflush (%rdi) # sched: [5:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_clflush:
 ; BTVER2-SSE:       # %bb.0:
@@ -751,17 +751,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cmppd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cmpeqpd %xmm0, %xmm1 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    cmpeqpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    orpd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cmpeqpd %xmm0, %xmm1 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    cmpeqpd (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    orpd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cmppd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcmpeqpd %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
-; BDVER2-NEXT:    vcmpeqpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    vorpd %xmm0, %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcmpeqpd %xmm1, %xmm0, %xmm1 # sched: [2:1.00]
+; BDVER2-NEXT:    vcmpeqpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    vorpd %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cmppd:
 ; BTVER2-SSE:       # %bb.0:
@@ -880,15 +880,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_cmpsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cmpeqsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    cmpeqsd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cmpeqsd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    cmpeqsd (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cmpsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcmpeqsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vcmpeqsd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcmpeqsd %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vcmpeqsd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cmpsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -1109,31 +1109,31 @@
 ;
 ; BDVER2-SSE-LABEL: test_comisd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    comisd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    comisd %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    comisd (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    comisd (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_comisd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcomisd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vcomisd %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-NEXT:    vcomisd (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-NEXT:    vcomisd (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_comisd:
 ; BTVER2-SSE:       # %bb.0:
@@ -1294,17 +1294,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtdq2pd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtdq2pd %xmm0, %xmm1 # sched: [4:1.00]
-; BDVER2-SSE-NEXT:    cvtdq2pd (%rdi), %xmm0 # sched: [10:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtdq2pd %xmm0, %xmm1 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    cvtdq2pd (%rdi), %xmm0 # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtdq2pd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtdq2pd %xmm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvtdq2pd (%rdi), %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtdq2pd (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    vcvtdq2pd %xmm0, %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtdq2pd:
 ; BTVER2-SSE:       # %bb.0:
@@ -1437,17 +1437,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtdq2ps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtdq2ps %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    cvtdq2ps %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-SSE-NEXT:    cvtdq2ps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtdq2ps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtdq2ps %xmm0, %xmm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    vcvtdq2ps (%rdi), %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtdq2ps %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtdq2ps:
 ; BTVER2-SSE:       # %bb.0:
@@ -1579,17 +1579,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtpd2dq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtpd2dq %xmm0, %xmm1 # sched: [4:1.00]
-; BDVER2-SSE-NEXT:    cvtpd2dq (%rdi), %xmm0 # sched: [10:1.00]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtpd2dq %xmm0, %xmm1 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    cvtpd2dq (%rdi), %xmm0 # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtpd2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtpd2dq %xmm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvtpd2dqx (%rdi), %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtpd2dqx (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    vcvtpd2dq %xmm0, %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtpd2dq:
 ; BTVER2-SSE:       # %bb.0:
@@ -1722,17 +1722,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtpd2ps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtpd2ps %xmm0, %xmm1 # sched: [4:1.00]
-; BDVER2-SSE-NEXT:    cvtpd2ps (%rdi), %xmm0 # sched: [10:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtpd2ps %xmm0, %xmm1 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    cvtpd2ps (%rdi), %xmm0 # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtpd2ps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtpd2ps %xmm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvtpd2psx (%rdi), %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtpd2psx (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    vcvtpd2ps %xmm0, %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtpd2ps:
 ; BTVER2-SSE:       # %bb.0:
@@ -1864,17 +1864,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtps2dq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtps2dq %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    cvtps2dq %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-SSE-NEXT:    cvtps2dq (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtps2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtps2dq %xmm0, %xmm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    vcvtps2dq (%rdi), %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtps2dq %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtps2dq:
 ; BTVER2-SSE:       # %bb.0:
@@ -2006,17 +2006,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtps2pd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtps2pd %xmm0, %xmm1 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    cvtps2pd (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtps2pd %xmm0, %xmm1 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    cvtps2pd (%rdi), %xmm0 # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtps2pd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtps2pd %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vcvtps2pd (%rdi), %xmm1 # sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtps2pd (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    vcvtps2pd %xmm0, %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtps2pd:
 ; BTVER2-SSE:       # %bb.0:
@@ -2148,17 +2148,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtsd2si:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtsd2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvtsd2si (%rdi), %eax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtsd2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvtsd2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsd2si:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtsd2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvtsd2si (%rdi), %eax # sched: [10:1.00]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtsd2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvtsd2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsd2si:
 ; BTVER2-SSE:       # %bb.0:
@@ -2291,17 +2291,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtsd2siq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtsd2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvtsd2si (%rdi), %rax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtsd2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvtsd2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsd2siq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtsd2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvtsd2si (%rdi), %rax # sched: [10:1.00]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvtsd2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvtsd2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsd2siq:
 ; BTVER2-SSE:       # %bb.0:
@@ -2449,18 +2449,18 @@
 ; BDVER2-SSE-LABEL: test_cvtsd2ss:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    cvtsd2ss %xmm0, %xmm1 # sched: [4:1.00]
-; BDVER2-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero sched: [6:0.50]
+; BDVER2-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero sched: [5:0.50]
 ; BDVER2-SSE-NEXT:    cvtsd2ss %xmm0, %xmm0 # sched: [4:1.00]
-; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addss %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsd2ss:
 ; BDVER2:       # %bb.0:
+; BDVER2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero sched: [5:0.50]
 ; BDVER2-NEXT:    vcvtsd2ss %xmm0, %xmm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero sched: [6:0.50]
 ; BDVER2-NEXT:    vcvtsd2ss %xmm1, %xmm1, %xmm1 # sched: [4:1.00]
-; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsd2ss:
 ; BTVER2-SSE:       # %bb.0:
@@ -2594,17 +2594,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtsi2sd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtsi2sdl %edi, %xmm1 # sched: [4:1.00]
 ; BDVER2-SSE-NEXT:    cvtsi2sdl (%rsi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtsi2sdl %edi, %xmm1 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsi2sd:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vcvtsi2sdl %edi, %xmm0, %xmm0 # sched: [4:1.00]
 ; BDVER2-NEXT:    vcvtsi2sdl (%rsi), %xmm1, %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsi2sd:
 ; BTVER2-SSE:       # %bb.0:
@@ -2734,17 +2734,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtsi2sdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtsi2sdq %rdi, %xmm1 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    cvtsi2sdq %rdi, %xmm1 # sched: [13:1.00]
 ; BDVER2-SSE-NEXT:    cvtsi2sdq (%rsi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtsi2sdq:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vcvtsi2sdq %rdi, %xmm0, %xmm0 # sched: [4:1.00]
 ; BDVER2-NEXT:    vcvtsi2sdq (%rsi), %xmm1, %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtsi2sdq:
 ; BTVER2-SSE:       # %bb.0:
@@ -2890,19 +2890,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvtss2sd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvtss2sd %xmm0, %xmm1 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-SSE-NEXT:    cvtss2sd %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvtss2sd %xmm0, %xmm1 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-SSE-NEXT:    cvtss2sd %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    addsd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvtss2sd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-NEXT:    vcvtss2sd %xmm1, %xmm1, %xmm1 # sched: [1:1.00]
-; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vcvtss2sd %xmm1, %xmm1, %xmm1 # sched: [4:1.00]
+; BDVER2-NEXT:    vaddsd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvtss2sd:
 ; BTVER2-SSE:       # %bb.0:
@@ -3038,17 +3038,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvttpd2dq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvttpd2dq %xmm0, %xmm1 # sched: [4:1.00]
-; BDVER2-SSE-NEXT:    cvttpd2dq (%rdi), %xmm0 # sched: [10:1.00]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvttpd2dq %xmm0, %xmm1 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    cvttpd2dq (%rdi), %xmm0 # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvttpd2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttpd2dq %xmm0, %xmm0 # sched: [4:1.00]
-; BDVER2-NEXT:    vcvttpd2dqx (%rdi), %xmm1 # sched: [10:1.00]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttpd2dqx (%rdi), %xmm1 # sched: [13:1.00]
+; BDVER2-NEXT:    vcvttpd2dq %xmm0, %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvttpd2dq:
 ; BTVER2-SSE:       # %bb.0:
@@ -3181,17 +3181,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvttps2dq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvttps2dq %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    cvttps2dq %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-SSE-NEXT:    cvttps2dq (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvttps2dq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttps2dq %xmm0, %xmm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    vcvttps2dq (%rdi), %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttps2dq %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvttps2dq:
 ; BTVER2-SSE:       # %bb.0:
@@ -3321,17 +3321,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvttsd2si:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvttsd2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvttsd2si (%rdi), %eax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvttsd2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvttsd2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvttsd2si:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttsd2si %xmm0, %ecx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvttsd2si (%rdi), %eax # sched: [10:1.00]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttsd2si (%rdi), %eax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvttsd2si %xmm0, %ecx # sched: [13:1.00]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvttsd2si:
 ; BTVER2-SSE:       # %bb.0:
@@ -3461,17 +3461,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_cvttsd2siq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    cvttsd2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    cvttsd2si (%rdi), %rax # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    cvttsd2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-SSE-NEXT:    cvttsd2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_cvttsd2siq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vcvttsd2si %xmm0, %rcx # sched: [5:1.00]
-; BDVER2-NEXT:    vcvttsd2si (%rdi), %rax # sched: [10:1.00]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vcvttsd2si (%rdi), %rax # sched: [18:1.00]
+; BDVER2-NEXT:    vcvttsd2si %xmm0, %rcx # sched: [13:1.00]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_cvttsd2siq:
 ; BTVER2-SSE:       # %bb.0:
@@ -3588,15 +3588,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_divpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    divpd %xmm1, %xmm0 # sched: [22:22.00]
-; BDVER2-SSE-NEXT:    divpd (%rdi), %xmm0 # sched: [28:22.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    divpd %xmm1, %xmm0 # sched: [9:9.50]
+; BDVER2-SSE-NEXT:    divpd (%rdi), %xmm0 # sched: [14:9.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_divpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdivpd %xmm1, %xmm0, %xmm0 # sched: [22:22.00]
-; BDVER2-NEXT:    vdivpd (%rdi), %xmm0, %xmm0 # sched: [28:22.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdivpd %xmm1, %xmm0, %xmm0 # sched: [9:9.50]
+; BDVER2-NEXT:    vdivpd (%rdi), %xmm0, %xmm0 # sched: [14:9.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_divpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -3708,15 +3708,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_divsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    divsd %xmm1, %xmm0 # sched: [22:22.00]
-; BDVER2-SSE-NEXT:    divsd (%rdi), %xmm0 # sched: [28:22.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    divsd %xmm1, %xmm0 # sched: [9:9.50]
+; BDVER2-SSE-NEXT:    divsd (%rdi), %xmm0 # sched: [14:9.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_divsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdivsd %xmm1, %xmm0, %xmm0 # sched: [22:22.00]
-; BDVER2-NEXT:    vdivsd (%rdi), %xmm0, %xmm0 # sched: [28:22.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdivsd %xmm1, %xmm0, %xmm0 # sched: [9:9.50]
+; BDVER2-NEXT:    vdivsd (%rdi), %xmm0, %xmm0 # sched: [14:9.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_divsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -3821,13 +3821,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_lfence:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    lfence # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    lfence # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_lfence:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    lfence # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    lfence # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_lfence:
 ; BTVER2-SSE:       # %bb.0:
@@ -3927,13 +3927,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_mfence:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    mfence # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    mfence # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mfence:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    mfence # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    mfence # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mfence:
 ; BTVER2-SSE:       # %bb.0:
@@ -4032,12 +4032,12 @@
 ; BDVER2-SSE-LABEL: test_maskmovdqu:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    maskmovdqu %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_maskmovdqu:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmaskmovdqu %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_maskmovdqu:
 ; BTVER2-SSE:       # %bb.0:
@@ -4144,15 +4144,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_maxpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    maxpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    maxpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    maxpd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    maxpd (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_maxpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaxpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmaxpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaxpd %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vmaxpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_maxpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4265,15 +4265,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_maxsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    maxsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    maxsd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    maxsd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    maxsd (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_maxsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmaxsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmaxsd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmaxsd %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vmaxsd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_maxsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4386,15 +4386,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_minpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    minpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    minpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    minpd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    minpd (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_minpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vminpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vminpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vminpd %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vminpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_minpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4507,15 +4507,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_minsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    minsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    minsd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    minsd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    minsd (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_minsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vminsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vminsd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vminsd %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vminsd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_minsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4641,17 +4641,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movapd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movapd (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addpd %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    movapd (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-SSE-NEXT:    movapd %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movapd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovapd (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovapd (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-NEXT:    vmovapd %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movapd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4780,17 +4780,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movdqa:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movdqa (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movdqa (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    movdqa %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movdqa:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovdqa (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    vpaddq %xmm0, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vmovdqa (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vpaddq %xmm0, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vmovdqa %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movdqa:
 ; BTVER2-SSE:       # %bb.0:
@@ -4919,17 +4919,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movdqu:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movdqu (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movdqu (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    movdqu %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movdqu:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovdqu (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    vpaddq %xmm0, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vmovdqu (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vpaddq %xmm0, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vmovdqu %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movdqu:
 ; BTVER2-SSE:       # %bb.0:
@@ -5097,23 +5097,23 @@
 ;
 ; BDVER2-SSE-LABEL: test_movd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movd %edi, %xmm1 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm0, %xmm1 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm0, %xmm2 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    movd %xmm2, %eax # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    movd %xmm1, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-SSE-NEXT:    movd %edi, %xmm1 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    paddd %xmm0, %xmm2 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movd %xmm2, %eax # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm0, %xmm1 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movd %xmm1, (%rsi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovd %edi, %xmm1 # sched: [1:1.00]
-; BDVER2-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
-; BDVER2-NEXT:    vpaddd %xmm2, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    vmovd %xmm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    vmovd %xmm1, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovd %edi, %xmm1 # sched: [10:0.50]
+; BDVER2-NEXT:    vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [5:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddd %xmm2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vmovd %xmm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    vmovd %xmm1, (%rsi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movd:
 ; BTVER2-SSE:       # %bb.0:
@@ -5298,23 +5298,23 @@
 ;
 ; BDVER2-SSE-LABEL: test_movd_64:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movq %rdi, %xmm1 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero sched: [6:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm1 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm2 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    movq %xmm2, %rax # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    movq %xmm1, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero sched: [5:0.50]
+; BDVER2-SSE-NEXT:    movq %rdi, %xmm1 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm2 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movq %xmm2, %rax # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm1 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movq %xmm1, (%rsi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movd_64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovq %rdi, %xmm1 # sched: [1:1.00]
-; BDVER2-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero sched: [6:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
-; BDVER2-NEXT:    vpaddq %xmm2, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    vmovq %xmm0, %rax # sched: [2:1.00]
-; BDVER2-NEXT:    vmovq %xmm1, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovq %rdi, %xmm1 # sched: [10:0.50]
+; BDVER2-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero sched: [5:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vmovq %xmm0, %rax # sched: [10:1.00]
+; BDVER2-NEXT:    vmovq %xmm1, (%rsi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movd_64:
 ; BTVER2-SSE:       # %bb.0:
@@ -5474,19 +5474,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_movhpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movhpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movapd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movhpd %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    movapd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movhpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovhpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    vmovapd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovhpd %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    vmovapd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movhpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -5636,19 +5636,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_movlpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movlpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movapd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movlpd %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    movapd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movlpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovlpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    vmovapd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovlpd %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    vmovapd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movlpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -5760,13 +5760,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_movmskpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movmskpd %xmm0, %eax # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movmskpd %xmm0, %eax # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movmskpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovmskpd %xmm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovmskpd %xmm0, %eax # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movmskpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -5875,15 +5875,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_movntdqa:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    movntdq %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movntdq %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movntdqa:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddq %xmm0, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    vmovntdq %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm0, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vmovntdq %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movntdqa:
 ; BTVER2-SSE:       # %bb.0:
@@ -5994,15 +5994,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_movntpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addpd %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movntpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movntpd %xmm0, (%rdi) # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movntpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovntpd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovntpd %xmm0, (%rdi) # sched: [3:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movntpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -6126,17 +6126,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movq_mem:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero sched: [6:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    movq %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero sched: [5:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movq %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movq_mem:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero sched: [6:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    vmovq %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero sched: [5:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vmovq %xmm0, (%rdi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movq_mem:
 ; BTVER2-SSE:       # %bb.0:
@@ -6256,15 +6256,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_movq_reg:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero sched: [1:0.33]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movq {{.*#+}} xmm0 = xmm0[0],zero sched: [2:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movq_reg:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero sched: [1:0.33]
-; BDVER2-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movq_reg:
 ; BTVER2-SSE:       # %bb.0:
@@ -6388,17 +6388,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movsd_mem:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addsd %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    movsd %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero sched: [5:0.50]
+; BDVER2-SSE-NEXT:    addsd %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    movsd %xmm0, (%rsi) # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movsd_mem:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero sched: [6:0.50]
-; BDVER2-NEXT:    vaddsd %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vmovsd %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero sched: [5:0.50]
+; BDVER2-NEXT:    vaddsd %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vmovsd %xmm0, (%rsi) # sched: [2:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movsd_mem:
 ; BTVER2-SSE:       # %bb.0:
@@ -6513,14 +6513,14 @@
 ;
 ; BDVER2-SSE-LABEL: test_movsd_reg:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movaps %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movaps %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movsd_reg:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0] sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0] sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movsd_reg:
 ; BTVER2-SSE:       # %bb.0:
@@ -6641,17 +6641,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movupd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movupd (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addpd %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    movupd (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-SSE-NEXT:    movupd %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movupd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovupd (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
+; BDVER2-NEXT:    vmovupd (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
 ; BDVER2-NEXT:    vmovupd %xmm0, (%rsi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movupd:
 ; BTVER2-SSE:       # %bb.0:
@@ -6768,14 +6768,14 @@
 ; BDVER2-SSE-LABEL: test_mulpd:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    mulpd %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    mulpd (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    mulpd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mulpd:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmulpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulpd (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulpd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mulpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -6888,14 +6888,14 @@
 ; BDVER2-SSE-LABEL: test_mulsd:
 ; BDVER2-SSE:       # %bb.0:
 ; BDVER2-SSE-NEXT:    mulsd %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    mulsd (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    mulsd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mulsd:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmulsd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vmulsd (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmulsd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mulsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -7020,17 +7020,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_orpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    orpd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    orpd (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    orpd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    orpd (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_orpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vorpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vorpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vorpd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vorpd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_orpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -7155,15 +7155,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_packssdw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    packssdw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    packssdw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    packssdw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_packssdw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpackssdw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_packssdw:
 ; BTVER2-SSE:       # %bb.0:
@@ -7281,15 +7281,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_packsswb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    packsswb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    packsswb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    packsswb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_packsswb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpacksswb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_packsswb:
 ; BTVER2-SSE:       # %bb.0:
@@ -7407,15 +7407,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_packuswb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    packuswb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    packuswb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    packuswb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_packuswb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpackuswb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_packuswb:
 ; BTVER2-SSE:       # %bb.0:
@@ -7533,15 +7533,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddb:
 ; BTVER2-SSE:       # %bb.0:
@@ -7657,15 +7657,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddd:
 ; BTVER2-SSE:       # %bb.0:
@@ -7777,15 +7777,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddq (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddq (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddq:
 ; BTVER2-SSE:       # %bb.0:
@@ -7901,15 +7901,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddsb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddsb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddsb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddsb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddsb:
 ; BTVER2-SSE:       # %bb.0:
@@ -8026,15 +8026,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddsw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddsw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddsw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddsw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -8151,15 +8151,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddusb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddusb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddusb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddusb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddusb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddusb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddusb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddusb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddusb:
 ; BTVER2-SSE:       # %bb.0:
@@ -8276,15 +8276,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddusw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddusw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddusw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddusw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddusw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddusw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddusw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddusw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddusw:
 ; BTVER2-SSE:       # %bb.0:
@@ -8401,15 +8401,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_paddw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    paddw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_paddw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpaddw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_paddw:
 ; BTVER2-SSE:       # %bb.0:
@@ -8534,17 +8534,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pand:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pand %xmm1, %xmm0 # sched: [1:0.33]
+; BDVER2-SSE-NEXT:    pand %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pand (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pand:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpand %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
+; BDVER2-NEXT:    vpand %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpand (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pand:
 ; BTVER2-SSE:       # %bb.0:
@@ -8690,19 +8690,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_pandn:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pandn %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm1 # sched: [1:0.33]
+; BDVER2-SSE-NEXT:    pandn %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pandn (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm1 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    movdqa %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm0, %xmm1 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movdqa %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pandn:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpandn %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
+; BDVER2-NEXT:    vpandn %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpandn (%rdi), %xmm0, %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pandn:
 ; BTVER2-SSE:       # %bb.0:
@@ -8829,15 +8829,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pavgb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pavgb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pavgb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pavgb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pavgb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpavgb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpavgb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpavgb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pavgb:
 ; BTVER2-SSE:       # %bb.0:
@@ -8963,15 +8963,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pavgw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pavgw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pavgw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pavgw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pavgw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpavgw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpavgw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpavgw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pavgw:
 ; BTVER2-SSE:       # %bb.0:
@@ -9108,17 +9108,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpeqb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpeqb %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpeqb %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpeqb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpeqb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomeqb %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomeqb %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomeqb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpeqb:
 ; BTVER2-SSE:       # %bb.0:
@@ -9251,17 +9251,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpeqd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpeqd %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpeqd %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpeqd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpeqd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomeqd %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomeqd %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomeqd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpeqd:
 ; BTVER2-SSE:       # %bb.0:
@@ -9394,17 +9394,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpeqw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpeqw %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpeqw %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpeqw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpeqw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomeqw %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomeqw %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomeqw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpeqw:
 ; BTVER2-SSE:       # %bb.0:
@@ -9543,18 +9543,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpgtb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm2 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpgtb %xmm1, %xmm2 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm2 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpgtb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pcmpgtb %xmm1, %xmm2 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpgtb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomgtb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpgtb:
 ; BTVER2-SSE:       # %bb.0:
@@ -9695,18 +9695,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpgtd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm2 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpgtd %xmm1, %xmm2 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm2 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpeqd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pcmpgtd %xmm1, %xmm2 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpgtd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomeqd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpgtd:
 ; BTVER2-SSE:       # %bb.0:
@@ -9847,18 +9847,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpgtw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm2 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpgtw %xmm1, %xmm2 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm2 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpgtw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pcmpgtw %xmm1, %xmm2 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpgtw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomgtw %xmm1, %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomgtw %xmm1, %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomgtw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpor %xmm0, %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpgtw:
 ; BTVER2-SSE:       # %bb.0:
@@ -9978,15 +9978,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pextrw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pextrw $6, %xmm0, %eax # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    pextrw $6, %xmm0, %eax # sched: [13:1.00]
 ; BDVER2-SSE-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pextrw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpextrw $6, %xmm0, %eax # sched: [3:1.00]
+; BDVER2-NEXT:    vpextrw $6, %xmm0, %eax # sched: [13:1.00]
 ; BDVER2-NEXT:    # kill: def $ax killed $ax killed $eax
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pextrw:
 ; BTVER2-SSE:       # %bb.0:
@@ -10100,15 +10100,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pinsrw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pinsrw $1, %edi, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    pinsrw $3, (%rsi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pinsrw $1, %edi, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    pinsrw $3, (%rsi), %xmm0 # sched: [6:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pinsrw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpinsrw $1, %edi, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpinsrw $3, (%rsi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpinsrw $1, %edi, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpinsrw $3, (%rsi), %xmm0, %xmm0 # sched: [6:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pinsrw:
 ; BTVER2-SSE:       # %bb.0:
@@ -10220,15 +10220,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaddwd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaddwd %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmaddwd (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmaddwd %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmaddwd (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaddwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaddwd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmaddwd (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmaddwd %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmaddwd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaddwd:
 ; BTVER2-SSE:       # %bb.0:
@@ -10346,15 +10346,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaxsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaxsw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmaxsw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmaxsw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaxsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpmaxsw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaxsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -10471,15 +10471,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaxub:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaxub %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmaxub %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmaxub (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaxub:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpmaxub (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaxub:
 ; BTVER2-SSE:       # %bb.0:
@@ -10596,15 +10596,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pminsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pminsw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pminsw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pminsw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pminsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpminsw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pminsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -10721,15 +10721,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pminub:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pminub %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pminub %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pminub (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pminub:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpminub %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpminub %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpminub (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pminub:
 ; BTVER2-SSE:       # %bb.0:
@@ -10831,13 +10831,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovmskb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovmskb %xmm0, %eax # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmovmskb %xmm0, %eax # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovmskb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovmskb %xmm0, %eax # sched: [2:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovmskb %xmm0, %eax # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovmskb:
 ; BTVER2-SSE:       # %bb.0:
@@ -10944,15 +10944,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmulhuw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmulhuw %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmulhuw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmulhuw %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmulhuw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmulhuw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmulhuw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmulhuw (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmulhuw %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmulhuw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmulhuw:
 ; BTVER2-SSE:       # %bb.0:
@@ -11065,15 +11065,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmulhw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmulhw %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmulhw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmulhw %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmulhw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmulhw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmulhw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmulhw (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmulhw %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmulhw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmulhw:
 ; BTVER2-SSE:       # %bb.0:
@@ -11186,15 +11186,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmullw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmullw %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmullw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmullw %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmullw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmullw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmullw (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmullw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmullw:
 ; BTVER2-SSE:       # %bb.0:
@@ -11306,15 +11306,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmuludq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmuludq %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmuludq (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmuludq %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmuludq (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmuludq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmuludq (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmuludq (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmuludq:
 ; BTVER2-SSE:       # %bb.0:
@@ -11441,17 +11441,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_por:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    por (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_por:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
+; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpor (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_por:
 ; BTVER2-SSE:       # %bb.0:
@@ -11568,15 +11568,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psadbw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psadbw %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    psadbw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psadbw %xmm1, %xmm0 # sched: [4:0.50]
+; BDVER2-SSE-NEXT:    psadbw (%rdi), %xmm0 # sched: [9:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psadbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpsadbw (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
+; BDVER2-NEXT:    vpsadbw (%rdi), %xmm0, %xmm0 # sched: [9:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psadbw:
 ; BTVER2-SSE:       # %bb.0:
@@ -11706,17 +11706,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pshufd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pshufd {{.*#+}} xmm0 = mem[3,2,1,0] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pshufd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,0,3,2] sched: [1:0.50]
 ; BDVER2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[3,2,1,0] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,0,3,2] sched: [2:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pshufd:
 ; BTVER2-SSE:       # %bb.0:
@@ -11849,17 +11849,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pshufhw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,5,4,7,6] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,5,4,7,6] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pshufhw {{.*#+}} xmm0 = mem[0,1,2,3,7,6,5,4] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pshufhw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,7,6] sched: [1:0.50]
 ; BDVER2-NEXT:    vpshufhw {{.*#+}} xmm1 = mem[0,1,2,3,7,6,5,4] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,7,6] sched: [2:0.50]
+; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pshufhw:
 ; BTVER2-SSE:       # %bb.0:
@@ -11992,17 +11992,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pshuflw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,0,3,2,4,5,6,7] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,0,3,2,4,5,6,7] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pshuflw {{.*#+}} xmm0 = mem[3,2,1,0,4,5,6,7] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pshuflw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7] sched: [1:0.50]
 ; BDVER2-NEXT:    vpshuflw {{.*#+}} xmm1 = mem[3,2,1,0,4,5,6,7] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7] sched: [2:0.50]
+; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pshuflw:
 ; BTVER2-SSE:       # %bb.0:
@@ -12132,17 +12132,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pslld:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pslld %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    pslld (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    pslld $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pslld %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    pslld (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    pslld $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pslld:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpslld %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpslld (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpslld $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpslld %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpslld (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpslld $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pslld:
 ; BTVER2-SSE:       # %bb.0:
@@ -12254,13 +12254,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_pslldq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11] sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pslldq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11] sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11] sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pslldq:
 ; BTVER2-SSE:       # %bb.0:
@@ -12379,17 +12379,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psllq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psllq %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psllq (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psllq $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psllq %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psllq (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psllq $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psllq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsllq %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsllq (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsllq $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsllq %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsllq (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsllq $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psllq:
 ; BTVER2-SSE:       # %bb.0:
@@ -12521,17 +12521,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psllw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psllw %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psllw (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psllw $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psllw %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psllw (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psllw $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psllw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsllw %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsllw (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsllw $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsllw %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsllw (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsllw $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psllw:
 ; BTVER2-SSE:       # %bb.0:
@@ -12663,17 +12663,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psrad:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psrad %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psrad (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psrad $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psrad %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psrad (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psrad $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psrad:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsrad %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsrad (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsrad $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsrad %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsrad (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsrad $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psrad:
 ; BTVER2-SSE:       # %bb.0:
@@ -12805,17 +12805,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psraw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psraw %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psraw (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psraw $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psraw %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psraw (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psraw $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psraw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsraw %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsraw (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsraw $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsraw %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsraw (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsraw $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psraw:
 ; BTVER2-SSE:       # %bb.0:
@@ -12947,17 +12947,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psrld:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psrld %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psrld (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psrld $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psrld %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psrld (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psrld $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psrld:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsrld %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsrld (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsrld $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsrld %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsrld (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsrld $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psrld:
 ; BTVER2-SSE:       # %bb.0:
@@ -13069,13 +13069,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_psrldq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psrldq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psrldq:
 ; BTVER2-SSE:       # %bb.0:
@@ -13194,17 +13194,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psrlq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psrlq %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psrlq (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psrlq $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psrlq %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psrlq (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psrlq $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psrlq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsrlq (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsrlq $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsrlq (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsrlq $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psrlq:
 ; BTVER2-SSE:       # %bb.0:
@@ -13336,17 +13336,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_psrlw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psrlw %xmm1, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    psrlw (%rdi), %xmm0 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    psrlw $2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    psrlw %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-SSE-NEXT:    psrlw (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER2-SSE-NEXT:    psrlw $2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psrlw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpsrlw (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    vpsrlw $2, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    vpsrlw (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER2-NEXT:    vpsrlw $2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psrlw:
 ; BTVER2-SSE:       # %bb.0:
@@ -13469,15 +13469,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubb:
 ; BTVER2-SSE:       # %bb.0:
@@ -13593,15 +13593,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubd %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubd:
 ; BTVER2-SSE:       # %bb.0:
@@ -13713,15 +13713,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubq %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubq %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubq (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubq (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubq:
 ; BTVER2-SSE:       # %bb.0:
@@ -13837,15 +13837,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubsb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubsb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubsb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubsb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubsb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubsb:
 ; BTVER2-SSE:       # %bb.0:
@@ -13962,15 +13962,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubsw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubsw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubsw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubsw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -14087,15 +14087,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubusb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubusb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubusb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubusb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubusb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubusb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubusb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubusb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubusb:
 ; BTVER2-SSE:       # %bb.0:
@@ -14212,15 +14212,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubusw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubusw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubusw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubusw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubusw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubusw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubusw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubusw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubusw:
 ; BTVER2-SSE:       # %bb.0:
@@ -14337,15 +14337,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psubw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psubw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psubw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psubw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psubw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsubw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsubw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsubw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psubw:
 ; BTVER2-SSE:       # %bb.0:
@@ -14461,15 +14461,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpckhbw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],mem[8],xmm0[9],mem[9],xmm0[10],mem[10],xmm0[11],mem[11],xmm0[12],mem[12],xmm0[13],mem[13],xmm0[14],mem[14],xmm0[15],mem[15] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpckhbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8],mem[8],xmm0[9],mem[9],xmm0[10],mem[10],xmm0[11],mem[11],xmm0[12],mem[12],xmm0[13],mem[13],xmm0[14],mem[14],xmm0[15],mem[15] sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpckhbw:
 ; BTVER2-SSE:       # %bb.0:
@@ -14596,17 +14596,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpckhdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpckhdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpckhdq:
 ; BTVER2-SSE:       # %bb.0:
@@ -14736,17 +14736,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpckhqdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpckhqdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpckhqdq {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpckhqdq:
 ; BTVER2-SSE:       # %bb.0:
@@ -14867,15 +14867,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpckhwd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpckhwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpckhwd:
 ; BTVER2-SSE:       # %bb.0:
@@ -14991,15 +14991,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpcklbw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3],xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpcklbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3],xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpcklbw:
 ; BTVER2-SSE:       # %bb.0:
@@ -15126,17 +15126,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpckldq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpckldq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpckldq:
 ; BTVER2-SSE:       # %bb.0:
@@ -15266,17 +15266,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpcklqdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpcklqdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpcklqdq:
 ; BTVER2-SSE:       # %bb.0:
@@ -15397,15 +15397,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_punpcklwd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_punpcklwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:0.50]
+; BDVER2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [2:0.50]
 ; BDVER2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3] sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_punpcklwd:
 ; BTVER2-SSE:       # %bb.0:
@@ -15530,17 +15530,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pxor:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pxor %xmm1, %xmm0 # sched: [1:0.33]
+; BDVER2-SSE-NEXT:    pxor %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pxor (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pxor:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpxor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
+; BDVER2-NEXT:    vpxor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpxor (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pxor:
 ; BTVER2-SSE:       # %bb.0:
@@ -15670,17 +15670,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_shufpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],mem[0] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],mem[0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_shufpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vshufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0] sched: [1:1.00]
-; BDVER2-NEXT:    vshufpd {{.*#+}} xmm1 = xmm1[1],mem[0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vshufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0] sched: [2:0.50]
+; BDVER2-NEXT:    vshufpd {{.*#+}} xmm1 = xmm1[1],mem[0] sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_shufpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -15811,17 +15811,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_sqrtpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    sqrtpd %xmm0, %xmm1 # sched: [21:21.00]
-; BDVER2-SSE-NEXT:    sqrtpd (%rdi), %xmm0 # sched: [27:21.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    sqrtpd %xmm0, %xmm1 # sched: [9:13.50]
+; BDVER2-SSE-NEXT:    sqrtpd (%rdi), %xmm0 # sched: [14:13.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_sqrtpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsqrtpd %xmm0, %xmm0 # sched: [21:21.00]
-; BDVER2-NEXT:    vsqrtpd (%rdi), %xmm1 # sched: [27:21.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsqrtpd (%rdi), %xmm1 # sched: [14:13.50]
+; BDVER2-NEXT:    vsqrtpd %xmm0, %xmm0 # sched: [9:13.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_sqrtpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -15967,19 +15967,19 @@
 ;
 ; BDVER2-SSE-LABEL: test_sqrtsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    sqrtsd %xmm0, %xmm0 # sched: [21:21.00]
-; BDVER2-SSE-NEXT:    movapd (%rdi), %xmm1 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    sqrtsd %xmm1, %xmm1 # sched: [21:21.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movapd (%rdi), %xmm1 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    sqrtsd %xmm0, %xmm0 # sched: [9:13.50]
+; BDVER2-SSE-NEXT:    sqrtsd %xmm1, %xmm1 # sched: [9:13.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_sqrtsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsqrtsd %xmm0, %xmm0, %xmm0 # sched: [21:21.00]
-; BDVER2-NEXT:    vmovapd (%rdi), %xmm1 # sched: [6:0.50]
-; BDVER2-NEXT:    vsqrtsd %xmm1, %xmm1, %xmm1 # sched: [21:21.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovapd (%rdi), %xmm1 # sched: [5:0.50]
+; BDVER2-NEXT:    vsqrtsd %xmm0, %xmm0, %xmm0 # sched: [9:13.50]
+; BDVER2-NEXT:    vsqrtsd %xmm1, %xmm1, %xmm1 # sched: [9:13.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_sqrtsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -16101,15 +16101,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_subpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    subpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    subpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    subpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    subpd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_subpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vsubpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsubpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vsubpd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_subpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -16221,15 +16221,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_subsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    subsd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    subsd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    subsd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    subsd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_subsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vsubsd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vsubsd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vsubsd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vsubsd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_subsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -16445,31 +16445,31 @@
 ;
 ; BDVER2-SSE-LABEL: test_ucomisd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    ucomisd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    ucomisd %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    ucomisd (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    ucomisd (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-SSE-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_ucomisd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vucomisd %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vucomisd %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %cl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-NEXT:    vucomisd (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-NEXT:    vucomisd (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-NEXT:    setnp %al # sched: [1:0.50]
 ; BDVER2-NEXT:    sete %dl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.33]
-; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andb %al, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    orb %cl, %dl # sched: [1:0.50]
+; BDVER2-NEXT:    movzbl %dl, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_ucomisd:
 ; BTVER2-SSE:       # %bb.0:
@@ -16628,17 +16628,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_unpckhpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_unpckhpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [1:1.00]
-; BDVER2-NEXT:    vunpckhpd {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [2:0.50]
+; BDVER2-NEXT:    vunpckhpd {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_unpckhpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -16776,18 +16776,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_unpcklpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movapd %xmm0, %xmm2 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    unpcklpd {{.*#+}} xmm2 = xmm2[0],xmm1[0] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],mem[0] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm2, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movapd %xmm0, %xmm2 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],mem[0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    unpcklpd {{.*#+}} xmm2 = xmm2[0],xmm1[0] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm2, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_unpcklpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm0[0],xmm1[0] sched: [1:1.00]
-; BDVER2-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],mem[0] sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm0[0],xmm1[0] sched: [2:0.50]
+; BDVER2-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],mem[0] sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_unpcklpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -16919,17 +16919,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_xorpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    xorpd %xmm1, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    xorpd (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    xorpd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    xorpd (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_xorpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vxorpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER2-NEXT:    vxorpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vxorpd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vxorpd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_xorpd:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/sse3-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/sse3-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/sse3-schedule.ll
@@ -14,9 +14,9 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2  | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2  | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse3 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=+sse3 -mattr=-avx2  | FileCheck %s --check-prefixes=CHECK,BDVER2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=+sse3 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2  | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2  | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-avx2  | FileCheck %s --check-prefixes=CHECK,ZNVER1
@@ -102,15 +102,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_addsubpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addsubpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    addsubpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addsubpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    addsubpd (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_addsubpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_addsubpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -223,15 +223,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_addsubps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    addsubps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    addsubps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addsubps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    addsubps (%rdi), %xmm0 # sched: [10:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_addsubps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    vaddsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddsubps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    vaddsubps (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_addsubps:
 ; BTVER2-SSE:       # %bb.0:
@@ -344,15 +344,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_haddpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    haddpd %xmm1, %xmm0 # sched: [5:2.00]
-; BDVER2-SSE-NEXT:    haddpd (%rdi), %xmm0 # sched: [11:2.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    haddpd %xmm1, %xmm0 # sched: [11:1.00]
+; BDVER2-SSE-NEXT:    haddpd (%rdi), %xmm0 # sched: [16:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_haddpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhaddpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhaddpd %xmm1, %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER2-NEXT:    vhaddpd (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_haddpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -465,15 +465,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_haddps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    haddps %xmm1, %xmm0 # sched: [5:2.00]
-; BDVER2-SSE-NEXT:    haddps (%rdi), %xmm0 # sched: [11:2.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    haddps %xmm1, %xmm0 # sched: [11:1.00]
+; BDVER2-SSE-NEXT:    haddps (%rdi), %xmm0 # sched: [16:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_haddps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhaddps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhaddps %xmm1, %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER2-NEXT:    vhaddps (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_haddps:
 ; BTVER2-SSE:       # %bb.0:
@@ -586,15 +586,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_hsubpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    hsubpd %xmm1, %xmm0 # sched: [5:2.00]
-; BDVER2-SSE-NEXT:    hsubpd (%rdi), %xmm0 # sched: [11:2.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    hsubpd %xmm1, %xmm0 # sched: [11:1.00]
+; BDVER2-SSE-NEXT:    hsubpd (%rdi), %xmm0 # sched: [16:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_hsubpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhsubpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhsubpd %xmm1, %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER2-NEXT:    vhsubpd (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_hsubpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -707,15 +707,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_hsubps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    hsubps %xmm1, %xmm0 # sched: [5:2.00]
-; BDVER2-SSE-NEXT:    hsubps (%rdi), %xmm0 # sched: [11:2.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    hsubps %xmm1, %xmm0 # sched: [11:1.00]
+; BDVER2-SSE-NEXT:    hsubps (%rdi), %xmm0 # sched: [16:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_hsubps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
-; BDVER2-NEXT:    vhsubps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vhsubps %xmm1, %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER2-NEXT:    vhsubps (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_hsubps:
 ; BTVER2-SSE:       # %bb.0:
@@ -817,13 +817,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_lddqu:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    lddqu (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    lddqu (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_lddqu:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vlddqu (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vlddqu (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_lddqu:
 ; BTVER2-SSE:       # %bb.0:
@@ -943,17 +943,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_monitor:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl %esi, %ecx # sched: [1:0.33]
 ; BDVER2-SSE-NEXT:    leaq (%rdi), %rax # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    monitor # sched: [100:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl %esi, %ecx # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    monitor # sched: [100:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_monitor:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %esi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    leaq (%rdi), %rax # sched: [1:0.50]
-; BDVER2-NEXT:    monitor # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl %esi, %ecx # sched: [1:0.50]
+; BDVER2-NEXT:    monitor # sched: [100:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_monitor:
 ; BTVER2-SSE:       # %bb.0:
@@ -1082,17 +1082,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movddup:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movddup {{.*#+}} xmm0 = mem[0,0] sched: [6:0.50]
-; BDVER2-SSE-NEXT:    subpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movddup {{.*#+}} xmm0 = mem[0,0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    subpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movddup:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
-; BDVER2-NEXT:    vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [6:0.50]
-; BDVER2-NEXT:    vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [7:0.50]
+; BDVER2-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [2:0.50]
+; BDVER2-NEXT:    vsubpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movddup:
 ; BTVER2-SSE:       # %bb.0:
@@ -1223,17 +1223,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movshdup:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movshdup:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
-; BDVER2-NEXT:    vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [6:0.50]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [7:0.50]
+; BDVER2-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [2:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movshdup:
 ; BTVER2-SSE:       # %bb.0:
@@ -1364,17 +1364,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_movsldup:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:0.50]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movsldup:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
-; BDVER2-NEXT:    vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [6:0.50]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [7:0.50]
+; BDVER2-NEXT:    vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [2:0.50]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movsldup:
 ; BTVER2-SSE:       # %bb.0:
@@ -1504,17 +1504,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_mwait:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl %esi, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movl %edi, %ecx # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    mwait # sched: [100:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl %esi, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movl %edi, %ecx # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    mwait # sched: [100:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mwait:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %esi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    movl %edi, %ecx # sched: [1:0.33]
-; BDVER2-NEXT:    mwait # sched: [100:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl %esi, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    movl %edi, %ecx # sched: [1:0.50]
+; BDVER2-NEXT:    mwait # sched: [100:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mwait:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
@@ -13,8 +13,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-sse4.2 | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse4.1 -mattr=-sse4.2 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx,+xop -mattr=+sse4.1 -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-sse4.2 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-sse4.2 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-sse4.2 | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
@@ -107,17 +107,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_blendpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    blendpd {{.*#+}} xmm1 = xmm0[0],xmm1[1] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    blendpd {{.*#+}} xmm1 = xmm0[0],xmm1[1] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    blendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_blendpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendpd {{.*#+}} xmm1 = xmm0[0],xmm1[1] sched: [1:0.50]
+; BDVER2-NEXT:    vblendpd {{.*#+}} xmm1 = xmm0[0],xmm1[1] sched: [2:0.50]
 ; BDVER2-NEXT:    vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [7:0.50]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_blendpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -240,17 +240,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_blendps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    blendps {{.*#+}} xmm1 = xmm1[0],mem[1],xmm1[2,3] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_blendps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:0.50]
+; BDVER2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [2:0.50]
 ; BDVER2-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0],mem[1],xmm1[2,3] sched: [7:0.50]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_blendps:
 ; BTVER2-SSE:       # %bb.0:
@@ -382,18 +382,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_blendvpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movapd %xmm0, %xmm3 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movaps %xmm2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    blendvpd %xmm0, %xmm1, %xmm3 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    blendvpd %xmm0, (%rdi), %xmm3 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    movapd %xmm3, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movapd %xmm0, %xmm3 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movaps %xmm2, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    blendvpd %xmm0, %xmm1, %xmm3 # sched: [2:2.00]
+; BDVER2-SSE-NEXT:    blendvpd %xmm0, (%rdi), %xmm3 # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    movapd %xmm3, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_blendvpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_blendvpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -527,18 +527,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_blendvps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movaps %xmm0, %xmm3 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    movaps %xmm2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    blendvps %xmm0, %xmm1, %xmm3 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    blendvps %xmm0, (%rdi), %xmm3 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    movaps %xmm3, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movaps %xmm0, %xmm3 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movaps %xmm2, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    blendvps %xmm0, %xmm1, %xmm3 # sched: [2:2.00]
+; BDVER2-SSE-NEXT:    blendvps %xmm0, (%rdi), %xmm3 # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    movaps %xmm3, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_blendvps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_blendvps:
 ; BTVER2-SSE:       # %bb.0:
@@ -651,15 +651,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_dppd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    dppd $7, %xmm1, %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    dppd $7, (%rdi), %xmm0 # sched: [15:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    dppd $7, %xmm1, %xmm0 # sched: [15:1.50]
+; BDVER2-SSE-NEXT:    dppd $7, (%rdi), %xmm0 # sched: [20:1.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_dppd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [15:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [15:1.50]
+; BDVER2-NEXT:    vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [20:1.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_dppd:
 ; BTVER2-SSE:       # %bb.0:
@@ -766,15 +766,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_dpps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    dpps $7, %xmm1, %xmm0 # sched: [12:2.00]
-; BDVER2-SSE-NEXT:    dpps $7, (%rdi), %xmm0 # sched: [18:2.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    dpps $7, %xmm1, %xmm0 # sched: [25:1.50]
+; BDVER2-SSE-NEXT:    dpps $7, (%rdi), %xmm0 # sched: [30:1.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_dpps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [12:2.00]
-; BDVER2-NEXT:    vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [18:2.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [25:1.50]
+; BDVER2-NEXT:    vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [30:1.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_dpps:
 ; BTVER2-SSE:       # %bb.0:
@@ -881,15 +881,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_extractps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    extractps $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    extractps $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    extractps $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    extractps $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_extractps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vextractps $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    vextractps $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vextractps $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-NEXT:    vextractps $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_extractps:
 ; BTVER2-SSE:       # %bb.0:
@@ -997,15 +997,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_insertps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    insertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:1.00]
-; BDVER2-SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [7:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    insertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [2:0.50]
+; BDVER2-SSE-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [7:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_insertps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:1.00]
-; BDVER2-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [7:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vinsertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [2:0.50]
+; BDVER2-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [7:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_insertps:
 ; BTVER2-SSE:       # %bb.0:
@@ -1100,13 +1100,13 @@
 ;
 ; BDVER2-SSE-LABEL: test_movntdqa:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movntdqa (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movntdqa (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_movntdqa:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmovntdqa (%rdi), %xmm0 # sched: [6:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmovntdqa (%rdi), %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_movntdqa:
 ; BTVER2-SSE:       # %bb.0:
@@ -1207,15 +1207,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_mpsadbw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    mpsadbw $7, %xmm1, %xmm0 # sched: [7:1.00]
-; BDVER2-SSE-NEXT:    mpsadbw $7, (%rdi), %xmm0 # sched: [13:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    mpsadbw $7, %xmm1, %xmm0 # sched: [9:2.00]
+; BDVER2-SSE-NEXT:    mpsadbw $7, (%rdi), %xmm0 # sched: [14:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_mpsadbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER2-NEXT:    vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [9:2.00]
+; BDVER2-NEXT:    vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [14:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_mpsadbw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1323,15 +1323,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_packusdw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    packusdw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    packusdw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    packusdw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_packusdw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpackusdw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_packusdw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1460,18 +1460,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_pblendvb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm3 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movaps %xmm2, %xmm0 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    pblendvb %xmm0, %xmm1, %xmm3 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    pblendvb %xmm0, (%rdi), %xmm3 # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    movdqa %xmm3, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movdqa %xmm0, %xmm3 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    movaps %xmm2, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pblendvb %xmm0, %xmm1, %xmm3 # sched: [2:2.00]
+; BDVER2-SSE-NEXT:    pblendvb %xmm0, (%rdi), %xmm3 # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    movdqa %xmm3, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pblendvb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
+; BDVER2-NEXT:    vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pblendvb:
 ; BTVER2-SSE:       # %bb.0:
@@ -1596,17 +1596,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pblendw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],mem[2,3],xmm1[4,5,6],mem[7] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pblendw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:0.50]
+; BDVER2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [2:0.50]
 ; BDVER2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],mem[2,3],xmm1[4,5,6],mem[7] sched: [7:0.50]
-; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pblendw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1717,15 +1717,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpeqq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpeqq %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpeqq %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pcmpeqq (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpeqq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomeqq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomeqq (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpeqq:
 ; BTVER2-SSE:       # %bb.0:
@@ -1833,15 +1833,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pextrb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pextrb $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    pextrb $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pextrb $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    pextrb $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pextrb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpextrb $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    vpextrb $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpextrb $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-NEXT:    vpextrb $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pextrb:
 ; BTVER2-SSE:       # %bb.0:
@@ -1960,17 +1960,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pextrd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    paddd %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    pextrd $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    pextrd $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    pextrd $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    pextrd $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pextrd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpaddd %xmm0, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    vpextrd $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    vpextrd $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpaddd %xmm0, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpextrd $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-NEXT:    vpextrd $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pextrd:
 ; BTVER2-SSE:       # %bb.0:
@@ -2081,15 +2081,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pextrq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pextrq $1, %xmm0, %rax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    pextrq $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pextrq $1, %xmm0, %rax # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    pextrq $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pextrq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpextrq $1, %xmm0, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    vpextrq $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpextrq $1, %xmm0, %rax # sched: [13:1.00]
+; BDVER2-NEXT:    vpextrq $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pextrq:
 ; BTVER2-SSE:       # %bb.0:
@@ -2195,15 +2195,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pextrw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pextrw $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    pextrw $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pextrw $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    pextrw $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pextrw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpextrw $3, %xmm0, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    vpextrw $1, %xmm0, (%rdi) # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpextrw $3, %xmm0, %eax # sched: [13:1.00]
+; BDVER2-NEXT:    vpextrw $1, %xmm0, (%rdi) # sched: [13:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pextrw:
 ; BTVER2-SSE:       # %bb.0:
@@ -2310,15 +2310,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phminposuw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phminposuw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    phminposuw %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phminposuw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    phminposuw %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phminposuw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphminposuw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    vphminposuw %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphminposuw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    vphminposuw %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phminposuw:
 ; BTVER2-SSE:       # %bb.0:
@@ -2425,15 +2425,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pinsrb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pinsrb $1, %edi, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    pinsrb $3, (%rsi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pinsrb $1, %edi, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    pinsrb $3, (%rsi), %xmm0 # sched: [6:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pinsrb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [6:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pinsrb:
 ; BTVER2-SSE:       # %bb.0:
@@ -2539,15 +2539,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pinsrd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pinsrd $1, %edi, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    pinsrd $3, (%rsi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pinsrd $1, %edi, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    pinsrd $3, (%rsi), %xmm0 # sched: [6:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pinsrd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [6:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pinsrd:
 ; BTVER2-SSE:       # %bb.0:
@@ -2665,17 +2665,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pinsrq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pinsrq $1, %rdi, %xmm0 # sched: [2:1.00]
-; BDVER2-SSE-NEXT:    pinsrq $1, (%rsi), %xmm1 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pinsrq $1, (%rsi), %xmm1 # sched: [6:0.50]
+; BDVER2-SSE-NEXT:    pinsrq $1, %rdi, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pinsrq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [2:1.00]
-; BDVER2-NEXT:    vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [6:0.50]
+; BDVER2-NEXT:    vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pinsrq:
 ; BTVER2-SSE:       # %bb.0:
@@ -2786,15 +2786,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaxsb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaxsb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmaxsb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmaxsb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaxsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaxsb:
 ; BTVER2-SSE:       # %bb.0:
@@ -2901,15 +2901,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaxsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaxsd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmaxsd %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmaxsd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaxsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaxsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -3016,15 +3016,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaxud:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaxud %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmaxud %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmaxud (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaxud:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpmaxud (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaxud:
 ; BTVER2-SSE:       # %bb.0:
@@ -3131,15 +3131,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaxuw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaxuw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmaxuw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmaxuw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaxuw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaxuw:
 ; BTVER2-SSE:       # %bb.0:
@@ -3246,15 +3246,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pminsb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pminsb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pminsb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pminsb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pminsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpminsb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pminsb:
 ; BTVER2-SSE:       # %bb.0:
@@ -3361,15 +3361,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pminsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pminsd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pminsd %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pminsd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pminsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpminsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpminsd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpminsd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pminsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -3476,15 +3476,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pminud:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pminud %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pminud %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pminud (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pminud:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpminud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpminud %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpminud (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pminud:
 ; BTVER2-SSE:       # %bb.0:
@@ -3591,15 +3591,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pminuw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pminuw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pminuw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pminuw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pminuw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpminuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpminuw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpminuw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pminuw:
 ; BTVER2-SSE:       # %bb.0:
@@ -3719,17 +3719,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovsxbw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovsxbw %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovsxbw %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovsxbw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovsxbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovsxbw %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovsxbw (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovsxbw %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovsxbw:
 ; BTVER2-SSE:       # %bb.0:
@@ -3854,17 +3854,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovsxbd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovsxbd %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovsxbd %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovsxbd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovsxbd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovsxbd %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovsxbd (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovsxbd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovsxbd:
 ; BTVER2-SSE:       # %bb.0:
@@ -3989,17 +3989,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovsxbq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovsxbq %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovsxbq %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovsxbq (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovsxbq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovsxbq %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovsxbq (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovsxbq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovsxbq:
 ; BTVER2-SSE:       # %bb.0:
@@ -4124,17 +4124,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovsxdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovsxdq %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovsxdq %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovsxdq (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovsxdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovsxdq %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovsxdq (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovsxdq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovsxdq:
 ; BTVER2-SSE:       # %bb.0:
@@ -4259,17 +4259,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovsxwd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovsxwd %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovsxwd %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovsxwd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovsxwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovsxwd %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovsxwd (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovsxwd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovsxwd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4394,17 +4394,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovsxwq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovsxwq %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovsxwq %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovsxwq (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovsxwq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovsxwq %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovsxwq (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovsxwq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovsxwq:
 ; BTVER2-SSE:       # %bb.0:
@@ -4529,17 +4529,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovzxbw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovzxbw {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddw %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovzxbw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [7:0.50]
-; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovzxbw:
 ; BTVER2-SSE:       # %bb.0:
@@ -4664,17 +4664,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovzxbd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovzxbd {{.*#+}} xmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovzxbd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovzxbd:
 ; BTVER2-SSE:       # %bb.0:
@@ -4799,17 +4799,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovzxbq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovzxbq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovzxbq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovzxbq {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovzxbq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovzxbq:
 ; BTVER2-SSE:       # %bb.0:
@@ -4934,17 +4934,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovzxdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovzxdq {{.*#+}} xmm0 = mem[0],zero,mem[1],zero sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovzxdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovzxdq:
 ; BTVER2-SSE:       # %bb.0:
@@ -5069,17 +5069,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovzxwd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovzxwd {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddd %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovzxwd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [7:0.50]
-; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovzxwd:
 ; BTVER2-SSE:       # %bb.0:
@@ -5204,17 +5204,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmovzxwq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pmovzxwq {{.*#+}} xmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [7:0.50]
-; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    paddq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmovzxwq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:0.50]
 ; BDVER2-NEXT:    vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [7:0.50]
-; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [2:0.50]
+; BDVER2-NEXT:    vpaddq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmovzxwq:
 ; BTVER2-SSE:       # %bb.0:
@@ -5338,17 +5338,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmuldq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmuldq %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmuldq (%rdi), %xmm2 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmuldq (%rdi), %xmm2 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    pmuldq %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    por %xmm2, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmuldq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmuldq (%rdi), %xmm2, %xmm1 # sched: [11:1.00]
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmuldq (%rdi), %xmm2, %xmm2 # sched: [9:1.00]
+; BDVER2-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpor %xmm2, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmuldq:
 ; BTVER2-SSE:       # %bb.0:
@@ -5460,15 +5460,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmulld:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmulld %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmulld (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmulld %xmm1, %xmm0 # sched: [5:2.00]
+; BDVER2-SSE-NEXT:    pmulld (%rdi), %xmm0 # sched: [10:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmulld:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmulld %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmulld (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmulld %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
+; BDVER2-NEXT:    vpmulld (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmulld:
 ; BTVER2-SSE:       # %bb.0:
@@ -5622,23 +5622,23 @@
 ;
 ; BDVER2-SSE-LABEL: test_ptest:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    ptest %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-SSE-NEXT:    ptest %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-SSE-NEXT:    setb %al # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    ptest (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-SSE-NEXT:    ptest (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-SSE-NEXT:    setb %cl # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movzbl %cl, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movzbl %cl, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_ptest:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vptest %xmm1, %xmm0 # sched: [2:1.00]
+; BDVER2-NEXT:    vptest %xmm1, %xmm0 # sched: [1:1.00]
 ; BDVER2-NEXT:    setb %al # sched: [1:0.50]
-; BDVER2-NEXT:    vptest (%rdi), %xmm0 # sched: [8:1.00]
+; BDVER2-NEXT:    vptest (%rdi), %xmm0 # sched: [6:1.00]
 ; BDVER2-NEXT:    setb %cl # sched: [1:0.50]
-; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.33]
-; BDVER2-NEXT:    movzbl %cl, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    andb %al, %cl # sched: [1:0.50]
+; BDVER2-NEXT:    movzbl %cl, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_ptest:
 ; BTVER2-SSE:       # %bb.0:
@@ -5776,17 +5776,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_roundpd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    roundpd $7, %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    roundpd $7, %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-SSE-NEXT:    roundpd $7, (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_roundpd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vroundpd $7, %xmm0, %xmm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    vroundpd $7, (%rdi), %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vroundpd $7, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_roundpd:
 ; BTVER2-SSE:       # %bb.0:
@@ -5912,17 +5912,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_roundps:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    roundps $7, %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    roundps $7, %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-SSE-NEXT:    roundps $7, (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_roundps:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vroundps $7, %xmm0, %xmm0 # sched: [3:1.00]
 ; BDVER2-NEXT:    vroundps $7, (%rdi), %xmm1 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vroundps $7, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_roundps:
 ; BTVER2-SSE:       # %bb.0:
@@ -6053,18 +6053,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_roundsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movapd %xmm0, %xmm2 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    roundsd $7, %xmm1, %xmm2 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    movapd %xmm0, %xmm2 # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    roundsd $7, (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addpd %xmm2, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    roundsd $7, %xmm1, %xmm2 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    addpd %xmm2, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_roundsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-NEXT:    vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-NEXT:    vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_roundsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -6197,18 +6197,18 @@
 ;
 ; BDVER2-SSE-LABEL: test_roundss:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movaps %xmm0, %xmm2 # sched: [1:1.00]
-; BDVER2-SSE-NEXT:    roundss $7, %xmm1, %xmm2 # sched: [3:1.00]
+; BDVER2-SSE-NEXT:    movaps %xmm0, %xmm2 # sched: [1:0.50]
 ; BDVER2-SSE-NEXT:    roundss $7, (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER2-SSE-NEXT:    addps %xmm2, %xmm0 # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    roundss $7, %xmm1, %xmm2 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    addps %xmm2, %xmm0 # sched: [5:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_roundss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
+; BDVER2-NEXT:    vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [4:1.00]
 ; BDVER2-NEXT:    vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
-; BDVER2-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_roundss:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/sse42-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/sse42-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/sse42-schedule.ll
@@ -13,8 +13,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx  | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse4.2,+pclmul -mattr=-avx  | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx,+xop -mattr=+sse4.2,+pclmul -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx  | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx  | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-avx  | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
@@ -107,17 +107,17 @@
 ;
 ; BDVER2-SSE-LABEL: crc32_32_8:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl %edi, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    crc32b %sil, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    crc32b (%rdx), %eax # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl %edi, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    crc32b %sil, %eax # sched: [3:2.00]
+; BDVER2-SSE-NEXT:    crc32b (%rdx), %eax # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: crc32_32_8:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    crc32b %sil, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    crc32b (%rdx), %eax # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    crc32b %sil, %eax # sched: [3:2.00]
+; BDVER2-NEXT:    crc32b (%rdx), %eax # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: crc32_32_8:
 ; BTVER2-SSE:       # %bb.0:
@@ -240,17 +240,17 @@
 ;
 ; BDVER2-SSE-LABEL: crc32_32_16:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl %edi, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    crc32w %si, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    crc32w (%rdx), %eax # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl %edi, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    crc32w %si, %eax # sched: [5:2.00]
+; BDVER2-SSE-NEXT:    crc32w (%rdx), %eax # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: crc32_32_16:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    crc32w %si, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    crc32w (%rdx), %eax # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    crc32w %si, %eax # sched: [5:2.00]
+; BDVER2-NEXT:    crc32w (%rdx), %eax # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: crc32_32_16:
 ; BTVER2-SSE:       # %bb.0:
@@ -373,17 +373,17 @@
 ;
 ; BDVER2-SSE-LABEL: crc32_32_32:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl %edi, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    crc32l %esi, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    crc32l (%rdx), %eax # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl %edi, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    crc32l %esi, %eax # sched: [6:2.00]
+; BDVER2-SSE-NEXT:    crc32l (%rdx), %eax # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: crc32_32_32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    crc32l %esi, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    crc32l (%rdx), %eax # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl %edi, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    crc32l %esi, %eax # sched: [6:2.00]
+; BDVER2-NEXT:    crc32l (%rdx), %eax # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: crc32_32_32:
 ; BTVER2-SSE:       # %bb.0:
@@ -506,17 +506,17 @@
 ;
 ; BDVER2-SSE-LABEL: crc32_64_8:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    crc32b %sil, %eax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    crc32b (%rdx), %eax # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    crc32b %sil, %eax # sched: [3:2.00]
+; BDVER2-SSE-NEXT:    crc32b (%rdx), %eax # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: crc32_64_8:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    crc32b %sil, %eax # sched: [3:1.00]
-; BDVER2-NEXT:    crc32b (%rdx), %eax # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    crc32b %sil, %eax # sched: [3:2.00]
+; BDVER2-NEXT:    crc32b (%rdx), %eax # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: crc32_64_8:
 ; BTVER2-SSE:       # %bb.0:
@@ -639,17 +639,17 @@
 ;
 ; BDVER2-SSE-LABEL: crc32_64_64:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    crc32q %rsi, %rax # sched: [3:1.00]
-; BDVER2-SSE-NEXT:    crc32q (%rdx), %rax # sched: [8:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    crc32q %rsi, %rax # sched: [10:2.00]
+; BDVER2-SSE-NEXT:    crc32q (%rdx), %rax # sched: [7:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: crc32_64_64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movq %rdi, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    crc32q %rsi, %rax # sched: [3:1.00]
-; BDVER2-NEXT:    crc32q (%rdx), %rax # sched: [8:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movq %rdi, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    crc32q %rsi, %rax # sched: [10:2.00]
+; BDVER2-NEXT:    crc32q (%rdx), %rax # sched: [7:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: crc32_64_64:
 ; BTVER2-SSE:       # %bb.0:
@@ -844,29 +844,29 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpestri:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpestri $7, %xmm1, %xmm0 # sched: [4:2.67]
-; BDVER2-SSE-NEXT:    movl %ecx, %esi # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpestri $7, (%rdi), %xmm0 # sched: [4:2.33]
+; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpestri $7, %xmm1, %xmm0 # sched: [15:4.00]
+; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movl %ecx, %esi # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpestri $7, (%rdi), %xmm0 # sched: [20:4.50]
 ; BDVER2-SSE-NEXT:    # kill: def $ecx killed $ecx def $rcx
 ; BDVER2-SSE-NEXT:    leal (%rcx,%rsi), %eax # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpestri:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-NEXT:    vpcmpestri $7, %xmm1, %xmm0 # sched: [4:2.67]
-; BDVER2-NEXT:    movl %ecx, %esi # sched: [1:0.33]
-; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-NEXT:    vpcmpestri $7, (%rdi), %xmm0 # sched: [4:2.33]
+; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-NEXT:    vpcmpestri $7, %xmm1, %xmm0 # sched: [15:4.00]
+; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-NEXT:    movl %ecx, %esi # sched: [1:0.50]
+; BDVER2-NEXT:    vpcmpestri $7, (%rdi), %xmm0 # sched: [20:4.50]
 ; BDVER2-NEXT:    # kill: def $ecx killed $ecx def $rcx
 ; BDVER2-NEXT:    leal (%rcx,%rsi), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpestri:
 ; BTVER2-SSE:       # %bb.0:
@@ -1050,23 +1050,23 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpestrm:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpestrm $7, %xmm1, %xmm0 # sched: [11:2.67]
-; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpestrm $7, (%rdi), %xmm0 # sched: [11:2.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpestrm $7, %xmm1, %xmm0 # sched: [10:4.00]
+; BDVER2-SSE-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpestrm $7, (%rdi), %xmm0 # sched: [15:4.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpestrm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-NEXT:    vpcmpestrm $7, %xmm1, %xmm0 # sched: [11:2.67]
-; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.33]
-; BDVER2-NEXT:    vpcmpestrm $7, (%rdi), %xmm0 # sched: [11:2.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-NEXT:    vpcmpestrm $7, %xmm1, %xmm0 # sched: [10:4.00]
+; BDVER2-NEXT:    movl $7, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    movl $7, %edx # sched: [1:0.50]
+; BDVER2-NEXT:    vpcmpestrm $7, (%rdi), %xmm0 # sched: [15:4.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpestrm:
 ; BTVER2-SSE:       # %bb.0:
@@ -1225,21 +1225,21 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpistri:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpistri $7, %xmm1, %xmm0 # sched: [11:3.00]
-; BDVER2-SSE-NEXT:    movl %ecx, %eax # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    pcmpistri $7, (%rdi), %xmm0 # sched: [17:3.00]
+; BDVER2-SSE-NEXT:    pcmpistri $7, %xmm1, %xmm0 # sched: [14:1.00]
+; BDVER2-SSE-NEXT:    movl %ecx, %eax # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pcmpistri $7, (%rdi), %xmm0 # sched: [19:1.00]
 ; BDVER2-SSE-NEXT:    # kill: def $ecx killed $ecx def $rcx
 ; BDVER2-SSE-NEXT:    leal (%rcx,%rax), %eax # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpistri:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcmpistri $7, %xmm1, %xmm0 # sched: [11:3.00]
-; BDVER2-NEXT:    movl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    vpcmpistri $7, (%rdi), %xmm0 # sched: [17:3.00]
+; BDVER2-NEXT:    vpcmpistri $7, %xmm1, %xmm0 # sched: [14:1.00]
+; BDVER2-NEXT:    movl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    vpcmpistri $7, (%rdi), %xmm0 # sched: [19:1.00]
 ; BDVER2-NEXT:    # kill: def $ecx killed $ecx def $rcx
 ; BDVER2-NEXT:    leal (%rcx,%rax), %eax # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpistri:
 ; BTVER2-SSE:       # %bb.0:
@@ -1359,15 +1359,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpistrm:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpistrm $7, %xmm1, %xmm0 # sched: [11:3.00]
-; BDVER2-SSE-NEXT:    pcmpistrm $7, (%rdi), %xmm0 # sched: [17:3.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pcmpistrm $7, %xmm1, %xmm0 # sched: [6:1.00]
+; BDVER2-SSE-NEXT:    pcmpistrm $7, (%rdi), %xmm0 # sched: [11:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpistrm:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcmpistrm $7, %xmm1, %xmm0 # sched: [11:3.00]
-; BDVER2-NEXT:    vpcmpistrm $7, (%rdi), %xmm0 # sched: [17:3.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpcmpistrm $7, %xmm1, %xmm0 # sched: [6:1.00]
+; BDVER2-NEXT:    vpcmpistrm $7, (%rdi), %xmm0 # sched: [11:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpistrm:
 ; BTVER2-SSE:       # %bb.0:
@@ -1474,15 +1474,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pcmpgtq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pcmpgtq %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pcmpgtq (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pcmpgtq %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    pcmpgtq (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pcmpgtq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpcomgtq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpcomgtq %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpcomgtq (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pcmpgtq:
 ; BTVER2-SSE:       # %bb.0:
@@ -1590,15 +1590,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pclmulqdq:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pclmulqdq $0, %xmm1, %xmm0 # sched: [14:6.00]
-; BDVER2-SSE-NEXT:    pclmulqdq $0, (%rdi), %xmm0 # sched: [14:5.67]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pclmulqdq $0, %xmm1, %xmm0 # sched: [12:1.00]
+; BDVER2-SSE-NEXT:    pclmulqdq $0, (%rdi), %xmm0 # sched: [17:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pclmulqdq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpclmulqdq $0, %xmm1, %xmm0, %xmm0 # sched: [14:6.00]
-; BDVER2-NEXT:    vpclmulqdq $0, (%rdi), %xmm0, %xmm0 # sched: [14:5.67]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpclmulqdq $0, %xmm1, %xmm0, %xmm0 # sched: [13:1.00]
+; BDVER2-NEXT:    vpclmulqdq $0, (%rdi), %xmm0, %xmm0 # sched: [17:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pclmulqdq:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/sse4a-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/sse4a-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/sse4a-schedule.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse4a | FileCheck %s --check-prefix=GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse4a | FileCheck %s --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=ZNVER1
 
@@ -12,8 +12,8 @@
 ;
 ; BDVER2-LABEL: test_extrq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    extrq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    extrq %xmm1, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_extrq:
 ; BTVER2:       # %bb.0:
@@ -37,8 +37,8 @@
 ;
 ; BDVER2-LABEL: test_extrqi:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    extrq $2, $3, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    extrq $2, $3, %xmm0 # sched: [3:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_extrqi:
 ; BTVER2:       # %bb.0:
@@ -62,8 +62,8 @@
 ;
 ; BDVER2-LABEL: test_insertq:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    insertq %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    insertq %xmm1, %xmm0 # sched: [3:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_insertq:
 ; BTVER2:       # %bb.0:
@@ -87,8 +87,8 @@
 ;
 ; BDVER2-LABEL: test_insertqi:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    insertq $6, $5, %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    insertq $6, $5, %xmm1, %xmm0 # sched: [3:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_insertqi:
 ; BTVER2:       # %bb.0:
@@ -112,8 +112,8 @@
 ;
 ; BDVER2-LABEL: test_movntsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movntsd %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movntsd %xmm0, (%rdi) # sched: [3:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movntsd:
 ; BTVER2:       # %bb.0:
@@ -137,8 +137,8 @@
 ;
 ; BDVER2-LABEL: test_movntss:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    movntss %xmm0, (%rdi) # sched: [1:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    movntss %xmm0, (%rdi) # sched: [3:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_movntss:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/ssse3-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/ssse3-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/ssse3-schedule.ll
@@ -14,8 +14,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,SKYLAKE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-sse4.1 | FileCheck %s --check-prefixes=CHECK,SKX-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,SKX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+ssse3 -mattr=-sse4.1 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+avx -mattr=+ssse3 -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-sse4.1 | FileCheck %s --check-prefixes=CHECK,BDVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-sse4.1 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2   | FileCheck %s --check-prefixes=CHECK,BTVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-sse4.1 | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
@@ -117,17 +117,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pabsb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pabsb %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pabsb %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pabsb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pabsb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpabsb %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpabsb (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpabsb %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pabsb:
 ; BTVER2-SSE:       # %bb.0:
@@ -260,17 +260,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pabsd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pabsd %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pabsd %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pabsd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pabsd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpabsd %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpabsd (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpabsd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pabsd:
 ; BTVER2-SSE:       # %bb.0:
@@ -403,17 +403,17 @@
 ;
 ; BDVER2-SSE-LABEL: test_pabsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pabsw %xmm0, %xmm1 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    pabsw %xmm0, %xmm1 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    pabsw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    por %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pabsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpabsw %xmm0, %xmm0 # sched: [1:0.50]
 ; BDVER2-NEXT:    vpabsw (%rdi), %xmm1 # sched: [7:0.50]
-; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpabsw %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pabsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -541,16 +541,16 @@
 ;
 ; BDVER2-SSE-LABEL: test_palignr:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    palignr {{.*#+}} xmm1 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [1:0.50]
+; BDVER2-SSE-NEXT:    palignr {{.*#+}} xmm1 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    palignr {{.*#+}} xmm1 = mem[14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [7:0.50]
-; BDVER2-SSE-NEXT:    movdqa %xmm1, %xmm0 # sched: [1:0.33]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    movdqa %xmm1, %xmm0 # sched: [2:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_palignr:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [1:0.50]
+; BDVER2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [2:0.50]
 ; BDVER2-NEXT:    vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_palignr:
 ; BTVER2-SSE:       # %bb.0:
@@ -664,15 +664,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phaddd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phaddd %xmm1, %xmm0 # sched: [3:1.50]
-; BDVER2-SSE-NEXT:    phaddd (%rdi), %xmm0 # sched: [9:1.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phaddd %xmm1, %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    phaddd (%rdi), %xmm0 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phaddd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER2-NEXT:    vphaddd (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vphaddd (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phaddd:
 ; BTVER2-SSE:       # %bb.0:
@@ -785,15 +785,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phaddsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phaddsw %xmm1, %xmm0 # sched: [3:1.50]
-; BDVER2-SSE-NEXT:    phaddsw (%rdi), %xmm0 # sched: [9:1.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phaddsw %xmm1, %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    phaddsw (%rdi), %xmm0 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phaddsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphaddsw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER2-NEXT:    vphaddsw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphaddsw %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vphaddsw (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phaddsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -906,15 +906,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phaddw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phaddw %xmm1, %xmm0 # sched: [3:1.50]
-; BDVER2-SSE-NEXT:    phaddw (%rdi), %xmm0 # sched: [9:1.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phaddw %xmm1, %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    phaddw (%rdi), %xmm0 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phaddw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphaddw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER2-NEXT:    vphaddw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphaddw %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vphaddw (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phaddw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1027,15 +1027,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phsubd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phsubd %xmm1, %xmm0 # sched: [3:1.50]
-; BDVER2-SSE-NEXT:    phsubd (%rdi), %xmm0 # sched: [9:1.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phsubd %xmm1, %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    phsubd (%rdi), %xmm0 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phsubd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphsubd %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER2-NEXT:    vphsubd (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphsubd %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vphsubd (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phsubd:
 ; BTVER2-SSE:       # %bb.0:
@@ -1148,15 +1148,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phsubsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phsubsw %xmm1, %xmm0 # sched: [3:1.50]
-; BDVER2-SSE-NEXT:    phsubsw (%rdi), %xmm0 # sched: [9:1.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phsubsw %xmm1, %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    phsubsw (%rdi), %xmm0 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phsubsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphsubsw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER2-NEXT:    vphsubsw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphsubsw %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vphsubsw (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phsubsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1269,15 +1269,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_phsubw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    phsubw %xmm1, %xmm0 # sched: [3:1.50]
-; BDVER2-SSE-NEXT:    phsubw (%rdi), %xmm0 # sched: [9:1.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    phsubw %xmm1, %xmm0 # sched: [5:0.50]
+; BDVER2-SSE-NEXT:    phsubw (%rdi), %xmm0 # sched: [10:0.50]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_phsubw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vphsubw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER2-NEXT:    vphsubw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vphsubw %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
+; BDVER2-NEXT:    vphsubw (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_phsubw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1390,15 +1390,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmaddubsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmaddubsw %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmaddubsw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmaddubsw %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmaddubsw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmaddubsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmaddubsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmaddubsw %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmaddubsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1512,15 +1512,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pmulhrsw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pmulhrsw %xmm1, %xmm0 # sched: [5:1.00]
-; BDVER2-SSE-NEXT:    pmulhrsw (%rdi), %xmm0 # sched: [11:1.00]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pmulhrsw %xmm1, %xmm0 # sched: [4:1.00]
+; BDVER2-SSE-NEXT:    pmulhrsw (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pmulhrsw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpmulhrsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER2-NEXT:    vpmulhrsw (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpmulhrsw %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER2-NEXT:    vpmulhrsw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pmulhrsw:
 ; BTVER2-SSE:       # %bb.0:
@@ -1633,15 +1633,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_pshufb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    pshufb %xmm1, %xmm0 # sched: [1:0.50]
-; BDVER2-SSE-NEXT:    pshufb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    pshufb %xmm1, %xmm0 # sched: [3:2.00]
+; BDVER2-SSE-NEXT:    pshufb (%rdi), %xmm0 # sched: [8:2.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_pshufb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpshufb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER2-NEXT:    vpshufb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    vpshufb %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BDVER2-NEXT:    vpshufb (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_pshufb:
 ; BTVER2-SSE:       # %bb.0:
@@ -1758,15 +1758,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psignb:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psignb %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psignb %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psignb (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psignb:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsignb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsignb %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsignb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psignb:
 ; BTVER2-SSE:       # %bb.0:
@@ -1883,15 +1883,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psignd:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psignd %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psignd %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psignd (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psignd:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsignd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsignd %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsignd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psignd:
 ; BTVER2-SSE:       # %bb.0:
@@ -2008,15 +2008,15 @@
 ;
 ; BDVER2-SSE-LABEL: test_psignw:
 ; BDVER2-SSE:       # %bb.0:
-; BDVER2-SSE-NEXT:    psignw %xmm1, %xmm0 # sched: [1:0.50]
+; BDVER2-SSE-NEXT:    psignw %xmm1, %xmm0 # sched: [2:0.50]
 ; BDVER2-SSE-NEXT:    psignw (%rdi), %xmm0 # sched: [7:0.50]
-; BDVER2-SSE-NEXT:    retq # sched: [1:1.00]
+; BDVER2-SSE-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER2-LABEL: test_psignw:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    vpsignw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER2-NEXT:    vpsignw %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER2-NEXT:    vpsignw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-SSE-LABEL: test_psignw:
 ; BTVER2-SSE:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/tbm-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/tbm-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/tbm-schedule.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+tbm | FileCheck %s --check-prefix=GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+tbm | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver3 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver4 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER4
 
@@ -16,12 +16,12 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_bextri_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    bextrl $3076, %edi, %ecx # imm = 0xC04
-; BDVER2-NEXT:    # sched: [2:1.00]
 ; BDVER2-NEXT:    bextrl $3076, (%rsi), %eax # imm = 0xC04
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    # sched: [6:0.50]
+; BDVER2-NEXT:    bextrl $3076, %edi, %ecx # imm = 0xC04
+; BDVER2-NEXT:    # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_bextri_u32:
 ; BDVER3:       # %bb.0:
@@ -57,12 +57,12 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_bextri_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    bextrl $3076, %edi, %ecx # imm = 0xC04
-; BDVER2-NEXT:    # sched: [2:1.00]
 ; BDVER2-NEXT:    bextrl $3076, (%rsi), %eax # imm = 0xC04
-; BDVER2-NEXT:    # sched: [7:1.00]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    # sched: [6:0.50]
+; BDVER2-NEXT:    bextrl $3076, %edi, %ecx # imm = 0xC04
+; BDVER2-NEXT:    # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_bextri_u64:
 ; BDVER3:       # %bb.0:
@@ -96,10 +96,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcfill_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcfilll %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcfilll (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcfilll %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcfill_u32:
 ; BDVER3:       # %bb.0:
@@ -133,10 +133,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcfill_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcfillq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcfillq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcfillq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcfill_u64:
 ; BDVER3:       # %bb.0:
@@ -170,10 +170,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blci_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcil %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcil (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcil %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blci_u32:
 ; BDVER3:       # %bb.0:
@@ -209,10 +209,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blci_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blciq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blciq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blciq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blci_u64:
 ; BDVER3:       # %bb.0:
@@ -248,10 +248,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcic_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcicl %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcicl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcicl %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcic_u32:
 ; BDVER3:       # %bb.0:
@@ -287,10 +287,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcic_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcicq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcicq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcicq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcic_u64:
 ; BDVER3:       # %bb.0:
@@ -326,10 +326,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcmsk_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcmskl %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcmskl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcmskl %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcmsk_u32:
 ; BDVER3:       # %bb.0:
@@ -363,10 +363,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcmsk_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcmskq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcmskq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcmskq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcmsk_u64:
 ; BDVER3:       # %bb.0:
@@ -400,10 +400,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcs_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcsl %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcsl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcsl %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcs_u32:
 ; BDVER3:       # %bb.0:
@@ -437,10 +437,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blcs_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blcsq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blcsq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blcsq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blcs_u64:
 ; BDVER3:       # %bb.0:
@@ -474,10 +474,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blsfill_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blsfilll %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blsfilll (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsfilll %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blsfill_u32:
 ; BDVER3:       # %bb.0:
@@ -511,10 +511,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blsfill_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blsfillq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blsfillq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsfillq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blsfill_u64:
 ; BDVER3:       # %bb.0:
@@ -548,10 +548,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blsic_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blsicl %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    blsicl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsicl %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blsic_u32:
 ; BDVER3:       # %bb.0:
@@ -587,10 +587,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_blsic_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    blsicq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    blsicq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    blsicq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_blsic_u64:
 ; BDVER3:       # %bb.0:
@@ -626,10 +626,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_t1mskc_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    t1mskcl %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    t1mskcl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    t1mskcl %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_t1mskc_u32:
 ; BDVER3:       # %bb.0:
@@ -665,10 +665,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_t1mskc_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    t1mskcq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    t1mskcq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    t1mskcq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_t1mskc_u64:
 ; BDVER3:       # %bb.0:
@@ -704,10 +704,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_tzmsk_u32:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    tzmskl %edi, %ecx # sched: [1:0.33]
 ; BDVER2-NEXT:    tzmskl (%rsi), %eax # sched: [6:0.50]
-; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    tzmskl %edi, %ecx # sched: [2:0.50]
+; BDVER2-NEXT:    addl %ecx, %eax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_tzmsk_u32:
 ; BDVER3:       # %bb.0:
@@ -743,10 +743,10 @@
 ;
 ; BDVER2-LABEL: test_x86_tbm_tzmsk_u64:
 ; BDVER2:       # %bb.0:
-; BDVER2-NEXT:    tzmskq %rdi, %rcx # sched: [1:0.33]
 ; BDVER2-NEXT:    tzmskq (%rsi), %rax # sched: [6:0.50]
-; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.33]
-; BDVER2-NEXT:    retq # sched: [1:1.00]
+; BDVER2-NEXT:    tzmskq %rdi, %rcx # sched: [2:0.50]
+; BDVER2-NEXT:    addq %rcx, %rax # sched: [1:0.50]
+; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_x86_tbm_tzmsk_u64:
 ; BDVER3:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/wide-fma-contraction.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/wide-fma-contraction.ll
+++ llvm/trunk/test/CodeGen/X86/wide-fma-contraction.ll
@@ -30,8 +30,8 @@
 ; CHECK-NOFMA-NEXT:    andl $-32, %esp
 ; CHECK-NOFMA-NEXT:    subl $32, %esp
 ; CHECK-NOFMA-NEXT:    vmulps %ymm2, %ymm0, %ymm0
-; CHECK-NOFMA-NEXT:    vaddps 8(%ebp), %ymm0, %ymm0
 ; CHECK-NOFMA-NEXT:    vmulps %ymm3, %ymm1, %ymm1
+; CHECK-NOFMA-NEXT:    vaddps 8(%ebp), %ymm0, %ymm0
 ; CHECK-NOFMA-NEXT:    vaddps 40(%ebp), %ymm1, %ymm1
 ; CHECK-NOFMA-NEXT:    movl %ebp, %esp
 ; CHECK-NOFMA-NEXT:    popl %ebp
Index: llvm/trunk/test/CodeGen/X86/x87-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/x87-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/x87-schedule.ll
@@ -8,7 +8,7 @@
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
-; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
 
@@ -72,9 +72,9 @@
 ; BDVER2-LABEL: test_f2xm1:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    f2xm1 # sched: [100:0.33]
+; BDVER2-NEXT:    f2xm1 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_f2xm1:
 ; BTVER2:       # %bb.0:
@@ -155,7 +155,7 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    fabs # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fabs:
 ; BTVER2:       # %bb.0:
@@ -276,12 +276,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fadd %st(0), %st(1) # sched: [3:1.00]
-; BDVER2-NEXT:    fadd %st(2) # sched: [3:1.00]
+; BDVER2-NEXT:    fadd %st(0), %st(1) # sched: [5:1.00]
+; BDVER2-NEXT:    fadd %st(2) # sched: [5:1.00]
 ; BDVER2-NEXT:    fadds (%ecx) # sched: [10:1.00]
 ; BDVER2-NEXT:    faddl (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fadd:
 ; BTVER2:       # %bb.0:
@@ -412,12 +412,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    faddp %st(1) # sched: [3:1.00]
-; BDVER2-NEXT:    faddp %st(2) # sched: [3:1.00]
-; BDVER2-NEXT:    fiadds (%ecx) # sched: [13:2.00]
-; BDVER2-NEXT:    fiaddl (%eax) # sched: [13:2.00]
+; BDVER2-NEXT:    faddp %st(1) # sched: [5:1.00]
+; BDVER2-NEXT:    faddp %st(2) # sched: [5:1.00]
+; BDVER2-NEXT:    fiadds (%ecx) # sched: [10:1.00]
+; BDVER2-NEXT:    fiaddl (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_faddp_fiadd:
 ; BTVER2:       # %bb.0:
@@ -523,10 +523,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fbld (%eax) # sched: [100:0.33]
-; BDVER2-NEXT:    fbstp (%eax) # sched: [100:0.33]
+; BDVER2-NEXT:    fbld (%eax) # sched: [100:0.50]
+; BDVER2-NEXT:    fbstp (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fbld_fbstp:
 ; BTVER2:       # %bb.0:
@@ -611,7 +611,7 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    fchs # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fchs:
 ; BTVER2:       # %bb.0:
@@ -698,10 +698,10 @@
 ; BDVER2-LABEL: test_fclex:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    wait # sched: [100:0.33]
-; BDVER2-NEXT:    fnclex # sched: [100:0.33]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
+; BDVER2-NEXT:    fnclex # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fclex:
 ; BTVER2:       # %bb.0:
@@ -782,9 +782,9 @@
 ; BDVER2-LABEL: test_fnclex:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fnclex # sched: [100:0.33]
+; BDVER2-NEXT:    fnclex # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fnclex:
 ; BTVER2:       # %bb.0:
@@ -919,16 +919,16 @@
 ; BDVER2-LABEL: test_fcmov:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fcmovb %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmovbe %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmove %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmovnb %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmovnbe %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmovne %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmovnu %st(1), %st(0) # sched: [3:2.00]
-; BDVER2-NEXT:    fcmovu %st(1), %st(0) # sched: [3:2.00]
+; BDVER2-NEXT:    fcmovb %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmovbe %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmove %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmovnb %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmovnbe %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmovne %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmovnu %st(1), %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    fcmovu %st(1), %st(0) # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fcmov:
 ; BTVER2:       # %bb.0:
@@ -1065,10 +1065,10 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    fcom %st(1) # sched: [1:1.00]
 ; BDVER2-NEXT:    fcom %st(3) # sched: [1:1.00]
-; BDVER2-NEXT:    fcoms (%ecx) # sched: [8:1.00]
-; BDVER2-NEXT:    fcoml (%eax) # sched: [8:1.00]
+; BDVER2-NEXT:    fcoms (%ecx) # sched: [6:1.00]
+; BDVER2-NEXT:    fcoml (%eax) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fcom:
 ; BTVER2:       # %bb.0:
@@ -1209,11 +1209,11 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    fcomp %st(1) # sched: [1:1.00]
 ; BDVER2-NEXT:    fcomp %st(3) # sched: [1:1.00]
-; BDVER2-NEXT:    fcomps (%ecx) # sched: [8:1.00]
-; BDVER2-NEXT:    fcompl (%eax) # sched: [8:1.00]
-; BDVER2-NEXT:    fcompp # sched: [100:0.33]
+; BDVER2-NEXT:    fcomps (%ecx) # sched: [6:1.00]
+; BDVER2-NEXT:    fcompl (%eax) # sched: [6:1.00]
+; BDVER2-NEXT:    fcompp # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fcomp_fcompp:
 ; BTVER2:       # %bb.0:
@@ -1312,10 +1312,10 @@
 ; BDVER2-LABEL: test_fcomi_fcomip:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fcomi %st(3) # sched: [3:1.00]
-; BDVER2-NEXT:    fcompi %st(3) # sched: [3:1.00]
+; BDVER2-NEXT:    fcomi %st(3) # sched: [1:1.00]
+; BDVER2-NEXT:    fcompi %st(3) # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fcomi_fcomip:
 ; BTVER2:       # %bb.0:
@@ -1396,9 +1396,9 @@
 ; BDVER2-LABEL: test_fcos:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fcos # sched: [100:0.33]
+; BDVER2-NEXT:    fcos # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fcos:
 ; BTVER2:       # %bb.0:
@@ -1477,9 +1477,9 @@
 ; BDVER2-LABEL: test_fdecstp:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fdecstp # sched: [1:1.00]
+; BDVER2-NEXT:    fdecstp # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fdecstp:
 ; BTVER2:       # %bb.0:
@@ -1600,12 +1600,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fdiv %st(0), %st(1) # sched: [14:14.00]
-; BDVER2-NEXT:    fdiv %st(2) # sched: [14:14.00]
-; BDVER2-NEXT:    fdivs (%ecx) # sched: [31:1.00]
-; BDVER2-NEXT:    fdivl (%eax) # sched: [31:1.00]
+; BDVER2-NEXT:    fdiv %st(0), %st(1) # sched: [9:9.50]
+; BDVER2-NEXT:    fdiv %st(2) # sched: [9:9.50]
+; BDVER2-NEXT:    fdivs (%ecx) # sched: [14:9.50]
+; BDVER2-NEXT:    fdivl (%eax) # sched: [14:9.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fdiv:
 ; BTVER2:       # %bb.0:
@@ -1736,12 +1736,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fdivp %st(1) # sched: [14:14.00]
-; BDVER2-NEXT:    fdivp %st(2) # sched: [14:14.00]
-; BDVER2-NEXT:    fidivs (%ecx) # sched: [34:1.00]
-; BDVER2-NEXT:    fidivl (%eax) # sched: [34:1.00]
+; BDVER2-NEXT:    fdivp %st(1) # sched: [9:9.50]
+; BDVER2-NEXT:    fdivp %st(2) # sched: [9:9.50]
+; BDVER2-NEXT:    fidivs (%ecx) # sched: [14:9.50]
+; BDVER2-NEXT:    fidivl (%eax) # sched: [14:9.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fdivp_fidiv:
 ; BTVER2:       # %bb.0:
@@ -1872,12 +1872,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fdivr %st(0), %st(1) # sched: [14:14.00]
-; BDVER2-NEXT:    fdivr %st(2) # sched: [14:14.00]
-; BDVER2-NEXT:    fdivrs (%ecx) # sched: [31:1.00]
-; BDVER2-NEXT:    fdivrl (%eax) # sched: [31:1.00]
+; BDVER2-NEXT:    fdivr %st(0), %st(1) # sched: [9:9.50]
+; BDVER2-NEXT:    fdivr %st(2) # sched: [9:9.50]
+; BDVER2-NEXT:    fdivrs (%ecx) # sched: [14:9.50]
+; BDVER2-NEXT:    fdivrl (%eax) # sched: [14:9.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fdivr:
 ; BTVER2:       # %bb.0:
@@ -2008,12 +2008,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fdivrp %st(1) # sched: [14:14.00]
-; BDVER2-NEXT:    fdivrp %st(2) # sched: [14:14.00]
-; BDVER2-NEXT:    fidivrs (%ecx) # sched: [34:1.00]
-; BDVER2-NEXT:    fidivrl (%eax) # sched: [34:1.00]
+; BDVER2-NEXT:    fdivrp %st(1) # sched: [9:9.50]
+; BDVER2-NEXT:    fdivrp %st(2) # sched: [9:9.50]
+; BDVER2-NEXT:    fidivrs (%ecx) # sched: [14:9.50]
+; BDVER2-NEXT:    fidivrl (%eax) # sched: [14:9.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fdivrp_fidivr:
 ; BTVER2:       # %bb.0:
@@ -2102,9 +2102,9 @@
 ; BDVER2-LABEL: test_ffree:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    ffree %st(0) # sched: [1:1.00]
+; BDVER2-NEXT:    ffree %st(0) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ffree:
 ; BTVER2:       # %bb.0:
@@ -2225,12 +2225,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    ficoms (%ecx) # sched: [11:2.00]
-; BDVER2-NEXT:    ficoml (%eax) # sched: [11:2.00]
-; BDVER2-NEXT:    ficomps (%ecx) # sched: [11:2.00]
-; BDVER2-NEXT:    ficompl (%eax) # sched: [11:2.00]
+; BDVER2-NEXT:    ficoms (%ecx) # sched: [6:1.00]
+; BDVER2-NEXT:    ficoml (%eax) # sched: [6:1.00]
+; BDVER2-NEXT:    ficomps (%ecx) # sched: [6:1.00]
+; BDVER2-NEXT:    ficompl (%eax) # sched: [6:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ficom:
 ; BTVER2:       # %bb.0:
@@ -2362,11 +2362,11 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %edx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    filds (%edx) # sched: [10:1.00]
-; BDVER2-NEXT:    fildl (%ecx) # sched: [10:1.00]
-; BDVER2-NEXT:    fildll (%eax) # sched: [10:1.00]
+; BDVER2-NEXT:    filds (%edx) # sched: [5:0.50]
+; BDVER2-NEXT:    fildl (%ecx) # sched: [5:0.50]
+; BDVER2-NEXT:    fildll (%eax) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fild:
 ; BTVER2:       # %bb.0:
@@ -2455,9 +2455,9 @@
 ; BDVER2-LABEL: test_fincstp:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fincstp # sched: [1:1.00]
+; BDVER2-NEXT:    fincstp # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fincstp:
 ; BTVER2:       # %bb.0:
@@ -2544,10 +2544,10 @@
 ; BDVER2-LABEL: test_finit:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    wait # sched: [100:0.33]
-; BDVER2-NEXT:    fninit # sched: [5:1.33]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
+; BDVER2-NEXT:    fninit # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_finit:
 ; BTVER2:       # %bb.0:
@@ -2628,9 +2628,9 @@
 ; BDVER2-LABEL: test_fninit:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fninit # sched: [5:1.33]
+; BDVER2-NEXT:    fninit # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fninit:
 ; BTVER2:       # %bb.0:
@@ -2792,16 +2792,16 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %edx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fists (%edx) # sched: [9:1.00]
-; BDVER2-NEXT:    fistl (%ecx) # sched: [9:1.00]
-; BDVER2-NEXT:    fistps (%edx) # sched: [9:1.00]
-; BDVER2-NEXT:    fistpl (%ecx) # sched: [9:1.00]
-; BDVER2-NEXT:    fistpll (%eax) # sched: [9:1.00]
-; BDVER2-NEXT:    fisttps (%edx) # sched: [5:1.00]
-; BDVER2-NEXT:    fisttpl (%ecx) # sched: [5:1.00]
-; BDVER2-NEXT:    fisttpll (%eax) # sched: [5:1.00]
+; BDVER2-NEXT:    fists (%edx) # sched: [1:0.50]
+; BDVER2-NEXT:    fistl (%ecx) # sched: [1:0.50]
+; BDVER2-NEXT:    fistps (%edx) # sched: [1:0.50]
+; BDVER2-NEXT:    fistpl (%ecx) # sched: [1:0.50]
+; BDVER2-NEXT:    fistpll (%eax) # sched: [1:0.50]
+; BDVER2-NEXT:    fisttps (%edx) # sched: [1:0.50]
+; BDVER2-NEXT:    fisttpl (%ecx) # sched: [1:0.50]
+; BDVER2-NEXT:    fisttpll (%eax) # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fist_fistp_fisttp:
 ; BTVER2:       # %bb.0:
@@ -2951,12 +2951,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %edx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fld %st(0) # sched: [1:1.00]
-; BDVER2-NEXT:    flds (%edx) # sched: [9:1.00]
-; BDVER2-NEXT:    fldl (%ecx) # sched: [9:1.00]
-; BDVER2-NEXT:    fldt (%eax) # sched: [9:1.00]
+; BDVER2-NEXT:    fld %st(0) # sched: [1:0.50]
+; BDVER2-NEXT:    flds (%edx) # sched: [5:0.50]
+; BDVER2-NEXT:    fldl (%ecx) # sched: [5:0.50]
+; BDVER2-NEXT:    fldt (%eax) # sched: [5:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fld:
 ; BTVER2:       # %bb.0:
@@ -3064,10 +3064,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fldcw (%eax) # sched: [8:2.00]
-; BDVER2-NEXT:    fldenv (%eax) # sched: [100:0.33]
+; BDVER2-NEXT:    fldcw (%eax) # sched: [5:0.50]
+; BDVER2-NEXT:    fldenv (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fldcw_fldenv:
 ; BTVER2:       # %bb.0:
@@ -3198,15 +3198,15 @@
 ; BDVER2-LABEL: test_fld1_fldl2e_fldl2t_fldlg2_fldln2_fldpi_fldz:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fld1 # sched: [1:1.00]
-; BDVER2-NEXT:    fldl2e # sched: [1:1.00]
-; BDVER2-NEXT:    fldl2t # sched: [1:1.00]
-; BDVER2-NEXT:    fldlg2 # sched: [1:1.00]
-; BDVER2-NEXT:    fldln2 # sched: [1:1.00]
-; BDVER2-NEXT:    fldpi # sched: [1:1.00]
-; BDVER2-NEXT:    fldz # sched: [1:1.00]
+; BDVER2-NEXT:    fld1 # sched: [3:1.00]
+; BDVER2-NEXT:    fldl2e # sched: [3:1.00]
+; BDVER2-NEXT:    fldl2t # sched: [3:1.00]
+; BDVER2-NEXT:    fldlg2 # sched: [3:1.00]
+; BDVER2-NEXT:    fldln2 # sched: [3:1.00]
+; BDVER2-NEXT:    fldpi # sched: [3:1.00]
+; BDVER2-NEXT:    fldz # sched: [3:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fld1_fldl2e_fldl2t_fldlg2_fldln2_fldpi_fldz:
 ; BTVER2:       # %bb.0:
@@ -3341,10 +3341,10 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    fmul %st(0), %st(1) # sched: [5:1.00]
 ; BDVER2-NEXT:    fmul %st(2) # sched: [5:1.00]
-; BDVER2-NEXT:    fmuls (%ecx) # sched: [12:1.00]
-; BDVER2-NEXT:    fmull (%eax) # sched: [12:1.00]
+; BDVER2-NEXT:    fmuls (%ecx) # sched: [10:1.00]
+; BDVER2-NEXT:    fmull (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fmul:
 ; BTVER2:       # %bb.0:
@@ -3477,10 +3477,10 @@
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    fmulp %st(1) # sched: [5:1.00]
 ; BDVER2-NEXT:    fmulp %st(2) # sched: [5:1.00]
-; BDVER2-NEXT:    fimuls (%ecx) # sched: [15:1.00]
-; BDVER2-NEXT:    fimull (%eax) # sched: [15:1.00]
+; BDVER2-NEXT:    fimuls (%ecx) # sched: [10:1.00]
+; BDVER2-NEXT:    fimull (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fmulp_fimul:
 ; BTVER2:       # %bb.0:
@@ -3569,9 +3569,9 @@
 ; BDVER2-LABEL: test_fnop:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fnop # sched: [1:1.00]
+; BDVER2-NEXT:    fnop # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fnop:
 ; BTVER2:       # %bb.0:
@@ -3650,9 +3650,9 @@
 ; BDVER2-LABEL: test_fpatan:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fpatan # sched: [100:0.33]
+; BDVER2-NEXT:    fpatan # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fpatan:
 ; BTVER2:       # %bb.0:
@@ -3739,10 +3739,10 @@
 ; BDVER2-LABEL: test_fprem_fprem1:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fprem # sched: [100:0.33]
-; BDVER2-NEXT:    fprem1 # sched: [100:0.33]
+; BDVER2-NEXT:    fprem # sched: [100:0.50]
+; BDVER2-NEXT:    fprem1 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fprem_fprem1:
 ; BTVER2:       # %bb.0:
@@ -3823,9 +3823,9 @@
 ; BDVER2-LABEL: test_fptan:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fptan # sched: [100:0.33]
+; BDVER2-NEXT:    fptan # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fptan:
 ; BTVER2:       # %bb.0:
@@ -3904,9 +3904,9 @@
 ; BDVER2-LABEL: test_frndint:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    frndint # sched: [100:0.33]
+; BDVER2-NEXT:    frndint # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_frndint:
 ; BTVER2:       # %bb.0:
@@ -3994,9 +3994,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    frstor (%eax) # sched: [100:0.33]
+; BDVER2-NEXT:    frstor (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_frstor:
 ; BTVER2:       # %bb.0:
@@ -4094,10 +4094,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    wait # sched: [100:0.33]
-; BDVER2-NEXT:    fnsave (%eax) # sched: [100:0.33]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
+; BDVER2-NEXT:    fnsave (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsave:
 ; BTVER2:       # %bb.0:
@@ -4189,9 +4189,9 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fnsave (%eax) # sched: [100:0.33]
+; BDVER2-NEXT:    fnsave (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fnsave:
 ; BTVER2:       # %bb.0:
@@ -4272,9 +4272,9 @@
 ; BDVER2-LABEL: test_fscale:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fscale # sched: [100:0.33]
+; BDVER2-NEXT:    fscale # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fscale:
 ; BTVER2:       # %bb.0:
@@ -4353,9 +4353,9 @@
 ; BDVER2-LABEL: test_fsin:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsin # sched: [100:0.33]
+; BDVER2-NEXT:    fsin # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsin:
 ; BTVER2:       # %bb.0:
@@ -4434,9 +4434,9 @@
 ; BDVER2-LABEL: test_fsincos:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsincos # sched: [100:0.33]
+; BDVER2-NEXT:    fsincos # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsincos:
 ; BTVER2:       # %bb.0:
@@ -4515,9 +4515,9 @@
 ; BDVER2-LABEL: test_fsqrt:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsqrt # sched: [24:24.00]
+; BDVER2-NEXT:    fsqrt # sched: [1:17.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsqrt:
 ; BTVER2:       # %bb.0:
@@ -4671,15 +4671,15 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %edx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fst %st(0) # sched: [1:1.00]
-; BDVER2-NEXT:    fsts (%edx) # sched: [6:1.00]
-; BDVER2-NEXT:    fstl (%ecx) # sched: [6:1.00]
-; BDVER2-NEXT:    fstp %st(0) # sched: [1:1.00]
-; BDVER2-NEXT:    fstpl (%edx) # sched: [6:1.00]
-; BDVER2-NEXT:    fstpl (%ecx) # sched: [6:1.00]
-; BDVER2-NEXT:    fstpt (%eax) # sched: [6:1.00]
+; BDVER2-NEXT:    fst %st(0) # sched: [1:0.50]
+; BDVER2-NEXT:    fsts (%edx) # sched: [1:0.50]
+; BDVER2-NEXT:    fstl (%ecx) # sched: [1:0.50]
+; BDVER2-NEXT:    fstp %st(0) # sched: [1:0.50]
+; BDVER2-NEXT:    fstpl (%edx) # sched: [1:0.50]
+; BDVER2-NEXT:    fstpl (%ecx) # sched: [1:0.50]
+; BDVER2-NEXT:    fstpt (%eax) # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fst_fstp:
 ; BTVER2:       # %bb.0:
@@ -4825,14 +4825,14 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    wait # sched: [100:0.33]
-; BDVER2-NEXT:    fnstcw (%eax) # sched: [7:1.00]
-; BDVER2-NEXT:    wait # sched: [100:0.33]
-; BDVER2-NEXT:    fnstenv (%eax) # sched: [100:0.33]
-; BDVER2-NEXT:    wait # sched: [100:0.33]
-; BDVER2-NEXT:    fnstsw (%eax) # sched: [7:1.00]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
+; BDVER2-NEXT:    fnstcw (%eax) # sched: [1:0.50]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
+; BDVER2-NEXT:    fnstenv (%eax) # sched: [100:0.50]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
+; BDVER2-NEXT:    fnstsw (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fstcw_fstenv_fstsw:
 ; BTVER2:       # %bb.0:
@@ -4948,11 +4948,11 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fnstcw (%eax) # sched: [7:1.00]
-; BDVER2-NEXT:    fnstenv (%eax) # sched: [100:0.33]
-; BDVER2-NEXT:    fnstsw (%eax) # sched: [7:1.00]
+; BDVER2-NEXT:    fnstcw (%eax) # sched: [1:0.50]
+; BDVER2-NEXT:    fnstenv (%eax) # sched: [100:0.50]
+; BDVER2-NEXT:    fnstsw (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fnstcw_fnstenv_fnstsw:
 ; BTVER2:       # %bb.0:
@@ -5079,12 +5079,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsub %st(0), %st(1) # sched: [3:1.00]
-; BDVER2-NEXT:    fsub %st(2) # sched: [3:1.00]
+; BDVER2-NEXT:    fsub %st(0), %st(1) # sched: [5:1.00]
+; BDVER2-NEXT:    fsub %st(2) # sched: [5:1.00]
 ; BDVER2-NEXT:    fsubs (%ecx) # sched: [10:1.00]
 ; BDVER2-NEXT:    fsubl (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsub:
 ; BTVER2:       # %bb.0:
@@ -5215,12 +5215,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsubp %st(1) # sched: [3:1.00]
-; BDVER2-NEXT:    fsubp %st(2) # sched: [3:1.00]
-; BDVER2-NEXT:    fisubs (%ecx) # sched: [13:2.00]
-; BDVER2-NEXT:    fisubl (%eax) # sched: [13:2.00]
+; BDVER2-NEXT:    fsubp %st(1) # sched: [5:1.00]
+; BDVER2-NEXT:    fsubp %st(2) # sched: [5:1.00]
+; BDVER2-NEXT:    fisubs (%ecx) # sched: [10:1.00]
+; BDVER2-NEXT:    fisubl (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsubp_fisub:
 ; BTVER2:       # %bb.0:
@@ -5351,12 +5351,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsubr %st(0), %st(1) # sched: [3:1.00]
-; BDVER2-NEXT:    fsubr %st(2) # sched: [3:1.00]
+; BDVER2-NEXT:    fsubr %st(0), %st(1) # sched: [5:1.00]
+; BDVER2-NEXT:    fsubr %st(2) # sched: [5:1.00]
 ; BDVER2-NEXT:    fsubrs (%ecx) # sched: [10:1.00]
 ; BDVER2-NEXT:    fsubrl (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsubr:
 ; BTVER2:       # %bb.0:
@@ -5487,12 +5487,12 @@
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %ecx # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fsubrp %st(1) # sched: [3:1.00]
-; BDVER2-NEXT:    fsubrp %st(2) # sched: [3:1.00]
-; BDVER2-NEXT:    fisubrs (%ecx) # sched: [13:2.00]
-; BDVER2-NEXT:    fisubrl (%eax) # sched: [13:2.00]
+; BDVER2-NEXT:    fsubrp %st(1) # sched: [5:1.00]
+; BDVER2-NEXT:    fsubrp %st(2) # sched: [5:1.00]
+; BDVER2-NEXT:    fisubrs (%ecx) # sched: [10:1.00]
+; BDVER2-NEXT:    fisubrl (%eax) # sched: [10:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fsubrp_fisubr:
 ; BTVER2:       # %bb.0:
@@ -5581,9 +5581,9 @@
 ; BDVER2-LABEL: test_ftst:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    ftst # sched: [3:1.00]
+; BDVER2-NEXT:    ftst # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_ftst:
 ; BTVER2:       # %bb.0:
@@ -5698,9 +5698,9 @@
 ; BDVER2-NEXT:    fucom %st(3) # sched: [1:1.00]
 ; BDVER2-NEXT:    fucomp %st(1) # sched: [1:1.00]
 ; BDVER2-NEXT:    fucomp %st(3) # sched: [1:1.00]
-; BDVER2-NEXT:    fucompp # sched: [3:1.00]
+; BDVER2-NEXT:    fucompp # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fucom_fucomp_fucompp:
 ; BTVER2:       # %bb.0:
@@ -5795,10 +5795,10 @@
 ; BDVER2-LABEL: test_fucomi_fucomip:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fucomi %st(3) # sched: [3:1.00]
-; BDVER2-NEXT:    fucompi %st(3) # sched: [3:1.00]
+; BDVER2-NEXT:    fucomi %st(3) # sched: [1:1.00]
+; BDVER2-NEXT:    fucompi %st(3) # sched: [1:1.00]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fucomi_fucomip:
 ; BTVER2:       # %bb.0:
@@ -5879,9 +5879,9 @@
 ; BDVER2-LABEL: test_fwait:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    wait # sched: [100:0.33]
+; BDVER2-NEXT:    wait # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fwait:
 ; BTVER2:       # %bb.0:
@@ -5960,9 +5960,9 @@
 ; BDVER2-LABEL: test_fxam:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fxam # sched: [100:0.33]
+; BDVER2-NEXT:    fxam # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fxam:
 ; BTVER2:       # %bb.0:
@@ -6049,10 +6049,10 @@
 ; BDVER2-LABEL: test_fxch:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fxch %st(1) # sched: [1:0.33]
-; BDVER2-NEXT:    fxch %st(3) # sched: [1:0.33]
+; BDVER2-NEXT:    fxch %st(1) # sched: [1:0.50]
+; BDVER2-NEXT:    fxch %st(3) # sched: [1:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fxch:
 ; BTVER2:       # %bb.0:
@@ -6150,10 +6150,10 @@
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    movl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fxrstor (%eax) # sched: [5:2.00]
-; BDVER2-NEXT:    fxsave (%eax) # sched: [100:0.33]
+; BDVER2-NEXT:    fxrstor (%eax) # sched: [100:0.50]
+; BDVER2-NEXT:    fxsave (%eax) # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fxrstor_fxsave:
 ; BTVER2:       # %bb.0:
@@ -6236,9 +6236,9 @@
 ; BDVER2-LABEL: test_fxtract:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fxtract # sched: [100:0.33]
+; BDVER2-NEXT:    fxtract # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fxtract:
 ; BTVER2:       # %bb.0:
@@ -6317,9 +6317,9 @@
 ; BDVER2-LABEL: test_fyl2x:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fyl2x # sched: [100:0.33]
+; BDVER2-NEXT:    fyl2x # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fyl2x:
 ; BTVER2:       # %bb.0:
@@ -6398,9 +6398,9 @@
 ; BDVER2-LABEL: test_fyl2xp1:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
-; BDVER2-NEXT:    fyl2xp1 # sched: [100:0.33]
+; BDVER2-NEXT:    fyl2xp1 # sched: [100:0.50]
 ; BDVER2-NEXT:    #NO_APP
-; BDVER2-NEXT:    retl # sched: [6:1.00]
+; BDVER2-NEXT:    retl # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_fyl2xp1:
 ; BTVER2:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/xop-schedule.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/xop-schedule.ll
+++ llvm/trunk/test/CodeGen/X86/xop-schedule.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+xop | FileCheck %s --check-prefix=GENERIC
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+xop | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+xop | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver1 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver2 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER12 --check-prefix=BDVER2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver3 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=bdver4 | FileCheck %s --check-prefix=BDVER --check-prefix=BDVER4
 
@@ -20,13 +20,13 @@
 ; BDVER12-LABEL: test_vfrczpd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vfrczpd %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER12-NEXT:    vfrczpd %ymm1, %ymm1 # sched: [3:1.00]
-; BDVER12-NEXT:    vfrczpd (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER12-NEXT:    vfrczpd (%rsi), %ymm1 # sched: [10:1.00]
+; BDVER12-NEXT:    vfrczpd %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfrczpd %ymm1, %ymm1 # sched: [10:2.00]
+; BDVER12-NEXT:    vfrczpd (%rdi), %xmm0 # sched: [15:1.00]
+; BDVER12-NEXT:    vfrczpd (%rsi), %ymm1 # sched: [15:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vfrczpd:
 ; BDVER3:       # %bb.0:
@@ -68,13 +68,13 @@
 ; BDVER12-LABEL: test_vfrczps:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vfrczps %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER12-NEXT:    vfrczps %ymm1, %ymm1 # sched: [3:1.00]
-; BDVER12-NEXT:    vfrczps (%rdi), %xmm0 # sched: [9:1.00]
-; BDVER12-NEXT:    vfrczps (%rsi), %ymm1 # sched: [10:1.00]
+; BDVER12-NEXT:    vfrczps %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfrczps %ymm1, %ymm1 # sched: [10:2.00]
+; BDVER12-NEXT:    vfrczps (%rdi), %xmm0 # sched: [15:1.00]
+; BDVER12-NEXT:    vfrczps (%rsi), %ymm1 # sched: [15:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vfrczps:
 ; BDVER3:       # %bb.0:
@@ -113,10 +113,10 @@
 ; BDVER12-LABEL: test_vfrczsd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vfrczsd %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER12-NEXT:    vfrczsd (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER12-NEXT:    vfrczsd %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfrczsd (%rdi), %xmm0 # sched: [15:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vfrczsd:
 ; BDVER3:       # %bb.0:
@@ -149,10 +149,10 @@
 ; BDVER12-LABEL: test_vfrczss:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vfrczss %xmm0, %xmm0 # sched: [3:1.00]
-; BDVER12-NEXT:    vfrczss (%rdi), %xmm0 # sched: [9:1.00]
+; BDVER12-NEXT:    vfrczss %xmm0, %xmm0 # sched: [10:1.00]
+; BDVER12-NEXT:    vfrczss (%rdi), %xmm0 # sched: [15:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vfrczss:
 ; BDVER3:       # %bb.0:
@@ -186,11 +186,11 @@
 ; BDVER12-LABEL: test_vpcmov_128:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpcmov %xmm2, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER12-NEXT:    vpcmov %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER12-NEXT:    vpcmov (%rdi), %xmm1, %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcmov %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpcmov_128:
 ; BDVER3:       # %bb.0:
@@ -227,12 +227,12 @@
 ; BDVER12-LABEL: test_vpcmov_256:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpcmov (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER12-NEXT:    vpcmov %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
+; BDVER12-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcmov (%rdi), %ymm1, %ymm0, %ymm0 # sched: [7:1.00]
+; BDVER12-NEXT:    vpcmov %ymm2, (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpcmov_256:
 ; BDVER3:       # %bb.0:
@@ -275,16 +275,16 @@
 ; BDVER12-LABEL: test_vpcom:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpcomb $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpcomd $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpcomq $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpcomw $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER12-NEXT:    vpcomb $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcomd $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcomq $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcomw $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER12-NEXT:    vpcomb $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcomd $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcomq $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcomw $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpcom:
 ; BDVER3:       # %bb.0:
@@ -335,16 +335,16 @@
 ; BDVER12-LABEL: test_vpcomu:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpcomub $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpcomud $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpcomuq $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpcomuw $3, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BDVER12-NEXT:    vpcomub $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcomud $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcomuq $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vpcomuw $3, %xmm1, %xmm0, %xmm0 # sched: [2:0.50]
 ; BDVER12-NEXT:    vpcomub $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcomud $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcomuq $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    vpcomuw $3, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpcomu:
 ; BDVER3:       # %bb.0:
@@ -390,11 +390,11 @@
 ; BDVER12-LABEL: test_vpermil2pd_128:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpermil2pd $3, %xmm2, %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpermil2pd $3, %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpermil2pd $3, (%rdi), %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER12-NEXT:    vpermil2pd $3, %xmm2, %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BDVER12-NEXT:    vpermil2pd $3, %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
+; BDVER12-NEXT:    vpermil2pd $3, (%rdi), %xmm1, %xmm0, %xmm0 # sched: [8:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpermil2pd_128:
 ; BDVER3:       # %bb.0:
@@ -431,12 +431,12 @@
 ; BDVER12-LABEL: test_vpermil2pd_256:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpermil2pd $3, %ymm2, %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpermil2pd $3, %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER12-NEXT:    vpermil2pd $3, (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:1.00]
+; BDVER12-NEXT:    vpermil2pd $3, %ymm2, %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
+; BDVER12-NEXT:    vpermil2pd $3, %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
+; BDVER12-NEXT:    vpermil2pd $3, (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:3.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpermil2pd_256:
 ; BDVER3:       # %bb.0:
@@ -474,11 +474,11 @@
 ; BDVER12-LABEL: test_vpermil2ps_128:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpermil2ps $3, %xmm2, %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpermil2ps $3, %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpermil2ps $3, (%rdi), %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
+; BDVER12-NEXT:    vpermil2ps $3, %xmm2, %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BDVER12-NEXT:    vpermil2ps $3, %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
+; BDVER12-NEXT:    vpermil2ps $3, (%rdi), %xmm1, %xmm0, %xmm0 # sched: [8:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpermil2ps_128:
 ; BDVER3:       # %bb.0:
@@ -515,12 +515,12 @@
 ; BDVER12-LABEL: test_vpermil2ps_256:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpermil2ps $3, %ymm2, %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpermil2ps $3, %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
-; BDVER12-NEXT:    vpermil2ps $3, (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:1.00]
+; BDVER12-NEXT:    vpermil2ps $3, %ymm2, %ymm1, %ymm0, %ymm0 # sched: [3:3.00]
+; BDVER12-NEXT:    vpermil2ps $3, %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:3.00]
+; BDVER12-NEXT:    vpermil2ps $3, (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:3.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    vzeroupper # sched: [100:0.33]
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    vzeroupper # sched: [46:4.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpermil2ps_256:
 ; BDVER3:       # %bb.0:
@@ -557,10 +557,10 @@
 ; BDVER12-LABEL: test_vphaddbd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddbd %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddbd (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddbd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddbd (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddbd:
 ; BDVER3:       # %bb.0:
@@ -593,10 +593,10 @@
 ; BDVER12-LABEL: test_vphaddbq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddbq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddbq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddbq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddbq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddbq:
 ; BDVER3:       # %bb.0:
@@ -629,10 +629,10 @@
 ; BDVER12-LABEL: test_vphaddbw:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddbw %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddbw (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddbw %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddbw (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddbw:
 ; BDVER3:       # %bb.0:
@@ -665,10 +665,10 @@
 ; BDVER12-LABEL: test_vphadddq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphadddq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphadddq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphadddq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphadddq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphadddq:
 ; BDVER3:       # %bb.0:
@@ -701,10 +701,10 @@
 ; BDVER12-LABEL: test_vphaddubd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddubd %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddubd (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddubd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddubd (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddubd:
 ; BDVER3:       # %bb.0:
@@ -737,10 +737,10 @@
 ; BDVER12-LABEL: test_vphaddubq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddubq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddubq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddubq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddubq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddubq:
 ; BDVER3:       # %bb.0:
@@ -773,10 +773,10 @@
 ; BDVER12-LABEL: test_vphaddubw:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddubw %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddubw (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddubw %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddubw (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddubw:
 ; BDVER3:       # %bb.0:
@@ -809,10 +809,10 @@
 ; BDVER12-LABEL: test_vphaddudq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddudq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddudq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddudq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddudq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddudq:
 ; BDVER3:       # %bb.0:
@@ -845,10 +845,10 @@
 ; BDVER12-LABEL: test_vphadduwd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphadduwd %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphadduwd (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphadduwd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphadduwd (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphadduwd:
 ; BDVER3:       # %bb.0:
@@ -881,10 +881,10 @@
 ; BDVER12-LABEL: test_vphadduwq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphadduwq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphadduwq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphadduwq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphadduwq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphadduwq:
 ; BDVER3:       # %bb.0:
@@ -917,10 +917,10 @@
 ; BDVER12-LABEL: test_vphaddwd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddwd %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddwd (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddwd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddwd (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddwd:
 ; BDVER3:       # %bb.0:
@@ -953,10 +953,10 @@
 ; BDVER12-LABEL: test_vphaddwq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphaddwq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphaddwq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphaddwq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphaddwq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphaddwq:
 ; BDVER3:       # %bb.0:
@@ -989,10 +989,10 @@
 ; BDVER12-LABEL: test_vphsubbw:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphsubbw %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphsubbw (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphsubbw %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphsubbw (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphsubbw:
 ; BDVER3:       # %bb.0:
@@ -1025,10 +1025,10 @@
 ; BDVER12-LABEL: test_vphsubdq:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphsubdq %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphsubdq (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphsubdq %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphsubdq (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphsubdq:
 ; BDVER3:       # %bb.0:
@@ -1061,10 +1061,10 @@
 ; BDVER12-LABEL: test_vphsubwd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vphsubwd %xmm0, %xmm0 # sched: [3:1.50]
-; BDVER12-NEXT:    vphsubwd (%rdi), %xmm0 # sched: [9:1.50]
+; BDVER12-NEXT:    vphsubwd %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vphsubwd (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vphsubwd:
 ; BDVER3:       # %bb.0:
@@ -1097,10 +1097,10 @@
 ; BDVER12-LABEL: test_vpmacsdd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacsdd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacsdd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacsdd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
+; BDVER12-NEXT:    vpmacsdd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacsdd:
 ; BDVER3:       # %bb.0:
@@ -1133,10 +1133,10 @@
 ; BDVER12-LABEL: test_vpmacsdqh:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacsdqh %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacsdqh %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacsdqh %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:2.00]
+; BDVER12-NEXT:    vpmacsdqh %xmm2, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacsdqh:
 ; BDVER3:       # %bb.0:
@@ -1169,10 +1169,10 @@
 ; BDVER12-LABEL: test_vpmacsdql:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacsdql %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacsdql %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacsdql %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:2.00]
+; BDVER12-NEXT:    vpmacsdql %xmm2, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacsdql:
 ; BDVER3:       # %bb.0:
@@ -1205,10 +1205,10 @@
 ; BDVER12-LABEL: test_vpmacssdd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacssdd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacssdd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacssdd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
+; BDVER12-NEXT:    vpmacssdd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacssdd:
 ; BDVER3:       # %bb.0:
@@ -1241,10 +1241,10 @@
 ; BDVER12-LABEL: test_vpmacssdqh:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacssdqh %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacssdqh %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacssdqh %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:2.00]
+; BDVER12-NEXT:    vpmacssdqh %xmm2, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacssdqh:
 ; BDVER3:       # %bb.0:
@@ -1277,10 +1277,10 @@
 ; BDVER12-LABEL: test_vpmacssdql:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacssdql %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacssdql %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacssdql %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:2.00]
+; BDVER12-NEXT:    vpmacssdql %xmm2, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacssdql:
 ; BDVER3:       # %bb.0:
@@ -1313,10 +1313,10 @@
 ; BDVER12-LABEL: test_vpmacsswd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacsswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacsswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacsswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER12-NEXT:    vpmacsswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacsswd:
 ; BDVER3:       # %bb.0:
@@ -1349,10 +1349,10 @@
 ; BDVER12-LABEL: test_vpmacssww:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacssww %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacssww %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacssww %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER12-NEXT:    vpmacssww %xmm2, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacssww:
 ; BDVER3:       # %bb.0:
@@ -1385,10 +1385,10 @@
 ; BDVER12-LABEL: test_vpmacswd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER12-NEXT:    vpmacswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacswd:
 ; BDVER3:       # %bb.0:
@@ -1421,10 +1421,10 @@
 ; BDVER12-LABEL: test_vpmacsww:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmacsww %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmacsww %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmacsww %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER12-NEXT:    vpmacsww %xmm2, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmacsww:
 ; BDVER3:       # %bb.0:
@@ -1457,10 +1457,10 @@
 ; BDVER12-LABEL: test_vpmadcsswd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmadcsswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmadcsswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmadcsswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER12-NEXT:    vpmadcsswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmadcsswd:
 ; BDVER3:       # %bb.0:
@@ -1493,10 +1493,10 @@
 ; BDVER12-LABEL: test_vpmadcswd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpmadcswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
-; BDVER12-NEXT:    vpmadcswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
+; BDVER12-NEXT:    vpmadcswd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [4:1.00]
+; BDVER12-NEXT:    vpmadcswd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpmadcswd:
 ; BDVER3:       # %bb.0:
@@ -1530,11 +1530,11 @@
 ; BDVER12-LABEL: test_vpperm:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpperm %xmm2, %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
-; BDVER12-NEXT:    vpperm (%rdi), %xmm1, %xmm0, %xmm0 # sched: [7:0.50]
-; BDVER12-NEXT:    vpperm %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
+; BDVER12-NEXT:    vpperm %xmm2, %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BDVER12-NEXT:    vpperm (%rdi), %xmm1, %xmm0, %xmm0 # sched: [8:2.00]
+; BDVER12-NEXT:    vpperm %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpperm:
 ; BDVER3:       # %bb.0:
@@ -1587,28 +1587,28 @@
 ; BDVER12-LABEL: test_vprot:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vprotb %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotq %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotw %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotb (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotq (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotw (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotb %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotd %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotq %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotw %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotb $7, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotd $7, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotq $7, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotw $7, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vprotb $7, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotd $7, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotq $7, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vprotw $7, (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER12-NEXT:    vprotb %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vprotd %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vprotq %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vprotw %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vprotb (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotd (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotq (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotw (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotb %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotd %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotq %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotw %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vprotb $7, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vprotd $7, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vprotq $7, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vprotw $7, %xmm0, %xmm0 # sched: [2:0.50]
+; BDVER12-NEXT:    vprotb $7, (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER12-NEXT:    vprotd $7, (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER12-NEXT:    vprotq $7, (%rdi), %xmm0 # sched: [7:0.50]
+; BDVER12-NEXT:    vprotw $7, (%rdi), %xmm0 # sched: [7:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vprot:
 ; BDVER3:       # %bb.0:
@@ -1687,20 +1687,20 @@
 ; BDVER12-LABEL: test_vpsha:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpshab %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshad %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshaq %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshaw %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshab (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshad (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshaq (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshaw (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshab %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshad %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshaq %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshaw %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER12-NEXT:    vpshab %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshad %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshaq %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshaw %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshab (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshad (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshaq (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshaw (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshab %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshad %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshaq %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshaw %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpsha:
 ; BDVER3:       # %bb.0:
@@ -1763,20 +1763,20 @@
 ; BDVER12-LABEL: test_vpshl:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
-; BDVER12-NEXT:    vpshlb %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshld %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshlq %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshlw %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
-; BDVER12-NEXT:    vpshlb (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshld (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshlq (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshlw (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshlb %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshld %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshlq %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
-; BDVER12-NEXT:    vpshlw %xmm0, (%rdi), %xmm0 # sched: [7:1.00]
+; BDVER12-NEXT:    vpshlb %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshld %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshlq %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshlw %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
+; BDVER12-NEXT:    vpshlb (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshld (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshlq (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshlw (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshlb %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshld %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshlq %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
+; BDVER12-NEXT:    vpshlw %xmm0, (%rdi), %xmm0 # sched: [8:0.50]
 ; BDVER12-NEXT:    #NO_APP
-; BDVER12-NEXT:    retq # sched: [1:1.00]
+; BDVER12-NEXT:    retq # sched: [5:1.00]
 ;
 ; BDVER3-LABEL: test_vpshl:
 ; BDVER3:       # %bb.0:
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/add-sequence.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/add-sequence.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/add-sequence.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1000 -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1000 -timeline < %s | FileCheck %s
 
 add %eax, %ecx
 add %esi, %eax
@@ -7,13 +7,13 @@
 
 # CHECK:      Iterations:        1000
 # CHECK-NEXT: Instructions:      3000
-# CHECK-NEXT: Total Cycles:      1004
+# CHECK-NEXT: Total Cycles:      1504
 # CHECK-NEXT: Total uOps:        3000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    2.99
-# CHECK-NEXT: IPC:               2.99
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.99
+# CHECK-NEXT: IPC:               1.99
+# CHECK-NEXT: Block RThroughput: 1.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -24,64 +24,76 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        addl	%eax, %ecx
-# CHECK-NEXT:  1      1     0.33                        addl	%esi, %eax
-# CHECK-NEXT:  1      1     0.33                        addl	%eax, %edx
+# CHECK-NEXT:  1      1     0.50                        addl	%eax, %ecx
+# CHECK-NEXT:  1      1     0.50                        addl	%esi, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	%eax, %edx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     1.50   1.50    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addl	%eax, %ecx
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     addl	%esi, %eax
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     addl	%eax, %edx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%esi, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%eax, %edx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123
+# CHECK-NEXT:                     01234567
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    .  .   addl	%eax, %ecx
-# CHECK-NEXT: [0,1]     DeER .    .  .   addl	%esi, %eax
-# CHECK-NEXT: [0,2]     D=eER.    .  .   addl	%eax, %edx
-# CHECK-NEXT: [1,0]     D=eER.    .  .   addl	%eax, %ecx
-# CHECK-NEXT: [1,1]     .DeER.    .  .   addl	%esi, %eax
-# CHECK-NEXT: [1,2]     .D=eER    .  .   addl	%eax, %edx
-# CHECK-NEXT: [2,0]     .D=eER    .  .   addl	%eax, %ecx
-# CHECK-NEXT: [2,1]     .D=eER    .  .   addl	%esi, %eax
-# CHECK-NEXT: [2,2]     . D=eER   .  .   addl	%eax, %edx
-# CHECK-NEXT: [3,0]     . D=eER   .  .   addl	%eax, %ecx
-# CHECK-NEXT: [3,1]     . D=eER   .  .   addl	%esi, %eax
-# CHECK-NEXT: [3,2]     . D==eER  .  .   addl	%eax, %edx
-# CHECK-NEXT: [4,0]     .  D=eER  .  .   addl	%eax, %ecx
-# CHECK-NEXT: [4,1]     .  D=eER  .  .   addl	%esi, %eax
-# CHECK-NEXT: [4,2]     .  D==eER .  .   addl	%eax, %edx
-# CHECK-NEXT: [5,0]     .  D==eER .  .   addl	%eax, %ecx
-# CHECK-NEXT: [5,1]     .   D=eER .  .   addl	%esi, %eax
-# CHECK-NEXT: [5,2]     .   D==eER.  .   addl	%eax, %edx
-# CHECK-NEXT: [6,0]     .   D==eER.  .   addl	%eax, %ecx
-# CHECK-NEXT: [6,1]     .   D==eER.  .   addl	%esi, %eax
-# CHECK-NEXT: [6,2]     .    D==eER  .   addl	%eax, %edx
-# CHECK-NEXT: [7,0]     .    D==eER  .   addl	%eax, %ecx
-# CHECK-NEXT: [7,1]     .    D==eER  .   addl	%esi, %eax
-# CHECK-NEXT: [7,2]     .    D===eER .   addl	%eax, %edx
-# CHECK-NEXT: [8,0]     .    .D==eER .   addl	%eax, %ecx
-# CHECK-NEXT: [8,1]     .    .D==eER .   addl	%esi, %eax
-# CHECK-NEXT: [8,2]     .    .D===eER.   addl	%eax, %edx
-# CHECK-NEXT: [9,0]     .    .D===eER.   addl	%eax, %ecx
-# CHECK-NEXT: [9,1]     .    . D==eER.   addl	%esi, %eax
-# CHECK-NEXT: [9,2]     .    . D===eER   addl	%eax, %edx
+# CHECK:      [0,0]     DeER .    .    . .   addl	%eax, %ecx
+# CHECK-NEXT: [0,1]     DeER .    .    . .   addl	%esi, %eax
+# CHECK-NEXT: [0,2]     D=eER.    .    . .   addl	%eax, %edx
+# CHECK-NEXT: [1,0]     D==eER    .    . .   addl	%eax, %ecx
+# CHECK-NEXT: [1,1]     .DeE-R    .    . .   addl	%esi, %eax
+# CHECK-NEXT: [1,2]     .D=eER    .    . .   addl	%eax, %edx
+# CHECK-NEXT: [2,0]     .D==eER   .    . .   addl	%eax, %ecx
+# CHECK-NEXT: [2,1]     .D==eER   .    . .   addl	%esi, %eax
+# CHECK-NEXT: [2,2]     . D==eER  .    . .   addl	%eax, %edx
+# CHECK-NEXT: [3,0]     . D===eER .    . .   addl	%eax, %ecx
+# CHECK-NEXT: [3,1]     . D==eE-R .    . .   addl	%esi, %eax
+# CHECK-NEXT: [3,2]     . D===eER .    . .   addl	%eax, %edx
+# CHECK-NEXT: [4,0]     .  D===eER.    . .   addl	%eax, %ecx
+# CHECK-NEXT: [4,1]     .  D===eER.    . .   addl	%esi, %eax
+# CHECK-NEXT: [4,2]     .  D====eER    . .   addl	%eax, %edx
+# CHECK-NEXT: [5,0]     .  D=====eER   . .   addl	%eax, %ecx
+# CHECK-NEXT: [5,1]     .   D===eE-R   . .   addl	%esi, %eax
+# CHECK-NEXT: [5,2]     .   D====eER   . .   addl	%eax, %edx
+# CHECK-NEXT: [6,0]     .   D=====eER  . .   addl	%eax, %ecx
+# CHECK-NEXT: [6,1]     .   D=====eER  . .   addl	%esi, %eax
+# CHECK-NEXT: [6,2]     .    D=====eER . .   addl	%eax, %edx
+# CHECK-NEXT: [7,0]     .    D======eER. .   addl	%eax, %ecx
+# CHECK-NEXT: [7,1]     .    D=====eE-R. .   addl	%esi, %eax
+# CHECK-NEXT: [7,2]     .    D======eER. .   addl	%eax, %edx
+# CHECK-NEXT: [8,0]     .    .D======eER .   addl	%eax, %ecx
+# CHECK-NEXT: [8,1]     .    .D======eER .   addl	%esi, %eax
+# CHECK-NEXT: [8,2]     .    .D=======eER.   addl	%eax, %edx
+# CHECK-NEXT: [9,0]     .    .D========eER   addl	%eax, %ecx
+# CHECK-NEXT: [9,1]     .    . D======eE-R   addl	%esi, %eax
+# CHECK-NEXT: [9,2]     .    . D=======eER   addl	%eax, %edx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -90,6 +102,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     10    2.5    0.1    0.0       addl	%eax, %ecx
-# CHECK-NEXT: 1.     10    2.2    0.1    0.0       addl	%esi, %eax
-# CHECK-NEXT: 2.     10    3.0    0.0    0.0       addl	%eax, %edx
+# CHECK-NEXT: 0.     10    5.0    0.6    0.0       addl	%eax, %ecx
+# CHECK-NEXT: 1.     10    4.2    0.5    0.5       addl	%esi, %eax
+# CHECK-NEXT: 2.     10    5.0    0.0    0.0       addl	%eax, %edx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/clear-super-register-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/clear-super-register-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/clear-super-register-1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=100 -resource-pressure=false -timeline -timeline-max-iterations=2 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=100 -resource-pressure=false -timeline -timeline-max-iterations=2 < %s | FileCheck %s
 
 ## Sets register RAX.
 imulq $5, %rcx, %rax
@@ -15,13 +15,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      400
-# CHECK-NEXT: Total Cycles:      803
-# CHECK-NEXT: Total uOps:        400
+# CHECK-NEXT: Total Cycles:      702
+# CHECK-NEXT: Total uOps:        1000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.50
-# CHECK-NEXT: IPC:               0.50
-# CHECK-NEXT: Block RThroughput: 3.0
+# CHECK-NEXT: uOps Per Cycle:    1.42
+# CHECK-NEXT: IPC:               0.57
+# CHECK-NEXT: Block RThroughput: 4.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -32,23 +32,23 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        imulq	$5, %rcx, %rax
-# CHECK-NEXT:  1      3     1.00                        lzcntl	%ecx, %eax
-# CHECK-NEXT:  1      1     0.33                        andq	%rcx, %rax
-# CHECK-NEXT:  1      3     1.00                        bsfq	%rax, %rcx
+# CHECK-NEXT:  1      6     4.00                        imulq	$5, %rcx, %rax
+# CHECK-NEXT:  2      2     0.50                        lzcntl	%ecx, %eax
+# CHECK-NEXT:  1      1     0.50                        andq	%rcx, %rax
+# CHECK-NEXT:  6      3     2.00                        bsfq	%rax, %rcx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     012345678
+# CHECK-NEXT:                     012345
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER    .    .  .   imulq	$5, %rcx, %rax
-# CHECK-NEXT: [0,1]     D=eeeER   .    .  .   lzcntl	%ecx, %eax
-# CHECK-NEXT: [0,2]     D====eER  .    .  .   andq	%rcx, %rax
-# CHECK-NEXT: [0,3]     D=====eeeER    .  .   bsfq	%rax, %rcx
-# CHECK-NEXT: [1,0]     .D=======eeeER .  .   imulq	$5, %rcx, %rax
-# CHECK-NEXT: [1,1]     .D========eeeER.  .   lzcntl	%ecx, %eax
-# CHECK-NEXT: [1,2]     .D===========eER  .   andq	%rcx, %rax
-# CHECK-NEXT: [1,3]     .D============eeeER   bsfq	%rax, %rcx
+# CHECK:      [0,0]     DeeeeeeER .    .   imulq	$5, %rcx, %rax
+# CHECK-NEXT: [0,1]     DeeE----R .    .   lzcntl	%ecx, %eax
+# CHECK-NEXT: [0,2]     D==eE---R .    .   andq	%rcx, %rax
+# CHECK-NEXT: [0,3]     .D==eeeER .    .   bsfq	%rax, %rcx
+# CHECK-NEXT: [1,0]     . D====eeeeeeER.   imulq	$5, %rcx, %rax
+# CHECK-NEXT: [1,1]     .  D====eeE---R.   lzcntl	%ecx, %eax
+# CHECK-NEXT: [1,2]     .  D======eE--R.   andq	%rcx, %rax
+# CHECK-NEXT: [1,3]     .   D======eeeER   bsfq	%rax, %rcx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -57,7 +57,7 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     4.5    0.5    0.0       imulq	$5, %rcx, %rax
-# CHECK-NEXT: 1.     2     5.5    1.5    0.0       lzcntl	%ecx, %eax
-# CHECK-NEXT: 2.     2     8.5    0.0    0.0       andq	%rcx, %rax
-# CHECK-NEXT: 3.     2     9.5    0.0    0.0       bsfq	%rax, %rcx
+# CHECK-NEXT: 0.     2     3.0    0.5    0.0       imulq	$5, %rcx, %rax
+# CHECK-NEXT: 1.     2     3.0    1.0    3.5       lzcntl	%ecx, %eax
+# CHECK-NEXT: 2.     2     5.0    0.0    2.5       andq	%rcx, %rax
+# CHECK-NEXT: 3.     2     5.0    0.0    0.0       bsfq	%rax, %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/clear-super-register-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/clear-super-register-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/clear-super-register-2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=100 -resource-pressure=false -timeline -timeline-max-iterations=2 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=100 -resource-pressure=false -timeline -timeline-max-iterations=2 < %s | FileCheck %s
 
 # In this test, the VDIVPS takes 38 cycles to write to register YMM3.  The first
 # VADDPS does not depend on the VDIVPS (the WAW dependency is eliminated at
@@ -33,13 +33,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      1800
-# CHECK-NEXT: Total Cycles:      2804
-# CHECK-NEXT: Total uOps:        2000
+# CHECK-NEXT: Total Cycles:      4003
+# CHECK-NEXT: Total uOps:        3400
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.71
-# CHECK-NEXT: IPC:               0.64
-# CHECK-NEXT: Block RThroughput: 28.0
+# CHECK-NEXT: uOps Per Cycle:    0.85
+# CHECK-NEXT: IPC:               0.45
+# CHECK-NEXT: Block RThroughput: 31.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -50,65 +50,63 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  3      29    28.00                       vdivps	%ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT:  1      1     1.00                        vandps	%xmm4, %xmm1, %xmm0
+# CHECK-NEXT:  2      9     19.00                       vdivps	%ymm0, %ymm1, %ymm3
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT:  1      2     0.50                        vandps	%xmm4, %xmm1, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456789          0123456789
-# CHECK-NEXT: Index     0123456789          0123456789          0123456789
+# CHECK-NEXT:                     0123456789          0123456789          0123456789          0123456789
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER   .    .    .    .    .   .   vdivps	%ymm0, %ymm1, %ymm3
-# CHECK-NEXT: [0,1]     DeeeE--------------------------R   .    .    .    .    .   .   vaddps	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT: [0,2]     .D==eeeE-----------------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,3]     .D===eeeE----------------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,4]     .D====eeeE---------------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,5]     .D=====eeeE--------------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,6]     . D=====eeeE-------------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,7]     . D======eeeE------------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,8]     . D=======eeeE-----------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,9]     . D========eeeE----------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,10]    .  D========eeeE---------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,11]    .  D=========eeeE--------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,12]    .  D==========eeeE-------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,13]    .  D===========eeeE------------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,14]    .   D===========eeeE-----------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,15]    .   D============eeeE----------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,16]    .   D=============eeeE---------R   .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [0,17]    .   D================eE--------R   .    .    .    .    .   .   vandps	%xmm4, %xmm1, %xmm0
-# CHECK-NEXT: [1,0]     .    D=======================eeeeeeeeeeeeeeeeeeeeeeeeeeeeeER   vdivps	%ymm0, %ymm1, %ymm3
-# CHECK-NEXT: [1,1]     .    D================eeeE---------------------------------R   vaddps	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT: [1,2]     .    .D==================eeeE------------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,3]     .    .D===================eeeE-----------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,4]     .    .D====================eeeE----------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,5]     .    .D=====================eeeE---------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,6]     .    . D=====================eeeE--------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,7]     .    . D======================eeeE-------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,8]     .    . D=======================eeeE------------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,9]     .    . D========================eeeE-----------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,10]    .    .  D========================eeeE----------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,11]    .    .  D=========================eeeE---------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,12]    .    .  D==========================eeeE--------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,13]    .    .  D===========================eeeE-------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,14]    .    .   D===========================eeeE------------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,15]    .    .   D============================eeeE-----------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,16]    .    .   D=============================eeeE----------------R   vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: [1,17]    .    .   D================================eE---------------R   vandps	%xmm4, %xmm1, %xmm0
+# CHECK:      [0,0]     DeeeeeeeeeER   .    .    .    .    .    .    .    .    .    .    .    .    .   .   vdivps	%ymm0, %ymm1, %ymm3
+# CHECK-NEXT: [0,1]     DeeeeeE----R   .    .    .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: [0,2]     .D====eeeeeER  .    .    .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,3]     .D======eeeeeER.    .    .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,4]     . D=======eeeeeER   .    .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,5]     . D=========eeeeeER .    .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,6]     .  D==========eeeeeER    .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,7]     .  D============eeeeeER  .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,8]     .   D=============eeeeeER.    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,9]     .   D===============eeeeeER   .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,10]    .    D================eeeeeER .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,11]    .    D==================eeeeeER    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,12]    .    .D===================eeeeeER  .    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,13]    .    .D=====================eeeeeER.    .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,14]    .    . D======================eeeeeER   .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,15]    .    . D========================eeeeeER .    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,16]    .    .  D=========================eeeeeER    .    .    .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [0,17]    .    .  D==============================eeER  .    .    .    .    .    .    .   .   vandps	%xmm4, %xmm1, %xmm0
+# CHECK-NEXT: [1,0]     .    .   D===============================eeeeeeeeeER   .    .    .    .    .   .   vdivps	%ymm0, %ymm1, %ymm3
+# CHECK-NEXT: [1,1]     .    .   D===============================eeeeeE----R   .    .    .    .    .   .   vaddps	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: [1,2]     .    .    D===================================eeeeeER  .    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,3]     .    .    D=====================================eeeeeER.    .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,4]     .    .    .D======================================eeeeeER   .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,5]     .    .    .D========================================eeeeeER .    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,6]     .    .    . D=========================================eeeeeER    .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,7]     .    .    . D===========================================eeeeeER  .    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,8]     .    .    .  D============================================eeeeeER.    .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,9]     .    .    .  D==============================================eeeeeER   .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,10]    .    .    .   D===============================================eeeeeER .    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,11]    .    .    .   D=================================================eeeeeER    .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,12]    .    .    .    D==================================================eeeeeER  .   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,13]    .    .    .    D====================================================eeeeeER.   .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,14]    .    .    .    .D=====================================================eeeeeER  .   vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: [1,15]    .    .    .    .D=======================================================eeeeeER.   vaddps	%ymm3, %ymm1, %ymm4
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -117,21 +115,21 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     12.5   4.0    0.0       vdivps	%ymm0, %ymm1, %ymm3
-# CHECK-NEXT: 1.     2     9.0    0.5    29.5      vaddps	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT: 2.     2     11.0   0.0    26.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 3.     2     12.0   1.0    25.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 4.     2     13.0   2.0    24.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 5.     2     14.0   3.0    23.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 6.     2     14.0   4.0    22.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 7.     2     15.0   5.0    21.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 8.     2     16.0   6.0    20.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 9.     2     17.0   7.0    19.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 10.    2     17.0   8.0    18.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 11.    2     18.0   9.0    17.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 12.    2     19.0   10.0   16.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 13.    2     20.0   11.0   15.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 14.    2     20.0   12.0   14.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 15.    2     21.0   13.0   13.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 16.    2     22.0   14.0   12.5      vaddps	%ymm3, %ymm1, %ymm4
-# CHECK-NEXT: 17.    2     25.0   0.0    11.5      vandps	%xmm4, %xmm1, %xmm0
+# CHECK-NEXT: 0.     2     16.5   0.5    0.0       vdivps	%ymm0, %ymm1, %ymm3
+# CHECK-NEXT: 1.     2     16.5   0.5    4.0       vaddps	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 2.     2     20.5   0.0    0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 3.     2     22.5   2.0    0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 4.     2     23.5   4.0    0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 5.     2     25.5   6.0    0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 6.     2     26.5   8.0    0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 7.     2     28.5   10.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 8.     2     29.5   12.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 9.     2     31.5   14.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 10.    2     32.5   16.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 11.    2     34.5   18.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 12.    2     35.5   20.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 13.    2     37.5   22.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 14.    2     38.5   23.5   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 15.    2     40.5   25.5   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 16.    2     41.5   27.0   0.0       vaddps	%ymm3, %ymm1, %ymm4
+# CHECK-NEXT: 17.    2     46.5   0.0    0.0       vandps	%xmm4, %xmm1, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-cmp.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-cmp.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-cmp.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
 
 # Perf stat reports an IPC of 1.97 for this block of code.
 
@@ -11,13 +11,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      3000
-# CHECK-NEXT: Total Cycles:      4503
-# CHECK-NEXT: Total uOps:        4500
+# CHECK-NEXT: Total Cycles:      1504
+# CHECK-NEXT: Total uOps:        3000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               0.67
-# CHECK-NEXT: Block RThroughput: 0.8
+# CHECK-NEXT: uOps Per Cycle:    1.99
+# CHECK-NEXT: IPC:               1.99
+# CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -28,38 +28,49 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        cmpl	%eax, %eax
-# CHECK-NEXT:  2      2     0.67                        cmovael	%ebx, %eax
+# CHECK-NEXT:  1      1     0.50                        cmpl	%eax, %eax
+# CHECK-NEXT:  1      1     0.50                        cmovael	%ebx, %eax
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     cmpl	%eax, %eax
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cmovael	%ebx, %eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	%eax, %eax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovael	%ebx, %eax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01
-# CHECK-NEXT: Index     0123456789
+# CHECK-NEXT: Index     0123456
 
-# CHECK:      [0,0]     DeER .    ..   cmpl	%eax, %eax
-# CHECK-NEXT: [0,1]     D=eeER    ..   cmovael	%ebx, %eax
-# CHECK-NEXT: [1,0]     D===eER   ..   cmpl	%eax, %eax
-# CHECK-NEXT: [1,1]     .D===eeER ..   cmovael	%ebx, %eax
-# CHECK-NEXT: [2,0]     .D=====eER..   cmpl	%eax, %eax
-# CHECK-NEXT: [2,1]     . D=====eeER   cmovael	%ebx, %eax
+# CHECK:      [0,0]     DeER ..   cmpl	%eax, %eax
+# CHECK-NEXT: [0,1]     D=eER..   cmovael	%ebx, %eax
+# CHECK-NEXT: [1,0]     DeE-R..   cmpl	%eax, %eax
+# CHECK-NEXT: [1,1]     D==eER.   cmovael	%ebx, %eax
+# CHECK-NEXT: [2,0]     .DeE-R.   cmpl	%eax, %eax
+# CHECK-NEXT: [2,1]     .D==eER   cmovael	%ebx, %eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -68,5 +79,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     3.7    0.3    0.0       cmpl	%eax, %eax
-# CHECK-NEXT: 1.     3     4.0    0.0    0.0       cmovael	%ebx, %eax
+# CHECK-NEXT: 0.     3     1.0    1.0    0.7       cmpl	%eax, %eax
+# CHECK-NEXT: 1.     3     2.7    0.0    0.0       cmovael	%ebx, %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpeq.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpeq.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpeq.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
 
 # perf stat reports an IPC of 2.00 for this block of code.
 
@@ -14,12 +14,12 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      6000
-# CHECK-NEXT: Total Cycles:      6003
+# CHECK-NEXT: Total Cycles:      3005
 # CHECK-NEXT: Total uOps:        6000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               1.00
+# CHECK-NEXT: uOps Per Cycle:    2.00
+# CHECK-NEXT: IPC:               2.00
 # CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
@@ -31,48 +31,60 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqq	%xmm3, %xmm3, %xmm0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     2.00    -     2.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     2.00   2.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpcmpeqb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpcmpeqw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpcmpeqd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpcmpeqq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     vpcmpeqb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -     vpcmpeqw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -      -     vpcmpeqd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     vpcmpeqq	%xmm3, %xmm3, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
+# CHECK-NEXT:                     0
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    .   .   vpcmpeqb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [0,1]     D=eER.    .   .   vpcmpeqw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [0,2]     D==eER    .   .   vpcmpeqd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [0,3]     D===eER   .   .   vpcmpeqq	%xmm3, %xmm3, %xmm0
-# CHECK-NEXT: [1,0]     .D===eER  .   .   vpcmpeqb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [1,1]     .D====eER .   .   vpcmpeqw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [1,2]     .D=====eER.   .   vpcmpeqd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [1,3]     .D======eER   .   vpcmpeqq	%xmm3, %xmm3, %xmm0
-# CHECK-NEXT: [2,0]     . D======eER  .   vpcmpeqb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [2,1]     . D=======eER .   vpcmpeqw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [2,2]     . D========eER.   vpcmpeqd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [2,3]     . D=========eER   vpcmpeqq	%xmm3, %xmm3, %xmm0
+# CHECK:      [0,0]     DeeER.    .   vpcmpeqb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [0,1]     D=eeER    .   vpcmpeqw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [0,2]     DeeE-R    .   vpcmpeqd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [0,3]     D==eeER   .   vpcmpeqq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: [1,0]     .DeeE-R   .   vpcmpeqb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [1,1]     .D==eeER  .   vpcmpeqw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [1,2]     .D=eeE-R  .   vpcmpeqd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [1,3]     .D===eeER .   vpcmpeqq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: [2,0]     . D=eeE-R .   vpcmpeqb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [2,1]     . D===eeER.   vpcmpeqw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [2,2]     . D==eeE-R.   vpcmpeqd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [2,3]     . D====eeER   vpcmpeqq	%xmm3, %xmm3, %xmm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -81,7 +93,7 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     4.0    0.3    0.0       vpcmpeqb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: 1.     3     5.0    0.0    0.0       vpcmpeqw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: 2.     3     6.0    0.0    0.0       vpcmpeqd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: 3.     3     7.0    0.0    0.0       vpcmpeqq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: 0.     3     1.3    1.3    0.7       vpcmpeqb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: 1.     3     3.0    3.0    0.0       vpcmpeqw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: 2.     3     2.0    2.0    1.0       vpcmpeqd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: 3.     3     4.0    0.0    0.0       vpcmpeqq	%xmm3, %xmm3, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpgt.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpgt.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpgt.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
 
 # perf stat reports an IPC of 2.00 for this block of code.
 
@@ -15,12 +15,12 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      6000
-# CHECK-NEXT: Total Cycles:      1501
+# CHECK-NEXT: Total Cycles:      1504
 # CHECK-NEXT: Total uOps:        6000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    4.00
-# CHECK-NEXT: IPC:               4.00
+# CHECK-NEXT: uOps Per Cycle:    3.99
+# CHECK-NEXT: IPC:               3.99
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
@@ -35,44 +35,56 @@
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtb	%xmm0, %xmm0, %xmm1
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtw	%xmm1, %xmm1, %xmm2
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT:  1      0     0.25                        vpcmpgtq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtq	%xmm3, %xmm3, %xmm0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -      -      -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     vpcmpgtq	%xmm3, %xmm3, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     0123
+# CHECK-NEXT: Index     0123456
 
-# CHECK:      [0,0]     DR .   vpcmpgtb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [0,1]     DR .   vpcmpgtw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [0,2]     DR .   vpcmpgtd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [0,3]     DR .   vpcmpgtq	%xmm3, %xmm3, %xmm0
-# CHECK-NEXT: [1,0]     .DR.   vpcmpgtb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [1,1]     .DR.   vpcmpgtw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [1,2]     .DR.   vpcmpgtd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [1,3]     .DR.   vpcmpgtq	%xmm3, %xmm3, %xmm0
-# CHECK-NEXT: [2,0]     . DR   vpcmpgtb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [2,1]     . DR   vpcmpgtw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [2,2]     . DR   vpcmpgtd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [2,3]     . DR   vpcmpgtq	%xmm3, %xmm3, %xmm0
+# CHECK:      [0,0]     DR   ..   vpcmpgtb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [0,1]     DR   ..   vpcmpgtw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [0,2]     DR   ..   vpcmpgtd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [0,3]     DeeER..   vpcmpgtq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: [1,0]     .D--R..   vpcmpgtb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [1,1]     .D--R..   vpcmpgtw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [1,2]     .D--R..   vpcmpgtd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [1,3]     .DeeER.   vpcmpgtq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: [2,0]     . D--R.   vpcmpgtb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [2,1]     . D--R.   vpcmpgtw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [2,2]     . D--R.   vpcmpgtd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [2,3]     . DeeER   vpcmpgtq	%xmm3, %xmm3, %xmm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -81,7 +93,7 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     0.0    0.0    0.0       vpcmpgtb	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: 1.     3     0.0    0.0    0.0       vpcmpgtw	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: 2.     3     0.0    0.0    0.0       vpcmpgtd	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: 3.     3     0.0    0.0    0.0       vpcmpgtq	%xmm3, %xmm3, %xmm0
+# CHECK-NEXT: 0.     3     0.0    0.0    1.3       vpcmpgtb	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: 1.     3     0.0    0.0    1.3       vpcmpgtw	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: 2.     3     0.0    0.0    1.3       vpcmpgtd	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: 3.     3     1.0    1.0    0.0       vpcmpgtq	%xmm3, %xmm3, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-sbb-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-sbb-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-sbb-1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
 
 # perf stat reports an IPC of 1.00 for this code block.
 
@@ -12,13 +12,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      3000
-# CHECK-NEXT: Total Cycles:      6003
-# CHECK-NEXT: Total uOps:        6000
+# CHECK-NEXT: Total Cycles:      3003
+# CHECK-NEXT: Total uOps:        3000
 
 # CHECK:      Dispatch Width:    4
 # CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               0.50
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: IPC:               1.00
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -29,38 +29,49 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      2     0.67                        sbbl	%edx, %edx
-# CHECK-NEXT:  2      2     0.67                        sbbl	%eax, %eax
+# CHECK-NEXT:  1      1     1.00                        sbbl	%edx, %edx
+# CHECK-NEXT:  1      1     1.00                        sbbl	%eax, %eax
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.33   1.33    -     1.33    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67    -      -     sbbl	%edx, %edx
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67    -      -     sbbl	%eax, %eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -     2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	%edx, %edx
+# CHECK-NEXT:  -      -      -      -      -     2.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	%eax, %eax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
-# CHECK-NEXT: Index     0123456789
+# CHECK-NEXT: Index     012345678
 
-# CHECK:      [0,0]     DeeER.    .   .   sbbl	%edx, %edx
-# CHECK-NEXT: [0,1]     D==eeER   .   .   sbbl	%eax, %eax
-# CHECK-NEXT: [1,0]     .D===eeER .   .   sbbl	%edx, %edx
-# CHECK-NEXT: [1,1]     .D=====eeER   .   sbbl	%eax, %eax
-# CHECK-NEXT: [2,0]     . D======eeER .   sbbl	%edx, %edx
-# CHECK-NEXT: [2,1]     . D========eeER   sbbl	%eax, %eax
+# CHECK:      [0,0]     DeER .  .   sbbl	%edx, %edx
+# CHECK-NEXT: [0,1]     D=eER.  .   sbbl	%eax, %eax
+# CHECK-NEXT: [1,0]     D==eER  .   sbbl	%edx, %edx
+# CHECK-NEXT: [1,1]     D===eER .   sbbl	%eax, %eax
+# CHECK-NEXT: [2,0]     .D===eER.   sbbl	%edx, %edx
+# CHECK-NEXT: [2,1]     .D====eER   sbbl	%eax, %eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -69,5 +80,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     4.0    0.3    0.0       sbbl	%edx, %edx
-# CHECK-NEXT: 1.     3     6.0    0.0    0.0       sbbl	%eax, %eax
+# CHECK-NEXT: 0.     3     2.7    0.3    0.0       sbbl	%edx, %edx
+# CHECK-NEXT: 1.     3     3.7    0.0    0.0       sbbl	%eax, %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-sbb-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-sbb-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependency-breaking-sbb-2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
 
 # perf stat reports a throughput of 1.51 IPC for this block of code.
 
@@ -13,13 +13,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      4500
-# CHECK-NEXT: Total Cycles:      7503
-# CHECK-NEXT: Total uOps:        6000
+# CHECK-NEXT: Total Cycles:      3006
+# CHECK-NEXT: Total uOps:        4500
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.80
-# CHECK-NEXT: IPC:               0.60
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.50
+# CHECK-NEXT: IPC:               1.50
+# CHECK-NEXT: Block RThroughput: 1.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -30,43 +30,55 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        imull	%edx, %eax
-# CHECK-NEXT:  1      1     0.33                        addl	%edx, %edx
-# CHECK-NEXT:  2      2     0.67                        sbbl	%eax, %eax
+# CHECK-NEXT:  1      4     1.00                        imull	%edx, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	%edx, %edx
+# CHECK-NEXT:  1      1     1.00                        sbbl	%eax, %eax
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.33   1.33    -     1.33    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imull	%edx, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.34    -      -     addl	%edx, %edx
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     sbbl	%eax, %eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	%edx, %eax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%edx, %edx
+# CHECK-NEXT:  -      -      -      -      -     2.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	%eax, %eax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234567
+# CHECK-NEXT:                     01
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER    .    . .   imull	%edx, %eax
-# CHECK-NEXT: [0,1]     DeE--R    .    . .   addl	%edx, %edx
-# CHECK-NEXT: [0,2]     D===eeER  .    . .   sbbl	%eax, %eax
-# CHECK-NEXT: [1,0]     .D====eeeER    . .   imull	%edx, %eax
-# CHECK-NEXT: [1,1]     .DeE------R    . .   addl	%edx, %edx
-# CHECK-NEXT: [1,2]     .D=======eeER  . .   sbbl	%eax, %eax
-# CHECK-NEXT: [2,0]     . D========eeeER .   imull	%edx, %eax
-# CHECK-NEXT: [2,1]     . DeE----------R .   addl	%edx, %edx
-# CHECK-NEXT: [2,2]     . D===========eeER   sbbl	%eax, %eax
+# CHECK:      [0,0]     D=eeeeER  ..   imull	%edx, %eax
+# CHECK-NEXT: [0,1]     DeE----R  ..   addl	%edx, %edx
+# CHECK-NEXT: [0,2]     D==eE--R  ..   sbbl	%eax, %eax
+# CHECK-NEXT: [1,0]     D===eeeeER..   imull	%edx, %eax
+# CHECK-NEXT: [1,1]     .DeE-----R..   addl	%edx, %edx
+# CHECK-NEXT: [1,2]     .D===eE--R..   sbbl	%eax, %eax
+# CHECK-NEXT: [2,0]     .D====eeeeER   imull	%edx, %eax
+# CHECK-NEXT: [2,1]     .D=eE------R   addl	%edx, %edx
+# CHECK-NEXT: [2,2]     . D====eE--R   sbbl	%eax, %eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -75,6 +87,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     5.0    0.3    0.0       imull	%edx, %eax
-# CHECK-NEXT: 1.     3     1.0    0.3    6.0       addl	%edx, %edx
-# CHECK-NEXT: 2.     3     8.0    0.0    0.0       sbbl	%eax, %eax
+# CHECK-NEXT: 0.     3     3.7    0.7    0.0       imull	%edx, %eax
+# CHECK-NEXT: 1.     3     1.3    0.3    5.0       addl	%edx, %edx
+# CHECK-NEXT: 2.     3     4.0    2.0    2.0       sbbl	%eax, %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependent-pmuld-paddd.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependent-pmuld-paddd.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dependent-pmuld-paddd.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=500 -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=500 -timeline < %s | FileCheck %s
 
 vpmuld %xmm0, %xmm0, %xmm1
 vpaddd %xmm1, %xmm1, %xmm0
@@ -7,7 +7,7 @@
 
 # CHECK:      Iterations:        500
 # CHECK-NEXT: Instructions:      1500
-# CHECK-NEXT: Total Cycles:      3004
+# CHECK-NEXT: Total Cycles:      3005
 # CHECK-NEXT: Total uOps:        1500
 
 # CHECK:      Dispatch Width:    4
@@ -24,64 +24,76 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      5     1.00                        vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  1      1     0.50                        vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT:  1      1     0.50                        vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  1      2     0.50                        vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT:  1      2     0.50                        vpaddd	%xmm0, %xmm0, %xmm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     2.00   1.00    -     1.50   1.50    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     vpaddd	%xmm0, %xmm0, %xmm3
 
 # CHECK:      Timeline view:
 # CHECK-NEXT:                     0123456789          0123456789          0123456789
-# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789          01234
 
-# CHECK:      [0,0]     DeeeeeER  .    .    .    .    .    .    .    .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [0,1]     D=====eER .    .    .    .    .    .    .    .    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [0,2]     D======eER.    .    .    .    .    .    .    .    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [1,0]     D======eeeeeER .    .    .    .    .    .    .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [1,1]     .D==========eER.    .    .    .    .    .    .    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [1,2]     .D===========eER    .    .    .    .    .    .    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [2,0]     .D===========eeeeeER.    .    .    .    .    .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [2,1]     .D================eER    .    .    .    .    .    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [2,2]     . D================eER   .    .    .    .    .    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [3,0]     . D================eeeeeER    .    .    .    .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [3,1]     . D=====================eER   .    .    .    .    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [3,2]     . D======================eER  .    .    .    .    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [4,0]     .  D=====================eeeeeER   .    .    .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [4,1]     .  D==========================eER  .    .    .    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [4,2]     .  D===========================eER .    .    .    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [5,0]     .  D===========================eeeeeER  .    .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [5,1]     .   D===============================eER .    .    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [5,2]     .   D================================eER.    .    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [6,0]     .   D================================eeeeeER .    .    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [6,1]     .   D=====================================eER.    .    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [6,2]     .    D=====================================eER    .    .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [7,0]     .    D=====================================eeeeeER.    .    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [7,1]     .    D==========================================eER    .    .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [7,2]     .    D===========================================eER   .    .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [8,0]     .    .D==========================================eeeeeER    .  .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [8,1]     .    .D===============================================eER   .  .   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [8,2]     .    .D================================================eER  .  .   vpaddd	%xmm0, %xmm0, %xmm3
-# CHECK-NEXT: [9,0]     .    .D================================================eeeeeER .   vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [9,1]     .    . D====================================================eER.   vpaddd	%xmm1, %xmm1, %xmm0
-# CHECK-NEXT: [9,2]     .    . D=====================================================eER   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK:      [0,0]     DeeeeER   .    .    .    .    .    .    .    .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [0,1]     D====eeER .    .    .    .    .    .    .    .    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [0,2]     D======eeER    .    .    .    .    .    .    .    .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [1,0]     D======eeeeER  .    .    .    .    .    .    .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [1,1]     .D=========eeER.    .    .    .    .    .    .    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [1,2]     .D===========eeER   .    .    .    .    .    .    .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [2,0]     .D===========eeeeER .    .    .    .    .    .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [2,1]     .D===============eeER    .    .    .    .    .    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [2,2]     . D================eeER  .    .    .    .    .    .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [3,0]     . D================eeeeER.    .    .    .    .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [3,1]     . D====================eeER   .    .    .    .    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [3,2]     . D======================eeER .    .    .    .    .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [4,0]     .  D=====================eeeeER    .    .    .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [4,1]     .  D=========================eeER  .    .    .    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [4,2]     .  D===========================eeER.    .    .    .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [5,0]     .  D===========================eeeeER   .    .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [5,1]     .   D==============================eeER .    .    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [5,2]     .   D================================eeER    .    .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [6,0]     .   D================================eeeeER  .    .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [6,1]     .   D====================================eeER.    .    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [6,2]     .    D=====================================eeER   .    .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [7,0]     .    D=====================================eeeeER .    .    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [7,1]     .    D=========================================eeER    .    .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [7,2]     .    D===========================================eeER  .    .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [8,0]     .    .D==========================================eeeeER.    .   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [8,1]     .    .D==============================================eeER   .   .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [8,2]     .    .D================================================eeER .   .   vpaddd	%xmm0, %xmm0, %xmm3
+# CHECK-NEXT: [9,0]     .    .D================================================eeeeER   .   vpmuldq	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [9,1]     .    . D===================================================eeER .   vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: [9,2]     .    . D=====================================================eeER   vpaddd	%xmm0, %xmm0, %xmm3
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -91,5 +103,5 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     10    25.0   0.1    0.0       vpmuldq	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: 1.     10    29.7   0.0    0.0       vpaddd	%xmm1, %xmm1, %xmm0
+# CHECK-NEXT: 1.     10    28.7   0.0    0.0       vpaddd	%xmm1, %xmm1, %xmm0
 # CHECK-NEXT: 2.     10    30.5   0.0    0.0       vpaddd	%xmm0, %xmm0, %xmm3
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dot-product.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dot-product.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/dot-product.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=300 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=300 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 vmulps   %xmm0, %xmm1, %xmm2
 vhaddps  %xmm2, %xmm2, %xmm3
@@ -7,13 +7,13 @@
 
 # CHECK:      Iterations:        300
 # CHECK-NEXT: Instructions:      900
-# CHECK-NEXT: Total Cycles:      1211
+# CHECK-NEXT: Total Cycles:      627
 # CHECK-NEXT: Total uOps:        2100
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.73
-# CHECK-NEXT: IPC:               0.74
-# CHECK-NEXT: Block RThroughput: 4.0
+# CHECK-NEXT: uOps Per Cycle:    3.35
+# CHECK-NEXT: IPC:               1.44
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -25,42 +25,54 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      5     1.00                        vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vhaddps	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT:  3      5     2.00                        vhaddps	%xmm3, %xmm3, %xmm4
+# CHECK-NEXT:  3      11    1.00                        vhaddps	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT:  3      11    1.00                        vhaddps	%xmm3, %xmm3, %xmm4
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   2.00    -     4.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.49   1.51    -      -      -      -     2.00   1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhaddps	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhaddps	%xmm3, %xmm3, %xmm4
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhaddps	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.49   0.51    -      -      -      -     1.00    -      -      -      -      -     vhaddps	%xmm3, %xmm3, %xmm4
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789
-# CHECK-NEXT: Index     0123456789          012
+# CHECK-NEXT:                     0123456789          012
+# CHECK-NEXT: Index     0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeeeER  .    .    . .   vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [0,1]     D=====eeeeeER  .    . .   vhaddps	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [0,2]     .D==========eeeeeER . .   vhaddps	%xmm3, %xmm3, %xmm4
-# CHECK-NEXT: [1,0]     .DeeeeeE----------R . .   vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [1,1]     . D=====eeeeeE----R . .   vhaddps	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [1,2]     .  D==========eeeeeER .   vhaddps	%xmm3, %xmm3, %xmm4
-# CHECK-NEXT: [2,0]     .  DeeeeeE----------R .   vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [2,1]     .   D=====eeeeeE----R .   vhaddps	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: [2,2]     .    D==========eeeeeER   vhaddps	%xmm3, %xmm3, %xmm4
+# CHECK:      [0,0]     DeeeeeER  .    .    .    .    . .   vmulps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [0,1]     D=====eeeeeeeeeeeER .    .    . .   vhaddps	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [0,2]     .D===============eeeeeeeeeeeER. .   vhaddps	%xmm3, %xmm3, %xmm4
+# CHECK-NEXT: [1,0]     .DeeeeeE---------------------R. .   vmulps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [1,1]     . D====eeeeeeeeeeeE----------R. .   vhaddps	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [1,2]     .  D==============eeeeeeeeeeeER .   vhaddps	%xmm3, %xmm3, %xmm4
+# CHECK-NEXT: [2,0]     .  DeeeeeE--------------------R .   vmulps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [2,1]     .   D====eeeeeeeeeeeE---------R .   vhaddps	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: [2,2]     .    D==============eeeeeeeeeeeER   vhaddps	%xmm3, %xmm3, %xmm4
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -69,6 +81,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.0    1.0    6.7       vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: 1.     3     6.0    0.7    2.7       vhaddps	%xmm2, %xmm2, %xmm3
-# CHECK-NEXT: 2.     3     11.0   1.0    0.0       vhaddps	%xmm3, %xmm3, %xmm4
+# CHECK-NEXT: 0.     3     1.0    1.0    13.7      vmulps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 1.     3     5.3    0.0    6.3       vhaddps	%xmm2, %xmm2, %xmm3
+# CHECK-NEXT: 2.     3     15.3   0.0    0.0       vhaddps	%xmm3, %xmm3, %xmm4
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/hadd-read-after-ld-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/hadd-read-after-ld-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/hadd-read-after-ld-1.s
@@ -1,18 +1,18 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
 
 vshufps $0, %xmm0, %xmm1, %xmm1
 vhaddps (%rdi), %xmm1, %xmm2
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      15
+# CHECK-NEXT: Total Cycles:      20
 # CHECK-NEXT: Total uOps:        5
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.33
-# CHECK-NEXT: IPC:               0.13
-# CHECK-NEXT: Block RThroughput: 3.0
+# CHECK-NEXT: uOps Per Cycle:    0.25
+# CHECK-NEXT: IPC:               0.10
+# CHECK-NEXT: Block RThroughput: 1.3
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -23,15 +23,15 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     1.00                        vshufps	$0, %xmm0, %xmm1, %xmm1
-# CHECK-NEXT:  4      11    2.00    *                   vhaddps	(%rdi), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vshufps	$0, %xmm0, %xmm1, %xmm1
+# CHECK-NEXT:  4      16    1.00    *                   vhaddps	(%rdi), %xmm1, %xmm2
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
+# CHECK-NEXT:                     0123456789
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    .   .   vshufps	$0, %xmm0, %xmm1, %xmm1
-# CHECK-NEXT: [0,1]     .DeeeeeeeeeeeER   vhaddps	(%rdi), %xmm1, %xmm2
+# CHECK:      [0,0]     DeeER.    .    .   .   vshufps	$0, %xmm0, %xmm1, %xmm1
+# CHECK-NEXT: [0,1]     .DeeeeeeeeeeeeeeeeER   vhaddps	(%rdi), %xmm1, %xmm2
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/hadd-read-after-ld-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/hadd-read-after-ld-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/hadd-read-after-ld-2.s
@@ -1,18 +1,18 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
 
 vshufps $0, %xmm0, %xmm1, %xmm1
 vhaddps (%rdi), %ymm1, %ymm2
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      16
-# CHECK-NEXT: Total uOps:        5
+# CHECK-NEXT: Total Cycles:      20
+# CHECK-NEXT: Total uOps:        11
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.31
-# CHECK-NEXT: IPC:               0.13
-# CHECK-NEXT: Block RThroughput: 3.0
+# CHECK-NEXT: uOps Per Cycle:    0.55
+# CHECK-NEXT: IPC:               0.10
+# CHECK-NEXT: Block RThroughput: 2.8
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -23,15 +23,15 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     1.00                        vshufps	$0, %xmm0, %xmm1, %xmm1
-# CHECK-NEXT:  4      12    2.00    *                   vhaddps	(%rdi), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vshufps	$0, %xmm0, %xmm1, %xmm1
+# CHECK-NEXT:  10     16    2.00    *                   vhaddps	(%rdi), %ymm1, %ymm2
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     012345
+# CHECK-NEXT:                     0123456789
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    .    .   vshufps	$0, %xmm0, %xmm1, %xmm1
-# CHECK-NEXT: [0,1]     .DeeeeeeeeeeeeER   vhaddps	(%rdi), %ymm1, %ymm2
+# CHECK:      [0,0]     DeeER.    .    .   .   vshufps	$0, %xmm0, %xmm1, %xmm1
+# CHECK-NEXT: [0,1]     .DeeeeeeeeeeeeeeeeER   vhaddps	(%rdi), %ymm1, %ymm2
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/instruction-info-view.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/instruction-info-view.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/instruction-info-view.s
@@ -1,8 +1,8 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -resource-pressure=false -instruction-info=true < %s | FileCheck %s --check-prefix=ENABLED
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -resource-pressure=false -instruction-info=false < %s | FileCheck %s -check-prefix=DISABLED
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -resource-pressure=false -instruction-info < %s | FileCheck %s -check-prefix=ENABLED
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -resource-pressure=false < %s | FileCheck %s -check-prefix=ENABLED
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -resource-pressure=false -instruction-info=true < %s | FileCheck %s --check-prefix=ENABLED
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -resource-pressure=false -instruction-info=false < %s | FileCheck %s -check-prefix=DISABLED
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -resource-pressure=false -instruction-info < %s | FileCheck %s -check-prefix=ENABLED
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -resource-pressure=false < %s | FileCheck %s -check-prefix=ENABLED
 
 vmulps   %xmm0, %xmm1, %xmm2
 vhaddps  %xmm2, %xmm2, %xmm3
@@ -13,14 +13,14 @@
 
 # ENABLED:       Iterations:        100
 # ENABLED-NEXT:  Instructions:      300
-# ENABLED-NEXT:  Total Cycles:      414
+# ENABLED-NEXT:  Total Cycles:      228
 # ENABLED-NEXT:  Total uOps:        700
 
 
 # ENABLED:       Dispatch Width:    4
-# ENABLED-NEXT:  uOps Per Cycle:    1.69
-# ENABLED-NEXT:  IPC:               0.72
-# ENABLED-NEXT:  Block RThroughput: 4.0
+# ENABLED-NEXT:  uOps Per Cycle:    3.07
+# ENABLED-NEXT:  IPC:               1.32
+# ENABLED-NEXT:  Block RThroughput: 2.0
 
 # ENABLED:       Instruction Info:
 # ENABLED-NEXT:  [1]: #uOps
@@ -32,5 +32,5 @@
 
 # ENABLED:       [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # ENABLED-NEXT:   1      5     1.00                        vmulps	%xmm0, %xmm1, %xmm2
-# ENABLED-NEXT:   3      5     2.00                        vhaddps	%xmm2, %xmm2, %xmm3
-# ENABLED-NEXT:   3      5     2.00                        vhaddps	%xmm3, %xmm3, %xmm4
+# ENABLED-NEXT:   3      11    1.00                        vhaddps	%xmm2, %xmm2, %xmm3
+# ENABLED-NEXT:   3      11    1.00                        vhaddps	%xmm3, %xmm3, %xmm4
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/load-store-alias.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/load-store-alias.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/load-store-alias.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=100 -timeline -timeline-max-iterations=1 -noalias=false < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=100 -timeline -timeline-max-iterations=1 -noalias=false < %s | FileCheck %s
 
 vmovaps (%rsi), %xmm0
 vmovaps %xmm0, (%rdi)
@@ -12,12 +12,12 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      800
-# CHECK-NEXT: Total Cycles:      2803
+# CHECK-NEXT: Total Cycles:      2403
 # CHECK-NEXT: Total uOps:        800
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.29
-# CHECK-NEXT: IPC:               0.29
+# CHECK-NEXT: uOps Per Cycle:    0.33
+# CHECK-NEXT: IPC:               0.33
 # CHECK-NEXT: Block RThroughput: 4.0
 
 # CHECK:      Instruction Info:
@@ -29,52 +29,64 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	16(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	32(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	48(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, 48(%rdi)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -      -     4.00    -      -     8.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -     8.00    -      -      -      -      -      -      -     4.00    -      -      -     4.00   3.99   4.01    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -     1.00   vmovaps	(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -      -     1.00   vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT:  -      -      -      -      -      -      -     1.00   vmovaps	16(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -      -     1.00   vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT:  -      -      -      -      -      -      -     1.00   vmovaps	32(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -      -     1.00   vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT:  -      -      -      -      -      -      -     1.00   vmovaps	48(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -      -     1.00   vmovaps	%xmm0, 48(%rdi)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -     1.00    -      -      -      -     0.99   0.01    -      -      -      -     vmovaps	(%rsi), %xmm0
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, (%rdi)
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -      -     vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, 16(%rdi)
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -      -     vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, 32(%rdi)
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -      -     vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, 48(%rdi)
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0
-# CHECK-NEXT: Index     0123456789          0123456789
+# CHECK-NEXT:                     0123456789
+# CHECK-NEXT: Index     0123456789          0123456
 
-# CHECK:      [0,0]     DeeeeeeER .    .    .    .    .   vmovaps	(%rsi), %xmm0
-# CHECK-NEXT: [0,1]     D======eER.    .    .    .    .   vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT: [0,2]     D=======eeeeeeER    .    .    .   vmovaps	16(%rsi), %xmm0
-# CHECK-NEXT: [0,3]     D=============eER   .    .    .   vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT: [0,4]     .D=============eeeeeeER  .    .   vmovaps	32(%rsi), %xmm0
-# CHECK-NEXT: [0,5]     .D===================eER .    .   vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT: [0,6]     .D====================eeeeeeER.   vmovaps	48(%rsi), %xmm0
-# CHECK-NEXT: [0,7]     .D==========================eER   vmovaps	%xmm0, 48(%rdi)
+# CHECK:      [0,0]     DeeeeeER  .    .    .    ..   vmovaps	(%rsi), %xmm0
+# CHECK-NEXT: [0,1]     D=====eER .    .    .    ..   vmovaps	%xmm0, (%rdi)
+# CHECK-NEXT: [0,2]     D======eeeeeER .    .    ..   vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT: [0,3]     D===========eER.    .    ..   vmovaps	%xmm0, 16(%rdi)
+# CHECK-NEXT: [0,4]     .D===========eeeeeER.    ..   vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT: [0,5]     .D================eER    ..   vmovaps	%xmm0, 32(%rdi)
+# CHECK-NEXT: [0,6]     .D=================eeeeeER.   vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT: [0,7]     .D======================eER   vmovaps	%xmm0, 48(%rdi)
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -84,10 +96,10 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       vmovaps	(%rsi), %xmm0
-# CHECK-NEXT: 1.     1     7.0    0.0    0.0       vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT: 2.     1     8.0    0.0    0.0       vmovaps	16(%rsi), %xmm0
-# CHECK-NEXT: 3.     1     14.0   0.0    0.0       vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT: 4.     1     14.0   0.0    0.0       vmovaps	32(%rsi), %xmm0
-# CHECK-NEXT: 5.     1     20.0   0.0    0.0       vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT: 6.     1     21.0   0.0    0.0       vmovaps	48(%rsi), %xmm0
-# CHECK-NEXT: 7.     1     27.0   0.0    0.0       vmovaps	%xmm0, 48(%rdi)
+# CHECK-NEXT: 1.     1     6.0    0.0    0.0       vmovaps	%xmm0, (%rdi)
+# CHECK-NEXT: 2.     1     7.0    0.0    0.0       vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT: 3.     1     12.0   0.0    0.0       vmovaps	%xmm0, 16(%rdi)
+# CHECK-NEXT: 4.     1     12.0   0.0    0.0       vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT: 5.     1     17.0   0.0    0.0       vmovaps	%xmm0, 32(%rdi)
+# CHECK-NEXT: 6.     1     18.0   0.0    0.0       vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT: 7.     1     23.0   0.0    0.0       vmovaps	%xmm0, 48(%rdi)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/memcpy-like-test.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/memcpy-like-test.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/memcpy-like-test.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=100 -timeline -timeline-max-iterations=1 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=100 -timeline -timeline-max-iterations=1 < %s | FileCheck %s
 
 vmovaps (%rsi), %xmm0
 vmovaps %xmm0, (%rdi)
@@ -12,7 +12,7 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      800
-# CHECK-NEXT: Total Cycles:      409
+# CHECK-NEXT: Total Cycles:      408
 # CHECK-NEXT: Total uOps:        800
 
 # CHECK:      Dispatch Width:    4
@@ -29,52 +29,64 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	16(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	32(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	48(%rsi), %xmm0
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, 48(%rdi)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -      -     4.00    -     3.94   4.06
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 4.05   3.95    -      -      -      -      -      -     3.95   0.05    -      -      -     4.00   3.95   4.05    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -     0.97   0.03   vmovaps	(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -      -     1.00   vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT:  -      -      -      -      -      -     0.03   0.97   vmovaps	16(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.97   0.03   vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -     vmovaps	32(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -      -     1.00   vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT:  -      -      -      -      -      -      -     1.00   vmovaps	48(%rsi), %xmm0
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.97   0.03   vmovaps	%xmm0, 48(%rdi)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -     0.97   0.03    -      -      -      -     0.97   0.03    -      -      -      -     vmovaps	(%rsi), %xmm0
+# CHECK-NEXT: 1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, (%rdi)
+# CHECK-NEXT: 1.00    -      -      -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -      -      -      -     vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT: 0.02   0.98    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, 16(%rdi)
+# CHECK-NEXT: 0.02   0.98    -      -      -      -      -      -     1.00    -      -      -      -      -     0.98   0.02    -      -      -      -     vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT: 1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, 32(%rdi)
+# CHECK-NEXT: 0.98   0.02    -      -      -      -      -      -     0.98   0.02    -      -      -      -     1.00    -      -      -      -      -     vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT: 0.03   0.97    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, 48(%rdi)
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     012
+# CHECK-NEXT:                     01
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeeeeER . .   vmovaps	(%rsi), %xmm0
-# CHECK-NEXT: [0,1]     D======eER. .   vmovaps	%xmm0, (%rdi)
-# CHECK-NEXT: [0,2]     DeeeeeeE-R. .   vmovaps	16(%rsi), %xmm0
-# CHECK-NEXT: [0,3]     D=======eER .   vmovaps	%xmm0, 16(%rdi)
-# CHECK-NEXT: [0,4]     .DeeeeeeE-R .   vmovaps	32(%rsi), %xmm0
-# CHECK-NEXT: [0,5]     .D=======eER.   vmovaps	%xmm0, 32(%rdi)
-# CHECK-NEXT: [0,6]     .DeeeeeeE--R.   vmovaps	48(%rsi), %xmm0
-# CHECK-NEXT: [0,7]     .D========eER   vmovaps	%xmm0, 48(%rdi)
+# CHECK:      [0,0]     DeeeeeER  ..   vmovaps	(%rsi), %xmm0
+# CHECK-NEXT: [0,1]     D=====eER ..   vmovaps	%xmm0, (%rdi)
+# CHECK-NEXT: [0,2]     DeeeeeE-R ..   vmovaps	16(%rsi), %xmm0
+# CHECK-NEXT: [0,3]     D======eER..   vmovaps	%xmm0, 16(%rdi)
+# CHECK-NEXT: [0,4]     .DeeeeeE-R..   vmovaps	32(%rsi), %xmm0
+# CHECK-NEXT: [0,5]     .D======eER.   vmovaps	%xmm0, 32(%rdi)
+# CHECK-NEXT: [0,6]     .DeeeeeE--R.   vmovaps	48(%rsi), %xmm0
+# CHECK-NEXT: [0,7]     .D=======eER   vmovaps	%xmm0, 48(%rdi)
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -84,10 +96,10 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       vmovaps	(%rsi), %xmm0
-# CHECK-NEXT: 1.     1     7.0    0.0    0.0       vmovaps	%xmm0, (%rdi)
+# CHECK-NEXT: 1.     1     6.0    0.0    0.0       vmovaps	%xmm0, (%rdi)
 # CHECK-NEXT: 2.     1     1.0    1.0    1.0       vmovaps	16(%rsi), %xmm0
-# CHECK-NEXT: 3.     1     8.0    0.0    0.0       vmovaps	%xmm0, 16(%rdi)
+# CHECK-NEXT: 3.     1     7.0    0.0    0.0       vmovaps	%xmm0, 16(%rdi)
 # CHECK-NEXT: 4.     1     1.0    1.0    1.0       vmovaps	32(%rsi), %xmm0
-# CHECK-NEXT: 5.     1     8.0    0.0    0.0       vmovaps	%xmm0, 32(%rdi)
+# CHECK-NEXT: 5.     1     7.0    0.0    0.0       vmovaps	%xmm0, 32(%rdi)
 # CHECK-NEXT: 6.     1     1.0    1.0    2.0       vmovaps	48(%rsi), %xmm0
-# CHECK-NEXT: 7.     1     9.0    0.0    0.0       vmovaps	%xmm0, 48(%rdi)
+# CHECK-NEXT: 7.     1     8.0    0.0    0.0       vmovaps	%xmm0, 48(%rdi)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/one-idioms.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/one-idioms.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/one-idioms.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=1 -register-file-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=1 -register-file-stats < %s | FileCheck %s
 
 # These are dependency-breaking one-idioms.
 # Much like zero-idioms, but they produce ones, and do consume resources.
@@ -29,13 +29,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      1500
-# CHECK-NEXT: Total Cycles:      903
+# CHECK-NEXT: Total Cycles:      754
 # CHECK-NEXT: Total uOps:        1500
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.66
-# CHECK-NEXT: IPC:               1.66
-# CHECK-NEXT: Block RThroughput: 6.0
+# CHECK-NEXT: uOps Per Cycle:    1.99
+# CHECK-NEXT: IPC:               1.99
+# CHECK-NEXT: Block RThroughput: 7.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -46,77 +46,99 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        pcmpeqb	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpeqd	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpeqw	%mm2, %mm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqb	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqd	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqq	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqw	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        pcmpeqb	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqd	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqw	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqb	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqd	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqq	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqw	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqw	%xmm3, %xmm3, %xmm5
 
 # CHECK:      Register File statistics:
 # CHECK-NEXT: Total number of mappings created:    1500
-# CHECK-NEXT: Max number of mappings used:         168
+# CHECK-NEXT: Max number of mappings used:         72
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 1500
+# CHECK-NEXT:    Max number of mappings used:      72
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     7.65    -     7.35    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     7.50   7.50    -      -     7.50   7.50    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqb	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqd	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqw	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     0.75    -     0.25    -      -     pcmpeqb	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -     0.49    -     0.51    -      -     pcmpeqd	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -     0.64    -     0.36    -      -     pcmpeqq	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -     0.21    -     0.79    -      -     pcmpeqw	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -     0.44    -     0.56    -      -     vpcmpeqb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -     0.26    -     0.74    -      -     vpcmpeqd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -     0.25    -     0.75    -      -     vpcmpeqq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpcmpeqw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -     0.25    -     0.75    -      -     vpcmpeqb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     0.55    -     0.45    -      -     vpcmpeqd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     0.44    -     0.56    -      -     vpcmpeqq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     0.37    -     0.63    -      -     vpcmpeqw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqb	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqd	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqw	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqb	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqd	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -      -      -      -     pcmpeqq	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqw	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     1.00    -      -      -      -      -     vpcmpeqd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqw	%xmm3, %xmm3, %xmm5
 
 # CHECK:      Timeline view:
 # CHECK-NEXT:                     01
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER    ..   pcmpeqb	%mm2, %mm2
-# CHECK-NEXT: [0,1]     D===eeeER ..   pcmpeqd	%mm2, %mm2
-# CHECK-NEXT: [0,2]     D======eeeER   pcmpeqw	%mm2, %mm2
-# CHECK-NEXT: [0,3]     DeE--------R   pcmpeqb	%xmm2, %xmm2
-# CHECK-NEXT: [0,4]     .DeE-------R   pcmpeqd	%xmm2, %xmm2
-# CHECK-NEXT: [0,5]     .D=eE------R   pcmpeqq	%xmm2, %xmm2
-# CHECK-NEXT: [0,6]     .D==eE-----R   pcmpeqw	%xmm2, %xmm2
-# CHECK-NEXT: [0,7]     .DeE-------R   vpcmpeqb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,8]     . DeE------R   vpcmpeqd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,9]     . D==eE----R   vpcmpeqq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,10]    . D===eE---R   vpcmpeqw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,11]    . D====eE--R   vpcmpeqb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,12]    .  D====eE-R   vpcmpeqd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,13]    .  D====eE-R   vpcmpeqq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,14]    .  D=====eER   vpcmpeqw	%xmm3, %xmm3, %xmm5
+# CHECK:      [0,0]     DeeER.    ..   pcmpeqb	%mm2, %mm2
+# CHECK-NEXT: [0,1]     DeeER.    ..   pcmpeqd	%mm2, %mm2
+# CHECK-NEXT: [0,2]     D=eeER    ..   pcmpeqw	%mm2, %mm2
+# CHECK-NEXT: [0,3]     D==eeER   ..   pcmpeqb	%xmm2, %xmm2
+# CHECK-NEXT: [0,4]     .DeeE-R   ..   pcmpeqd	%xmm2, %xmm2
+# CHECK-NEXT: [0,5]     .D==eeER  ..   pcmpeqq	%xmm2, %xmm2
+# CHECK-NEXT: [0,6]     .D=eeE-R  ..   pcmpeqw	%xmm2, %xmm2
+# CHECK-NEXT: [0,7]     .D===eeER ..   vpcmpeqb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,8]     . D=eeE-R ..   vpcmpeqd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,9]     . D===eeER..   vpcmpeqq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,10]    . D==eeE-R..   vpcmpeqw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,11]    . D===eeER..   vpcmpeqb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,12]    .  D===eeER.   vpcmpeqd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,13]    .  D===eeER.   vpcmpeqq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,14]    .  D====eeER   vpcmpeqw	%xmm3, %xmm3, %xmm5
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -126,17 +148,17 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       pcmpeqb	%mm2, %mm2
-# CHECK-NEXT: 1.     1     4.0    0.0    0.0       pcmpeqd	%mm2, %mm2
-# CHECK-NEXT: 2.     1     7.0    0.0    0.0       pcmpeqw	%mm2, %mm2
-# CHECK-NEXT: 3.     1     1.0    1.0    8.0       pcmpeqb	%xmm2, %xmm2
-# CHECK-NEXT: 4.     1     1.0    0.0    7.0       pcmpeqd	%xmm2, %xmm2
-# CHECK-NEXT: 5.     1     2.0    0.0    6.0       pcmpeqq	%xmm2, %xmm2
-# CHECK-NEXT: 6.     1     3.0    0.0    5.0       pcmpeqw	%xmm2, %xmm2
-# CHECK-NEXT: 7.     1     1.0    1.0    7.0       vpcmpeqb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 8.     1     1.0    0.0    6.0       vpcmpeqd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 9.     1     3.0    1.0    4.0       vpcmpeqq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 10.    1     4.0    0.0    3.0       vpcmpeqw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 11.    1     5.0    0.0    2.0       vpcmpeqb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 12.    1     5.0    1.0    1.0       vpcmpeqd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 13.    1     5.0    1.0    1.0       vpcmpeqq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 14.    1     6.0    2.0    0.0       vpcmpeqw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 1.     1     1.0    1.0    0.0       pcmpeqd	%mm2, %mm2
+# CHECK-NEXT: 2.     1     2.0    2.0    0.0       pcmpeqw	%mm2, %mm2
+# CHECK-NEXT: 3.     1     3.0    3.0    0.0       pcmpeqb	%xmm2, %xmm2
+# CHECK-NEXT: 4.     1     1.0    1.0    1.0       pcmpeqd	%xmm2, %xmm2
+# CHECK-NEXT: 5.     1     3.0    0.0    0.0       pcmpeqq	%xmm2, %xmm2
+# CHECK-NEXT: 6.     1     2.0    2.0    1.0       pcmpeqw	%xmm2, %xmm2
+# CHECK-NEXT: 7.     1     4.0    4.0    0.0       vpcmpeqb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 8.     1     2.0    2.0    1.0       vpcmpeqd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 9.     1     4.0    0.0    0.0       vpcmpeqq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 10.    1     3.0    3.0    1.0       vpcmpeqw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 11.    1     4.0    4.0    0.0       vpcmpeqb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 12.    1     4.0    4.0    0.0       vpcmpeqd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 13.    1     4.0    0.0    0.0       vpcmpeqq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 14.    1     5.0    5.0    0.0       vpcmpeqw	%xmm3, %xmm3, %xmm5
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=false -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=false -timeline < %s | FileCheck %s
 
 imul   %rax, %rbx
 lzcnt  %ax,  %bx
@@ -7,13 +7,13 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      3
-# CHECK-NEXT: Total Cycles:      8
-# CHECK-NEXT: Total uOps:        3
+# CHECK-NEXT: Total Cycles:      11
+# CHECK-NEXT: Total uOps:        4
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.38
-# CHECK-NEXT: IPC:               0.38
-# CHECK-NEXT: Block RThroughput: 2.0
+# CHECK-NEXT: uOps Per Cycle:    0.36
+# CHECK-NEXT: IPC:               0.27
+# CHECK-NEXT: Block RThroughput: 4.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -24,16 +24,17 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        imulq	%rax, %rbx
-# CHECK-NEXT:  1      3     1.00                        lzcntw	%ax, %bx
-# CHECK-NEXT:  1      1     0.33                        addl	%ecx, %ebx
+# CHECK-NEXT:  1      6     4.00                        imulq	%rax, %rbx
+# CHECK-NEXT:  2      2     0.50                        lzcntw	%ax, %bx
+# CHECK-NEXT:  1      1     0.50                        addl	%ecx, %ebx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     01234567
+# CHECK-NEXT:                     0
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER .   imulq	%rax, %rbx
-# CHECK-NEXT: [0,1]     D=eeeER.   lzcntw	%ax, %bx
-# CHECK-NEXT: [0,2]     D====eER   addl	%ecx, %ebx
+# CHECK:      [0,0]     DeeeeeeER .   imulq	%rax, %rbx
+# CHECK-NEXT: [0,1]     D=====eeER.   lzcntw	%ax, %bx
+# CHECK-NEXT: [0,2]     D=======eER   addl	%ecx, %ebx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -43,5 +44,5 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       imulq	%rax, %rbx
-# CHECK-NEXT: 1.     1     2.0    2.0    0.0       lzcntw	%ax, %bx
-# CHECK-NEXT: 2.     1     5.0    0.0    0.0       addl	%ecx, %ebx
+# CHECK-NEXT: 1.     1     6.0    0.0    0.0       lzcntw	%ax, %bx
+# CHECK-NEXT: 2.     1     8.0    0.0    0.0       addl	%ecx, %ebx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-3.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-3.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-3.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 # perf stat reports a throughput of 1.00 IPC for this code snippet.
 
@@ -12,13 +12,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      4500
-# CHECK-NEXT: Total Cycles:      1504
+# CHECK-NEXT: Total Cycles:      4503
 # CHECK-NEXT: Total uOps:        4500
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    2.99
-# CHECK-NEXT: IPC:               2.99
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.00
+# CHECK-NEXT: IPC:               1.00
+# CHECK-NEXT: Block RThroughput: 1.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -29,42 +29,55 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        addw	%cx, %dx
-# CHECK-NEXT:  1      1     0.33                        movw	%ax, %dx
-# CHECK-NEXT:  1      1     0.33                        xorw	%bx, %dx
+# CHECK-NEXT:  1      1     0.50                        addw	%cx, %dx
+# CHECK-NEXT:  1      1     0.50                        movw	%ax, %dx
+# CHECK-NEXT:  1      1     0.50                        xorw	%bx, %dx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     1.50   1.50    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.67    -      -     0.33    -      -     addw	%cx, %dx
-# CHECK-NEXT:  -      -      -     0.67    -     0.33    -      -     movw	%ax, %dx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorw	%bx, %dx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	%cx, %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movw	%ax, %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	%bx, %dx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     0123456
+# CHECK-NEXT:                     01
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER ..   addw	%cx, %dx
-# CHECK-NEXT: [0,1]     DeER ..   movw	%ax, %dx
-# CHECK-NEXT: [0,2]     D=eER..   xorw	%bx, %dx
-# CHECK-NEXT: [1,0]     D==eER.   addw	%cx, %dx
-# CHECK-NEXT: [1,1]     .DeE-R.   movw	%ax, %dx
-# CHECK-NEXT: [1,2]     .D=eER.   xorw	%bx, %dx
-# CHECK-NEXT: [2,0]     .D==eER   addw	%cx, %dx
-# CHECK-NEXT: [2,1]     .DeE--R   movw	%ax, %dx
-# CHECK-NEXT: [2,2]     . DeE-R   xorw	%bx, %dx
+# CHECK:      [0,0]     DeER .    ..   addw	%cx, %dx
+# CHECK-NEXT: [0,1]     D=eER.    ..   movw	%ax, %dx
+# CHECK-NEXT: [0,2]     D==eER    ..   xorw	%bx, %dx
+# CHECK-NEXT: [1,0]     D===eER   ..   addw	%cx, %dx
+# CHECK-NEXT: [1,1]     .D===eER  ..   movw	%ax, %dx
+# CHECK-NEXT: [1,2]     .D====eER ..   xorw	%bx, %dx
+# CHECK-NEXT: [2,0]     .D=====eER..   addw	%cx, %dx
+# CHECK-NEXT: [2,1]     .D======eER.   movw	%ax, %dx
+# CHECK-NEXT: [2,2]     . D======eER   xorw	%bx, %dx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -73,6 +86,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     2.3    0.3    0.0       addw	%cx, %dx
-# CHECK-NEXT: 1.     3     1.0    1.0    1.0       movw	%ax, %dx
-# CHECK-NEXT: 2.     3     1.7    0.0    0.3       xorw	%bx, %dx
+# CHECK-NEXT: 0.     3     3.7    0.3    0.0       addw	%cx, %dx
+# CHECK-NEXT: 1.     3     4.3    0.0    0.0       movw	%ax, %dx
+# CHECK-NEXT: 2.     3     5.0    0.0    0.0       xorw	%bx, %dx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-4.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-4.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-4.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 # perf stat reports a throughput of 0.60 IPC for this code snippet.
 
@@ -12,13 +12,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      4500
-# CHECK-NEXT: Total Cycles:      3005
-# CHECK-NEXT: Total uOps:        4500
+# CHECK-NEXT: Total Cycles:      9003
+# CHECK-NEXT: Total uOps:        6000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.50
-# CHECK-NEXT: IPC:               1.50
-# CHECK-NEXT: Block RThroughput: 2.0
+# CHECK-NEXT: uOps Per Cycle:    0.67
+# CHECK-NEXT: IPC:               0.50
+# CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -29,43 +29,55 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        imulw	%ax, %bx
-# CHECK-NEXT:  1      3     1.00                        lzcntw	%ax, %bx
-# CHECK-NEXT:  1      1     0.33                        addw	%cx, %bx
+# CHECK-NEXT:  1      4     1.00                        imulw	%ax, %bx
+# CHECK-NEXT:  2      2     0.50                        lzcntw	%ax, %bx
+# CHECK-NEXT:  1      1     0.50                        addw	%cx, %bx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     0.50   2.00    -     0.50    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     1.50   1.50    -      -      -      -      -      -      -      -      -      -      -      -     1.00
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imulw	%ax, %bx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     lzcntw	%ax, %bx
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     addw	%cx, %bx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	%ax, %bx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntw	%ax, %bx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	%cx, %bx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01
-# CHECK-NEXT: Index     0123456789
+# CHECK-NEXT:                     0123456789
+# CHECK-NEXT: Index     0123456789          0
 
-# CHECK:      [0,0]     DeeeER    ..   imulw	%ax, %bx
-# CHECK-NEXT: [0,1]     D=eeeER   ..   lzcntw	%ax, %bx
-# CHECK-NEXT: [0,2]     D====eER  ..   addw	%cx, %bx
-# CHECK-NEXT: [1,0]     D=====eeeER.   imulw	%ax, %bx
-# CHECK-NEXT: [1,1]     .D=eeeE---R.   lzcntw	%ax, %bx
-# CHECK-NEXT: [1,2]     .D====eE--R.   addw	%cx, %bx
-# CHECK-NEXT: [2,0]     .D=====eeeER   imulw	%ax, %bx
-# CHECK-NEXT: [2,1]     .D==eeeE---R   lzcntw	%ax, %bx
-# CHECK-NEXT: [2,2]     . D====eE--R   addw	%cx, %bx
+# CHECK:      [0,0]     DeeeeER   .    .    .   imulw	%ax, %bx
+# CHECK-NEXT: [0,1]     D===eeER  .    .    .   lzcntw	%ax, %bx
+# CHECK-NEXT: [0,2]     D=====eER .    .    .   addw	%cx, %bx
+# CHECK-NEXT: [1,0]     .D=====eeeeER  .    .   imulw	%ax, %bx
+# CHECK-NEXT: [1,1]     .D========eeER .    .   lzcntw	%ax, %bx
+# CHECK-NEXT: [1,2]     .D==========eER.    .   addw	%cx, %bx
+# CHECK-NEXT: [2,0]     . D==========eeeeER .   imulw	%ax, %bx
+# CHECK-NEXT: [2,1]     . D=============eeER.   lzcntw	%ax, %bx
+# CHECK-NEXT: [2,2]     . D===============eER   addw	%cx, %bx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -74,6 +86,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     4.3    0.3    0.0       imulw	%ax, %bx
-# CHECK-NEXT: 1.     3     2.3    2.3    2.0       lzcntw	%ax, %bx
-# CHECK-NEXT: 2.     3     5.0    0.0    1.3       addw	%cx, %bx
+# CHECK-NEXT: 0.     3     6.0    0.3    0.0       imulw	%ax, %bx
+# CHECK-NEXT: 1.     3     9.0    0.0    0.0       lzcntw	%ax, %bx
+# CHECK-NEXT: 2.     3     11.0   0.0    0.0       addw	%cx, %bx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-5.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-5.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-5.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 # perf stat reports a throughput of 1.00 IPC for this code snippet.
 
@@ -7,13 +7,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      1500
-# CHECK-NEXT: Total Cycles:      1505
-# CHECK-NEXT: Total uOps:        1500
+# CHECK-NEXT: Total Cycles:      1504
+# CHECK-NEXT: Total uOps:        3000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
+# CHECK-NEXT: uOps Per Cycle:    1.99
 # CHECK-NEXT: IPC:               1.00
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: Block RThroughput: 0.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -24,32 +24,44 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        lzcntw	%ax, %bx
+# CHECK-NEXT:  2      2     0.50                        lzcntw	%ax, %bx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     lzcntw	%ax, %bx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntw	%ax, %bx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     01234567
+# CHECK-NEXT: Index     0123456
 
-# CHECK:      [0,0]     DeeeER .   lzcntw	%ax, %bx
-# CHECK-NEXT: [1,0]     D=eeeER.   lzcntw	%ax, %bx
-# CHECK-NEXT: [2,0]     D==eeeER   lzcntw	%ax, %bx
+# CHECK:      [0,0]     DeeER..   lzcntw	%ax, %bx
+# CHECK-NEXT: [1,0]     D=eeER.   lzcntw	%ax, %bx
+# CHECK-NEXT: [2,0]     .D=eeER   lzcntw	%ax, %bx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -58,4 +70,4 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     2.0    2.0    0.0       lzcntw	%ax, %bx
+# CHECK-NEXT: 0.     3     1.7    0.3    0.0       lzcntw	%ax, %bx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-6.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-6.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update-6.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1500 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 # perf stat reports a throughput of 0.60 IPC for this code snippet.
 # Each lzcnt has a false dependency on %ecx; the first lzcnt has to wait on the
@@ -13,13 +13,13 @@
 
 # CHECK:      Iterations:        1500
 # CHECK-NEXT: Instructions:      4500
-# CHECK-NEXT: Total Cycles:      4510
+# CHECK-NEXT: Total Cycles:      10503
 # CHECK-NEXT: Total uOps:        7500
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.66
-# CHECK-NEXT: IPC:               1.00
-# CHECK-NEXT: Block RThroughput: 3.0
+# CHECK-NEXT: uOps Per Cycle:    0.71
+# CHECK-NEXT: IPC:               0.43
+# CHECK-NEXT: Block RThroughput: 1.3
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -30,43 +30,55 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        imull	%edx, %ecx
-# CHECK-NEXT:  2      8     1.00    *                   lzcntw	(%rsp), %cx
-# CHECK-NEXT:  2      8     1.00    *                   lzcntw	2(%rsp), %cx
+# CHECK-NEXT:  1      4     1.00                        imull	%edx, %ecx
+# CHECK-NEXT:  2      6     0.50    *                   lzcntw	(%rsp), %cx
+# CHECK-NEXT:  2      6     0.50    *                   lzcntw	2(%rsp), %cx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     3.00    -      -      -     2.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -     2.00    -      -      -     1.50   1.50    -      -      -      -      -      -      -      -      -      -      -      -     1.00
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imull	%edx, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -     1.00   lzcntw	(%rsp), %cx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -     1.00   lzcntw	2(%rsp), %cx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	%edx, %ecx
+# CHECK-NEXT:  -     1.00    -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntw	(%rsp), %cx
+# CHECK-NEXT:  -     1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntw	2(%rsp), %cx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     012345678
-# CHECK-NEXT: Index     0123456789
+# CHECK-NEXT:                     0123456789
+# CHECK-NEXT: Index     0123456789          0123
 
-# CHECK:      [0,0]     DeeeER    .    .  .   imull	%edx, %ecx
-# CHECK-NEXT: [0,1]     D=eeeeeeeeER   .  .   lzcntw	(%rsp), %cx
-# CHECK-NEXT: [0,2]     .D=eeeeeeeeER  .  .   lzcntw	2(%rsp), %cx
-# CHECK-NEXT: [1,0]     .D=========eeeER  .   imull	%edx, %ecx
-# CHECK-NEXT: [1,1]     . D=eeeeeeeeE--R  .   lzcntw	(%rsp), %cx
-# CHECK-NEXT: [1,2]     . D==eeeeeeeeE-R  .   lzcntw	2(%rsp), %cx
-# CHECK-NEXT: [2,0]     .  D==========eeeER   imull	%edx, %ecx
-# CHECK-NEXT: [2,1]     .  D==eeeeeeeeE---R   lzcntw	(%rsp), %cx
-# CHECK-NEXT: [2,2]     .   D==eeeeeeeeE--R   lzcntw	2(%rsp), %cx
+# CHECK:      [0,0]     DeeeeER   .    .    .  .   imull	%edx, %ecx
+# CHECK-NEXT: [0,1]     DeeeeeeER .    .    .  .   lzcntw	(%rsp), %cx
+# CHECK-NEXT: [0,2]     .DeeeeeeER.    .    .  .   lzcntw	2(%rsp), %cx
+# CHECK-NEXT: [1,0]     .D======eeeeER .    .  .   imull	%edx, %ecx
+# CHECK-NEXT: [1,1]     . D=====eeeeeeER    .  .   lzcntw	(%rsp), %cx
+# CHECK-NEXT: [1,2]     . D======eeeeeeER   .  .   lzcntw	2(%rsp), %cx
+# CHECK-NEXT: [2,0]     .  D===========eeeeER  .   imull	%edx, %ecx
+# CHECK-NEXT: [2,1]     .  D===========eeeeeeER.   lzcntw	(%rsp), %cx
+# CHECK-NEXT: [2,2]     .   D===========eeeeeeER   lzcntw	2(%rsp), %cx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -75,6 +87,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     7.3    0.3    0.0       imull	%edx, %ecx
-# CHECK-NEXT: 1.     3     2.3    2.3    1.7       lzcntw	(%rsp), %cx
-# CHECK-NEXT: 2.     3     2.7    2.7    1.0       lzcntw	2(%rsp), %cx
+# CHECK-NEXT: 0.     3     6.7    0.3    0.0       imull	%edx, %ecx
+# CHECK-NEXT: 1.     3     6.3    0.0    0.0       lzcntw	(%rsp), %cx
+# CHECK-NEXT: 2.     3     6.7    0.0    0.0       lzcntw	2(%rsp), %cx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/partial-reg-update.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=false -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=false -timeline < %s | FileCheck %s
 
 imul %ax, %cx
 add  %al, %cl
@@ -7,12 +7,12 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      3
-# CHECK-NEXT: Total Cycles:      8
+# CHECK-NEXT: Total Cycles:      9
 # CHECK-NEXT: Total uOps:        3
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.38
-# CHECK-NEXT: IPC:               0.38
+# CHECK-NEXT: uOps Per Cycle:    0.33
+# CHECK-NEXT: IPC:               0.33
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
@@ -24,16 +24,16 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        imulw	%ax, %cx
-# CHECK-NEXT:  1      1     0.33                        addb	%al, %cl
-# CHECK-NEXT:  1      1     0.33                        addl	%ecx, %ebx
+# CHECK-NEXT:  1      4     1.00                        imulw	%ax, %cx
+# CHECK-NEXT:  1      1     0.50                        addb	%al, %cl
+# CHECK-NEXT:  1      1     0.50                        addl	%ecx, %ebx
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     01234567
+# CHECK-NEXT: Index     012345678
 
-# CHECK:      [0,0]     DeeeER .   imulw	%ax, %cx
-# CHECK-NEXT: [0,1]     D===eER.   addb	%al, %cl
-# CHECK-NEXT: [0,2]     D====eER   addl	%ecx, %ebx
+# CHECK:      [0,0]     DeeeeER .   imulw	%ax, %cx
+# CHECK-NEXT: [0,1]     D====eER.   addb	%al, %cl
+# CHECK-NEXT: [0,2]     D=====eER   addl	%ecx, %ebx
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -43,5 +43,5 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       imulw	%ax, %cx
-# CHECK-NEXT: 1.     1     4.0    0.0    0.0       addb	%al, %cl
-# CHECK-NEXT: 2.     1     5.0    0.0    0.0       addl	%ecx, %ebx
+# CHECK-NEXT: 1.     1     5.0    0.0    0.0       addb	%al, %cl
+# CHECK-NEXT: 2.     1     6.0    0.0    0.0       addl	%ecx, %ebx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/pipes-fpu.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/pipes-fpu.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/pipes-fpu.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
 
 # VALU0/VALU1
 vpmulld     %xmm0, %xmm1, %xmm2
@@ -19,13 +19,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      800
-# CHECK-NEXT: Total Cycles:      4256
-# CHECK-NEXT: Total uOps:        1000
+# CHECK-NEXT: Total Cycles:      3244
+# CHECK-NEXT: Total uOps:        1500
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.23
-# CHECK-NEXT: IPC:               0.19
-# CHECK-NEXT: Block RThroughput: 42.0
+# CHECK-NEXT: uOps Per Cycle:    0.46
+# CHECK-NEXT: IPC:               0.25
+# CHECK-NEXT: Block RThroughput: 32.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -36,51 +36,72 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      5     1.00                        vpmulld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.33                        vpand	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  1      14    6.00                        vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      14    14.00                       vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  3      29    28.00                       vsqrtps	%ymm0, %ymm2
+# CHECK-NEXT:  1      5     2.00                        vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT:  6      13    1.00                        vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     10.50                       vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      9     21.00                       vsqrtps	%ymm0, %ymm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -     42.00  6.03   3.96    -     17.01   -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     32.71  32.29   -     2.00   3.00   1.00   6.00   6.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmulld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.05   0.06    -     0.89    -      -     vpand	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.98   0.90    -     15.12   -      -     vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -     28.00  2.00    -      -     1.00    -      -     vsqrtps	%ymm0, %ymm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00   2.00    -     2.00   1.00    -      -      -      -     vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     10.71  10.29   -      -      -      -      -     1.00    -      -      -      -     vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     21.00  21.00   -      -      -      -      -     2.00    -      -      -      -     vsqrtps	%ymm0, %ymm2
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456
-# CHECK-NEXT: Index     0123456789          0123456789
+# CHECK-NEXT:                     0123456789          0123456789          012345678
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeeeER  .    .    .    .    .    ..   vpmulld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [0,1]     DeE----R  .    .    .    .    .    ..   vpand	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [0,2]     DeeeE--R  .    .    .    .    .    ..   vcvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT: [0,3]     D=eeeeeeeeeeeeeeER  .    .    .    ..   vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [0,4]     .DeeeE-----------R  .    .    .    ..   vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: [0,5]     .DeeeeeeeeeeeeeeER  .    .    .    ..   vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT: [0,6]     .D=eeeE----------R  .    .    .    ..   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK:      [0,0]     DeeeeeER  .    .    .    .    .    .    .    .    .    .  .   vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [0,1]     D=eeE--R  .    .    .    .    .    .    .    .    .    .  .   vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [0,2]     D==eeeeER .    .    .    .    .    .    .    .    .    .  .   vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT: [0,3]     .D=eeeeeeeeeeeeeER  .    .    .    .    .    .    .    .  .   vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [0,4]     . D=eeeeeE-------R  .    .    .    .    .    .    .    .  .   vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [0,5]     . D=eeeeeeeeeE---R  .    .    .    .    .    .    .    .  .   vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT: [0,6]     .  D=eeeeeE------R  .    .    .    .    .    .    .    .  .   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [0,7]     .  D==eeeeeeeeeE--R .    .    .    .    .    .    .    .  .   vsqrtps	%ymm0, %ymm2
+# CHECK-NEXT: [1,0]     .   D===eeeeeE----R .    .    .    .    .    .    .    .  .   vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [1,1]     .   DeeE----------R .    .    .    .    .    .    .    .  .   vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [1,2]     .   D====eeeeE----R .    .    .    .    .    .    .    .  .   vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT: [1,3]     .    D=eeeeeeeeeeeeeER   .    .    .    .    .    .    .  .   vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [1,4]     .    .D==================eeeeeER   .    .    .    .    .  .   vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: [1,5]     .    .D===================eeeeeeeeeER   .    .    .    .  .   vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT: [1,6]     .    . D=======================================eeeeeER .  .   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [1,7]     .    . D========================================eeeeeeeeeER   vsqrtps	%ymm0, %ymm2
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -89,11 +110,11 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     1.0    1.0    79.0      vpmulld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: 1.     2     1.0    1.0    82.5      vpand	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: 2.     2     1.5    1.5    80.0      vcvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT: 3.     2     1.5    1.5    74.0      vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT: 4.     2     2.0    2.0    84.0      vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT: 5.     2     9.5    9.5    65.0      vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT: 6.     2     2.5    2.5    83.0      vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: 7.     2     147.5  147.5  0.0       vsqrtps	%ymm0, %ymm2
+# CHECK-NEXT: 0.     2     2.5    2.5    2.0       vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 1.     2     1.5    1.5    6.0       vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 2.     2     4.0    4.0    2.0       vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT: 3.     2     2.0    2.0    0.0       vpclmulqdq	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 4.     2     10.5   10.5   3.5       vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 5.     2     11.0   11.0   1.5       vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT: 6.     2     21.0   21.0   3.0       vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 7.     2     22.0   22.0   1.0       vsqrtps	%ymm0, %ymm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/pr37790.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/pr37790.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/pr37790.s
@@ -1,18 +1,18 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -lqueue=2 -iterations=2 -resource-pressure=false -timeline -timeline-max-cycles=104 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -lqueue=2 -iterations=2 -resource-pressure=false -timeline -timeline-max-cycles=104 < %s | FileCheck %s
 
 int3
 stmxcsr (%rsp)
 
 # CHECK:      Iterations:        2
 # CHECK-NEXT: Instructions:      4
-# CHECK-NEXT: Total Cycles:      213
-# CHECK-NEXT: Total uOps:        10
+# CHECK-NEXT: Total Cycles:      205
+# CHECK-NEXT: Total uOps:        6
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.05
+# CHECK-NEXT: uOps Per Cycle:    0.03
 # CHECK-NEXT: IPC:               0.02
-# CHECK-NEXT: Block RThroughput: 1.3
+# CHECK-NEXT: Block RThroughput: 0.8
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -23,14 +23,15 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      100   0.33    *      *      U     int3
-# CHECK-NEXT:  4      5     1.00    *      *      U     stmxcsr	(%rsp)
+# CHECK-NEXT:  1      100   0.50    *      *      U     int3
+# CHECK-NEXT:  2      1     0.50    *      *      U     stmxcsr	(%rsp)
 
 # CHECK:      Timeline view:
 # CHECK-NEXT:                     0123456789          0123456789          0123456789          0123456789          0123456789
-# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456789          0123456789          012
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456789          0123456789          0123
 
-# CHECK:      [0,0]     DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER   int3
+# CHECK:      [0,0]     DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER.   int3
+# CHECK-NEXT: [0,1]     D====================================================================================================eER   stmxcsr	(%rsp)
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -39,5 +40,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     3.0    0.5    0.0       int3
-# CHECK-NEXT: 1.     2     100.0  0.0    0.0       stmxcsr	(%rsp)
+# CHECK-NEXT: 0.     2     1.0    0.5    0.0       int3
+# CHECK-NEXT: 1.     2     100.5  0.0    0.0       stmxcsr	(%rsp)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/rank.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/rank.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/rank.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 add %eax, %ecx
 add %eax, %edx
@@ -12,13 +12,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      800
-# CHECK-NEXT: Total Cycles:      403
+# CHECK-NEXT: Total Cycles:      503
 # CHECK-NEXT: Total uOps:        800
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.99
-# CHECK-NEXT: IPC:               1.99
-# CHECK-NEXT: Block RThroughput: 2.7
+# CHECK-NEXT: uOps Per Cycle:    1.59
+# CHECK-NEXT: IPC:               1.59
+# CHECK-NEXT: Block RThroughput: 4.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -29,68 +29,80 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        addl	%eax, %ecx
-# CHECK-NEXT:  1      1     0.33                        addl	%eax, %edx
-# CHECK-NEXT:  1      1     0.33                        addl	%eax, %ebx
-# CHECK-NEXT:  1      1     0.33                        addl	%edx, %esi
-# CHECK-NEXT:  1      1     0.33                        addl	%ebx, %eax
-# CHECK-NEXT:  1      1     0.33                        addl	%edx, %esi
-# CHECK-NEXT:  1      1     0.33                        addl	%ebx, %eax
-# CHECK-NEXT:  1      1     0.33                        addl	%ebx, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	%eax, %ecx
+# CHECK-NEXT:  1      1     0.50                        addl	%eax, %edx
+# CHECK-NEXT:  1      1     0.50                        addl	%eax, %ebx
+# CHECK-NEXT:  1      1     0.50                        addl	%edx, %esi
+# CHECK-NEXT:  1      1     0.50                        addl	%ebx, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	%edx, %esi
+# CHECK-NEXT:  1      1     0.50                        addl	%ebx, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	%ebx, %eax
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     2.66   2.67    -     2.67    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.34    -      -     addl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.34    -     0.33    -      -     addl	%eax, %edx
-# CHECK-NEXT:  -      -     0.34   0.33    -     0.33    -      -     addl	%eax, %ebx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.34    -      -     addl	%edx, %esi
-# CHECK-NEXT:  -      -     0.33   0.34    -     0.33    -      -     addl	%ebx, %eax
-# CHECK-NEXT:  -      -     0.34   0.33    -     0.33    -      -     addl	%edx, %esi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.34    -      -     addl	%ebx, %eax
-# CHECK-NEXT:  -      -     0.33   0.34    -     0.33    -      -     addl	%ebx, %eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.01   0.99    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%eax, %edx
+# CHECK-NEXT:  -      -      -      -      -     0.99   0.01    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%eax, %ebx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%edx, %esi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%ebx, %eax
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%edx, %esi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%ebx, %eax
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%ebx, %eax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
+# CHECK-NEXT:                     01234567
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    .   .   addl	%eax, %ecx
-# CHECK-NEXT: [0,1]     DeER .    .   .   addl	%eax, %edx
-# CHECK-NEXT: [0,2]     DeER .    .   .   addl	%eax, %ebx
-# CHECK-NEXT: [0,3]     D=eER.    .   .   addl	%edx, %esi
-# CHECK-NEXT: [0,4]     .DeER.    .   .   addl	%ebx, %eax
-# CHECK-NEXT: [0,5]     .D=eER    .   .   addl	%edx, %esi
-# CHECK-NEXT: [0,6]     .D=eER    .   .   addl	%ebx, %eax
-# CHECK-NEXT: [0,7]     .D==eER   .   .   addl	%ebx, %eax
-# CHECK-NEXT: [1,0]     . D==eER  .   .   addl	%eax, %ecx
-# CHECK-NEXT: [1,1]     . D==eER  .   .   addl	%eax, %edx
-# CHECK-NEXT: [1,2]     . D==eER  .   .   addl	%eax, %ebx
-# CHECK-NEXT: [1,3]     . D===eER .   .   addl	%edx, %esi
-# CHECK-NEXT: [1,4]     .  D==eER .   .   addl	%ebx, %eax
-# CHECK-NEXT: [1,5]     .  D===eER.   .   addl	%edx, %esi
-# CHECK-NEXT: [1,6]     .  D===eER.   .   addl	%ebx, %eax
-# CHECK-NEXT: [1,7]     .  D====eER   .   addl	%ebx, %eax
-# CHECK-NEXT: [2,0]     .   D====eER  .   addl	%eax, %ecx
-# CHECK-NEXT: [2,1]     .   D====eER  .   addl	%eax, %edx
-# CHECK-NEXT: [2,2]     .   D====eER  .   addl	%eax, %ebx
-# CHECK-NEXT: [2,3]     .   D=====eER .   addl	%edx, %esi
-# CHECK-NEXT: [2,4]     .    D====eER .   addl	%ebx, %eax
-# CHECK-NEXT: [2,5]     .    D=====eER.   addl	%edx, %esi
-# CHECK-NEXT: [2,6]     .    D=====eER.   addl	%ebx, %eax
-# CHECK-NEXT: [2,7]     .    D======eER   addl	%ebx, %eax
+# CHECK:      [0,0]     DeER .    .    . .   addl	%eax, %ecx
+# CHECK-NEXT: [0,1]     DeER .    .    . .   addl	%eax, %edx
+# CHECK-NEXT: [0,2]     D=eER.    .    . .   addl	%eax, %ebx
+# CHECK-NEXT: [0,3]     D=eER.    .    . .   addl	%edx, %esi
+# CHECK-NEXT: [0,4]     .D=eER    .    . .   addl	%ebx, %eax
+# CHECK-NEXT: [0,5]     .D=eER    .    . .   addl	%edx, %esi
+# CHECK-NEXT: [0,6]     .D==eER   .    . .   addl	%ebx, %eax
+# CHECK-NEXT: [0,7]     .D===eER  .    . .   addl	%ebx, %eax
+# CHECK-NEXT: [1,0]     . D====eER.    . .   addl	%eax, %ecx
+# CHECK-NEXT: [1,1]     . D===eE-R.    . .   addl	%eax, %edx
+# CHECK-NEXT: [1,2]     . D===eE-R.    . .   addl	%eax, %ebx
+# CHECK-NEXT: [1,3]     . D====eER.    . .   addl	%edx, %esi
+# CHECK-NEXT: [1,4]     .  D====eER    . .   addl	%ebx, %eax
+# CHECK-NEXT: [1,5]     .  D====eER    . .   addl	%edx, %esi
+# CHECK-NEXT: [1,6]     .  D=====eER   . .   addl	%ebx, %eax
+# CHECK-NEXT: [1,7]     .  D======eER  . .   addl	%ebx, %eax
+# CHECK-NEXT: [2,0]     .   D=======eER. .   addl	%eax, %ecx
+# CHECK-NEXT: [2,1]     .   D======eE-R. .   addl	%eax, %edx
+# CHECK-NEXT: [2,2]     .   D======eE-R. .   addl	%eax, %ebx
+# CHECK-NEXT: [2,3]     .   D=======eER. .   addl	%edx, %esi
+# CHECK-NEXT: [2,4]     .    D=======eER .   addl	%ebx, %eax
+# CHECK-NEXT: [2,5]     .    D=======eER .   addl	%edx, %esi
+# CHECK-NEXT: [2,6]     .    D========eER.   addl	%ebx, %eax
+# CHECK-NEXT: [2,7]     .    D=========eER   addl	%ebx, %eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -99,11 +111,11 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     3.0    0.3    0.0       addl	%eax, %ecx
-# CHECK-NEXT: 1.     3     3.0    0.3    0.0       addl	%eax, %edx
-# CHECK-NEXT: 2.     3     3.0    0.3    0.0       addl	%eax, %ebx
-# CHECK-NEXT: 3.     3     4.0    0.0    0.0       addl	%edx, %esi
-# CHECK-NEXT: 4.     3     3.0    0.0    0.0       addl	%ebx, %eax
-# CHECK-NEXT: 5.     3     4.0    0.0    0.0       addl	%edx, %esi
-# CHECK-NEXT: 6.     3     4.0    0.0    0.0       addl	%ebx, %eax
-# CHECK-NEXT: 7.     3     5.0    0.0    0.0       addl	%ebx, %eax
+# CHECK-NEXT: 0.     3     4.7    1.0    0.0       addl	%eax, %ecx
+# CHECK-NEXT: 1.     3     4.0    0.3    0.7       addl	%eax, %edx
+# CHECK-NEXT: 2.     3     4.3    0.7    0.7       addl	%eax, %ebx
+# CHECK-NEXT: 3.     3     5.0    0.0    0.0       addl	%edx, %esi
+# CHECK-NEXT: 4.     3     5.0    0.7    0.0       addl	%ebx, %eax
+# CHECK-NEXT: 5.     3     5.0    0.0    0.0       addl	%edx, %esi
+# CHECK-NEXT: 6.     3     6.0    0.0    0.0       addl	%ebx, %eax
+# CHECK-NEXT: 7.     3     7.0    0.0    0.0       addl	%ebx, %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/rcu-statistics.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/rcu-statistics.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/rcu-statistics.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -resource-pressure=false -retire-stats -iterations=1 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -resource-pressure=false -retire-stats -iterations=1 < %s | FileCheck %s
 
   vsqrtps %xmm0, %xmm2
   vaddps  %xmm0, %xmm1, %xmm2
@@ -20,13 +20,13 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      16
-# CHECK-NEXT: Total Cycles:      20
+# CHECK-NEXT: Total Cycles:      22
 # CHECK-NEXT: Total uOps:        16
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.80
-# CHECK-NEXT: IPC:               0.80
-# CHECK-NEXT: Block RThroughput: 15.0
+# CHECK-NEXT: uOps Per Cycle:    0.73
+# CHECK-NEXT: IPC:               0.73
+# CHECK-NEXT: Block RThroughput: 18.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -37,25 +37,26 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      14    14.00                       vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     10.50                       vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
 
 # CHECK:      Retire Control Unit - number of cycles where we saw N instructions retired:
 # CHECK-NEXT: [# retired], [# cycles]
-# CHECK-NEXT:  0,           16  (80.0%)
-# CHECK-NEXT:  1,           3  (15.0%)
-# CHECK-NEXT:  13,          1  (5.0%)
+# CHECK-NEXT:  0,           11  (50.0%)
+# CHECK-NEXT:  1,           9  (40.9%)
+# CHECK-NEXT:  3,           1  (4.5%)
+# CHECK-NEXT:  4,           1  (4.5%)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
 
 # The vmul can start executing 3cy in advance. That is beause the first use
 # operand (i.e. %xmm1) is a ReadAfterLd. That means, the memory operand is
@@ -10,12 +10,12 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      14
-# CHECK-NEXT: Total uOps:        3
+# CHECK-NEXT: Total Cycles:      13
+# CHECK-NEXT: Total uOps:        2
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.21
-# CHECK-NEXT: IPC:               0.14
+# CHECK-NEXT: uOps Per Cycle:    0.15
+# CHECK-NEXT: IPC:               0.15
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
@@ -27,15 +27,15 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  2      11    1.00    *                   vmulps	(%rdi), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  1      10    1.00    *                   vmulps	(%rdi), %xmm1, %xmm2
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123
+# CHECK-NEXT:                     012
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER    .  .   vaddps	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT: [0,1]     DeeeeeeeeeeeER   vmulps	(%rdi), %xmm1, %xmm2
+# CHECK:      [0,0]     DeeeeeER  . .   vaddps	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT: [0,1]     DeeeeeeeeeeER   vmulps	(%rdi), %xmm1, %xmm2
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=0 -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=0 -timeline < %s | FileCheck %s
 
   imull  %esi
   imull  (%rdi)
@@ -9,12 +9,12 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      13
-# CHECK-NEXT: Total uOps:        7
+# CHECK-NEXT: Total Cycles:      12
+# CHECK-NEXT: Total uOps:        2
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.54
-# CHECK-NEXT: IPC:               0.15
+# CHECK-NEXT: uOps Per Cycle:    0.17
+# CHECK-NEXT: IPC:               0.17
 # CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
@@ -26,15 +26,15 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  3      4     1.00                        imull	%esi
-# CHECK-NEXT:  4      9     1.00    *                   imull	(%rdi)
+# CHECK-NEXT:  1      4     1.00                        imull	%esi
+# CHECK-NEXT:  1      8     1.00    *                   imull	(%rdi)
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     012
+# CHECK-NEXT:                     01
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeeER   . .   imull	%esi
-# CHECK-NEXT: [0,1]     .DeeeeeeeeeER   imull	(%rdi)
+# CHECK:      [0,0]     DeeeeER   ..   imull	%esi
+# CHECK-NEXT: [0,1]     D=eeeeeeeeER   imull	(%rdi)
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -44,4 +44,4 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       imull	%esi
-# CHECK-NEXT: 1.     1     1.0    1.0    0.0       imull	(%rdi)
+# CHECK-NEXT: 1.     1     2.0    1.0    0.0       imull	(%rdi)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-3.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-3.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/read-advance-3.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=0 -timeline -dispatch=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=0 -timeline -dispatch=3 < %s | FileCheck %s
 
   add %rdi, %rsi
   add (%rsp), %rsi
@@ -7,13 +7,13 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      3
-# CHECK-NEXT: Total Cycles:      9
-# CHECK-NEXT: Total uOps:        4
+# CHECK-NEXT: Total Cycles:      8
+# CHECK-NEXT: Total uOps:        3
 
 # CHECK:      Dispatch Width:    3
-# CHECK-NEXT: uOps Per Cycle:    0.44
-# CHECK-NEXT: IPC:               0.33
-# CHECK-NEXT: Block RThroughput: 1.3
+# CHECK-NEXT: uOps Per Cycle:    0.38
+# CHECK-NEXT: IPC:               0.38
+# CHECK-NEXT: Block RThroughput: 1.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -24,16 +24,16 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        addq	%rdi, %rsi
-# CHECK-NEXT:  2      6     0.50    *                   addq	(%rsp), %rsi
-# CHECK-NEXT:  1      1     0.33                        addq	%rdx, %r8
+# CHECK-NEXT:  1      1     0.50                        addq	%rdi, %rsi
+# CHECK-NEXT:  1      5     0.50    *                   addq	(%rsp), %rsi
+# CHECK-NEXT:  1      1     0.50                        addq	%rdx, %r8
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     012345678
+# CHECK-NEXT: Index     01234567
 
-# CHECK:      [0,0]     DeER .  .   addq	%rdi, %rsi
-# CHECK-NEXT: [0,1]     DeeeeeeER   addq	(%rsp), %rsi
-# CHECK-NEXT: [0,2]     .DeE----R   addq	%rdx, %r8
+# CHECK:      [0,0]     DeER . .   addq	%rdi, %rsi
+# CHECK-NEXT: [0,1]     DeeeeeER   addq	(%rsp), %rsi
+# CHECK-NEXT: [0,2]     D=eE---R   addq	%rdx, %r8
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -44,4 +44,4 @@
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       addq	%rdi, %rsi
 # CHECK-NEXT: 1.     1     1.0    0.0    0.0       addq	(%rsp), %rsi
-# CHECK-NEXT: 2.     1     1.0    1.0    4.0       addq	%rdx, %r8
+# CHECK-NEXT: 2.     1     2.0    2.0    3.0       addq	%rdx, %r8
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
 
 # The register move from XMM0 to XMM1 can be eliminated at register renaming
 # stage. So, it should not consume pipeline resources.
@@ -10,12 +10,12 @@
 
 # CHECK:      Iterations:        3
 # CHECK-NEXT: Instructions:      9
-# CHECK-NEXT: Total Cycles:      9
+# CHECK-NEXT: Total Cycles:      11
 # CHECK-NEXT: Total uOps:        9
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               1.00
+# CHECK-NEXT: uOps Per Cycle:    0.82
+# CHECK-NEXT: IPC:               0.82
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
@@ -28,45 +28,68 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      0     0.25                        vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT:  1      1     1.00                        vmovaps	%xmm0, %xmm1
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT:  1      1     0.50                        vmovaps	%xmm0, %xmm1
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm1, %xmm1, %xmm2
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    9
-# CHECK-NEXT: Max number of mappings used:         8
+# CHECK-NEXT: Total number of mappings created:    6
+# CHECK-NEXT: Max number of mappings used:         6
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 6
+# CHECK-NEXT:    Max number of mappings used:      6
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   1.33    -      -      -      -     1.00   1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovaps	%xmm0, %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.33   0.67    -      -      -      -      -     1.00    -      -      -      -     vmovaps	%xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.33   0.67    -      -      -      -     1.00    -      -      -      -      -     vaddps	%xmm1, %xmm1, %xmm2
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     012345678
+# CHECK-NEXT:                     0
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DR   .  .   vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [0,1]     DeER .  .   vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: [0,2]     D=eeeER .   vaddps	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [1,0]     D-----R .   vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [1,1]     .DeE--R .   vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: [1,2]     .D=eeeER.   vaddps	%xmm1, %xmm1, %xmm2
-# CHECK-NEXT: [2,0]     .D-----R.   vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [2,1]     .D=eE--R.   vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: [2,2]     . D=eeeER   vaddps	%xmm1, %xmm1, %xmm2
+# CHECK:      [0,0]     DR   .    .   vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [0,1]     DeER .    .   vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: [0,2]     D=eeeeeER .   vaddps	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [1,0]     D-------R .   vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [1,1]     .DeE----R .   vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: [1,2]     .D=eeeeeER.   vaddps	%xmm1, %xmm1, %xmm2
+# CHECK-NEXT: [2,0]     .D-------R.   vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [2,1]     .D=eE----R.   vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: [2,2]     . D=eeeeeER   vaddps	%xmm1, %xmm1, %xmm2
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -75,6 +98,6 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     0.0    0.0    3.3       vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: 1.     3     1.3    1.3    1.3       vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: 0.     3     0.0    0.0    4.7       vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: 1.     3     1.3    1.3    2.7       vmovaps	%xmm0, %xmm1
 # CHECK-NEXT: 2.     3     2.0    0.0    0.0       vaddps	%xmm1, %xmm1, %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
 
 pxor %mm0, %mm0
 movq %mm0, %mm1
@@ -14,13 +14,13 @@
 
 # CHECK:      Iterations:        3
 # CHECK-NEXT: Instructions:      27
-# CHECK-NEXT: Total Cycles:      22
+# CHECK-NEXT: Total Cycles:      18
 # CHECK-NEXT: Total uOps:        27
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.23
-# CHECK-NEXT: IPC:               1.23
-# CHECK-NEXT: Block RThroughput: 4.0
+# CHECK-NEXT: uOps Per Cycle:    1.50
+# CHECK-NEXT: IPC:               1.50
+# CHECK-NEXT: Block RThroughput: 3.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -31,77 +31,99 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        pxor	%mm0, %mm0
-# CHECK-NEXT:  1      1     0.50                        movq	%mm0, %mm1
+# CHECK-NEXT:  1      0     0.25                        pxor	%mm0, %mm0
+# CHECK-NEXT:  1      2     0.50                        movq	%mm0, %mm1
 # CHECK-NEXT:  1      0     0.25                        xorps	%xmm0, %xmm0
-# CHECK-NEXT:  1      1     1.00                        movaps	%xmm0, %xmm1
-# CHECK-NEXT:  1      1     1.00                        movups	%xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        movapd	%xmm2, %xmm3
-# CHECK-NEXT:  1      1     1.00                        movupd	%xmm3, %xmm4
-# CHECK-NEXT:  1      1     0.33                        movdqa	%xmm4, %xmm5
-# CHECK-NEXT:  1      1     0.33                        movdqu	%xmm5, %xmm0
+# CHECK-NEXT:  1      1     0.50                        movaps	%xmm0, %xmm1
+# CHECK-NEXT:  1      1     0.50                        movups	%xmm1, %xmm2
+# CHECK-NEXT:  1      1     0.50                        movapd	%xmm2, %xmm3
+# CHECK-NEXT:  1      1     0.50                        movupd	%xmm3, %xmm4
+# CHECK-NEXT:  1      2     0.50                        movdqa	%xmm4, %xmm5
+# CHECK-NEXT:  1      2     0.50                        movdqu	%xmm5, %xmm0
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    27
-# CHECK-NEXT: Max number of mappings used:         21
+# CHECK-NEXT: Total number of mappings created:    21
+# CHECK-NEXT: Max number of mappings used:         16
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 21
+# CHECK-NEXT:    Max number of mappings used:      16
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.67   1.67    -     4.67    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.67   1.33    -     3.00    -      -     3.33   3.67    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     0.67    -     0.33    -      -     pxor	%mm0, %mm0
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movq	%mm0, %mm1
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorps	%xmm0, %xmm0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movaps	%xmm0, %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movups	%xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movapd	%xmm2, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movupd	%xmm3, %xmm4
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     movdqa	%xmm4, %xmm5
-# CHECK-NEXT:  -      -     0.67    -      -     0.33    -      -     movdqu	%xmm5, %xmm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pxor	%mm0, %mm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     movq	%mm0, %mm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorps	%xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   0.33    -      -      -      -     0.33   0.67    -      -      -      -     movaps	%xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   0.33    -      -      -      -     0.33   0.67    -      -      -      -     movups	%xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   0.33    -      -      -      -     1.00    -      -      -      -      -     movapd	%xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   0.33    -      -      -      -     0.33   0.67    -      -      -      -     movupd	%xmm3, %xmm4
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     movdqa	%xmm4, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     0.33   0.67    -      -      -      -     movdqu	%xmm5, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789
-# CHECK-NEXT: Index     0123456789          01
+# CHECK-NEXT:                     01234567
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    .    .    ..   pxor	%mm0, %mm0
-# CHECK-NEXT: [0,1]     D=eER.    .    .    ..   movq	%mm0, %mm1
-# CHECK-NEXT: [0,2]     D---R.    .    .    ..   xorps	%xmm0, %xmm0
-# CHECK-NEXT: [0,3]     D=eER.    .    .    ..   movaps	%xmm0, %xmm1
-# CHECK-NEXT: [0,4]     .D=eER    .    .    ..   movups	%xmm1, %xmm2
-# CHECK-NEXT: [0,5]     .D==eER   .    .    ..   movapd	%xmm2, %xmm3
-# CHECK-NEXT: [0,6]     .D===eER  .    .    ..   movupd	%xmm3, %xmm4
-# CHECK-NEXT: [0,7]     .D====eER .    .    ..   movdqa	%xmm4, %xmm5
-# CHECK-NEXT: [0,8]     . D====eER.    .    ..   movdqu	%xmm5, %xmm0
-# CHECK-NEXT: [1,0]     . DeE----R.    .    ..   pxor	%mm0, %mm0
-# CHECK-NEXT: [1,1]     . D=eE---R.    .    ..   movq	%mm0, %mm1
-# CHECK-NEXT: [1,2]     . D=====ER.    .    ..   xorps	%xmm0, %xmm0
-# CHECK-NEXT: [1,3]     .  D====eER    .    ..   movaps	%xmm0, %xmm1
-# CHECK-NEXT: [1,4]     .  D=====eER   .    ..   movups	%xmm1, %xmm2
-# CHECK-NEXT: [1,5]     .  D======eER  .    ..   movapd	%xmm2, %xmm3
-# CHECK-NEXT: [1,6]     .  D=======eER .    ..   movupd	%xmm3, %xmm4
-# CHECK-NEXT: [1,7]     .   D=======eER.    ..   movdqa	%xmm4, %xmm5
-# CHECK-NEXT: [1,8]     .   D========eER    ..   movdqu	%xmm5, %xmm0
-# CHECK-NEXT: [2,0]     .   DeE--------R    ..   pxor	%mm0, %mm0
-# CHECK-NEXT: [2,1]     .   D=eE-------R    ..   movq	%mm0, %mm1
-# CHECK-NEXT: [2,2]     .    D========ER    ..   xorps	%xmm0, %xmm0
-# CHECK-NEXT: [2,3]     .    D========eER   ..   movaps	%xmm0, %xmm1
-# CHECK-NEXT: [2,4]     .    D=========eER  ..   movups	%xmm1, %xmm2
-# CHECK-NEXT: [2,5]     .    D==========eER ..   movapd	%xmm2, %xmm3
-# CHECK-NEXT: [2,6]     .    .D==========eER..   movupd	%xmm3, %xmm4
-# CHECK-NEXT: [2,7]     .    .D===========eER.   movdqa	%xmm4, %xmm5
-# CHECK-NEXT: [2,8]     .    .D============eER   movdqu	%xmm5, %xmm0
+# CHECK:      [0,0]     DR   .    .    . .   pxor	%mm0, %mm0
+# CHECK-NEXT: [0,1]     DeeER.    .    . .   movq	%mm0, %mm1
+# CHECK-NEXT: [0,2]     D---R.    .    . .   xorps	%xmm0, %xmm0
+# CHECK-NEXT: [0,3]     DeE-R.    .    . .   movaps	%xmm0, %xmm1
+# CHECK-NEXT: [0,4]     .DeER.    .    . .   movups	%xmm1, %xmm2
+# CHECK-NEXT: [0,5]     .D=eER    .    . .   movapd	%xmm2, %xmm3
+# CHECK-NEXT: [0,6]     .D==eER   .    . .   movupd	%xmm3, %xmm4
+# CHECK-NEXT: [0,7]     .D===eeER .    . .   movdqa	%xmm4, %xmm5
+# CHECK-NEXT: [0,8]     . D====eeER    . .   movdqu	%xmm5, %xmm0
+# CHECK-NEXT: [1,0]     . D-------R    . .   pxor	%mm0, %mm0
+# CHECK-NEXT: [1,1]     . DeeE----R    . .   movq	%mm0, %mm1
+# CHECK-NEXT: [1,2]     . D-------R    . .   xorps	%xmm0, %xmm0
+# CHECK-NEXT: [1,3]     .  DeE-----R   . .   movaps	%xmm0, %xmm1
+# CHECK-NEXT: [1,4]     .  D=eE----R   . .   movups	%xmm1, %xmm2
+# CHECK-NEXT: [1,5]     .  D==eE---R   . .   movapd	%xmm2, %xmm3
+# CHECK-NEXT: [1,6]     .  D===eE--R   . .   movupd	%xmm3, %xmm4
+# CHECK-NEXT: [1,7]     .   D===eeE-R  . .   movdqa	%xmm4, %xmm5
+# CHECK-NEXT: [1,8]     .   D=====eeER . .   movdqu	%xmm5, %xmm0
+# CHECK-NEXT: [2,0]     .   D--------R . .   pxor	%mm0, %mm0
+# CHECK-NEXT: [2,1]     .   D=eeE----R . .   movq	%mm0, %mm1
+# CHECK-NEXT: [2,2]     .    D-------R . .   xorps	%xmm0, %xmm0
+# CHECK-NEXT: [2,3]     .    D==eE----R. .   movaps	%xmm0, %xmm1
+# CHECK-NEXT: [2,4]     .    D===eE---R. .   movups	%xmm1, %xmm2
+# CHECK-NEXT: [2,5]     .    D====eE--R. .   movapd	%xmm2, %xmm3
+# CHECK-NEXT: [2,6]     .    .D====eE-R. .   movupd	%xmm3, %xmm4
+# CHECK-NEXT: [2,7]     .    .D=====eeER .   movdqa	%xmm4, %xmm5
+# CHECK-NEXT: [2,8]     .    .D=======eeER   movdqu	%xmm5, %xmm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -110,12 +132,12 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.0    1.0    4.0       pxor	%mm0, %mm0
-# CHECK-NEXT: 1.     3     2.0    0.0    3.3       movq	%mm0, %mm1
-# CHECK-NEXT: 2.     3     5.0    0.0    1.0       xorps	%xmm0, %xmm0
-# CHECK-NEXT: 3.     3     5.3    0.7    0.0       movaps	%xmm0, %xmm1
-# CHECK-NEXT: 4.     3     6.0    0.0    0.0       movups	%xmm1, %xmm2
-# CHECK-NEXT: 5.     3     7.0    0.0    0.0       movapd	%xmm2, %xmm3
-# CHECK-NEXT: 6.     3     7.7    0.0    0.0       movupd	%xmm3, %xmm4
-# CHECK-NEXT: 7.     3     8.3    0.0    0.0       movdqa	%xmm4, %xmm5
-# CHECK-NEXT: 8.     3     9.0    0.0    0.0       movdqu	%xmm5, %xmm0
+# CHECK-NEXT: 0.     3     0.0    0.0    5.0       pxor	%mm0, %mm0
+# CHECK-NEXT: 1.     3     1.3    1.3    2.7       movq	%mm0, %mm1
+# CHECK-NEXT: 2.     3     0.0    0.0    5.7       xorps	%xmm0, %xmm0
+# CHECK-NEXT: 3.     3     1.7    1.7    3.3       movaps	%xmm0, %xmm1
+# CHECK-NEXT: 4.     3     2.3    0.0    2.3       movups	%xmm1, %xmm2
+# CHECK-NEXT: 5.     3     3.3    0.0    1.7       movapd	%xmm2, %xmm3
+# CHECK-NEXT: 6.     3     4.0    0.0    1.0       movupd	%xmm3, %xmm4
+# CHECK-NEXT: 7.     3     4.7    0.0    0.3       movdqa	%xmm4, %xmm5
+# CHECK-NEXT: 8.     3     6.3    0.0    0.0       movdqu	%xmm5, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-3.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-3.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-3.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
 
 vxorps  %xmm0, %xmm0, %xmm0
 vmovaps %xmm0, %xmm1
@@ -11,13 +11,13 @@
 
 # CHECK:      Iterations:        3
 # CHECK-NEXT: Instructions:      21
-# CHECK-NEXT: Total Cycles:      21
+# CHECK-NEXT: Total Cycles:      17
 # CHECK-NEXT: Total uOps:        21
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               1.00
-# CHECK-NEXT: Block RThroughput: 4.0
+# CHECK-NEXT: uOps Per Cycle:    1.24
+# CHECK-NEXT: IPC:               1.24
+# CHECK-NEXT: Block RThroughput: 3.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -29,66 +29,88 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      0     0.25                        vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT:  1      1     1.00                        vmovaps	%xmm0, %xmm1
-# CHECK-NEXT:  1      1     1.00                        vmovups	%xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovapd	%xmm2, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vmovupd	%xmm3, %xmm4
-# CHECK-NEXT:  1      1     0.33                        vmovdqa	%xmm4, %xmm5
-# CHECK-NEXT:  1      1     0.33                        vmovdqu	%xmm5, %xmm0
+# CHECK-NEXT:  1      1     0.50                        vmovaps	%xmm0, %xmm1
+# CHECK-NEXT:  1      1     0.50                        vmovups	%xmm1, %xmm2
+# CHECK-NEXT:  1      1     0.50                        vmovapd	%xmm2, %xmm3
+# CHECK-NEXT:  1      1     0.50                        vmovupd	%xmm3, %xmm4
+# CHECK-NEXT:  1      2     0.50                        vmovdqa	%xmm4, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vmovdqu	%xmm5, %xmm0
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    21
-# CHECK-NEXT: Max number of mappings used:         17
+# CHECK-NEXT: Total number of mappings created:    18
+# CHECK-NEXT: Max number of mappings used:         15
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 18
+# CHECK-NEXT:    Max number of mappings used:      15
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     4.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00   1.33   0.67    -      -     3.00   3.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovaps	%xmm0, %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovups	%xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovapd	%xmm2, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovupd	%xmm3, %xmm4
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmovdqa	%xmm4, %xmm5
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovdqu	%xmm5, %xmm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   0.33    -      -      -      -     0.67   0.33    -      -      -      -     vmovaps	%xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.33   0.67    -      -      -      -     0.67   0.33    -      -      -      -     vmovups	%xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.33   0.67    -      -      -      -     0.33   0.67    -      -      -      -     vmovapd	%xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.67   0.33    -      -      -      -     0.33   0.67    -      -      -      -     vmovupd	%xmm3, %xmm4
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.67   0.33    -      -     0.33   0.67    -      -      -      -     vmovdqa	%xmm4, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.67   0.33    -      -     0.67   0.33    -      -      -      -     vmovdqu	%xmm5, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789
-# CHECK-NEXT: Index     0123456789          0
+# CHECK-NEXT:                     0123456
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DR   .    .    .    .   vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [0,1]     DeER .    .    .    .   vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: [0,2]     D=eER.    .    .    .   vmovups	%xmm1, %xmm2
-# CHECK-NEXT: [0,3]     D==eER    .    .    .   vmovapd	%xmm2, %xmm3
-# CHECK-NEXT: [0,4]     .D==eER   .    .    .   vmovupd	%xmm3, %xmm4
-# CHECK-NEXT: [0,5]     .D===eER  .    .    .   vmovdqa	%xmm4, %xmm5
-# CHECK-NEXT: [0,6]     .D====eER .    .    .   vmovdqu	%xmm5, %xmm0
-# CHECK-NEXT: [1,0]     .D=====ER .    .    .   vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [1,1]     . D====eER.    .    .   vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: [1,2]     . D=====eER    .    .   vmovups	%xmm1, %xmm2
-# CHECK-NEXT: [1,3]     . D======eER   .    .   vmovapd	%xmm2, %xmm3
-# CHECK-NEXT: [1,4]     . D=======eER  .    .   vmovupd	%xmm3, %xmm4
-# CHECK-NEXT: [1,5]     .  D=======eER .    .   vmovdqa	%xmm4, %xmm5
-# CHECK-NEXT: [1,6]     .  D========eER.    .   vmovdqu	%xmm5, %xmm0
-# CHECK-NEXT: [2,0]     .  D=========ER.    .   vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [2,1]     .  D=========eER    .   vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: [2,2]     .   D=========eER   .   vmovups	%xmm1, %xmm2
-# CHECK-NEXT: [2,3]     .   D==========eER  .   vmovapd	%xmm2, %xmm3
-# CHECK-NEXT: [2,4]     .   D===========eER .   vmovupd	%xmm3, %xmm4
-# CHECK-NEXT: [2,5]     .   D============eER.   vmovdqa	%xmm4, %xmm5
-# CHECK-NEXT: [2,6]     .    D============eER   vmovdqu	%xmm5, %xmm0
+# CHECK:      [0,0]     DR   .    .    ..   vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [0,1]     DeER .    .    ..   vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: [0,2]     D=eER.    .    ..   vmovups	%xmm1, %xmm2
+# CHECK-NEXT: [0,3]     D==eER    .    ..   vmovapd	%xmm2, %xmm3
+# CHECK-NEXT: [0,4]     .D==eER   .    ..   vmovupd	%xmm3, %xmm4
+# CHECK-NEXT: [0,5]     .D===eeER .    ..   vmovdqa	%xmm4, %xmm5
+# CHECK-NEXT: [0,6]     .D=====eeER    ..   vmovdqu	%xmm5, %xmm0
+# CHECK-NEXT: [1,0]     .D--------R    ..   vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [1,1]     . DeE-----R    ..   vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: [1,2]     . D=eE----R    ..   vmovups	%xmm1, %xmm2
+# CHECK-NEXT: [1,3]     . D==eE----R   ..   vmovapd	%xmm2, %xmm3
+# CHECK-NEXT: [1,4]     . D===eE---R   ..   vmovupd	%xmm3, %xmm4
+# CHECK-NEXT: [1,5]     .  D===eeE-R   ..   vmovdqa	%xmm4, %xmm5
+# CHECK-NEXT: [1,6]     .  D=====eeER  ..   vmovdqu	%xmm5, %xmm0
+# CHECK-NEXT: [2,0]     .  D--------R  ..   vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [2,1]     .  D==eE----R  ..   vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: [2,2]     .   D===eE--R  ..   vmovups	%xmm1, %xmm2
+# CHECK-NEXT: [2,3]     .   D====eE--R ..   vmovapd	%xmm2, %xmm3
+# CHECK-NEXT: [2,4]     .   D=====eE-R ..   vmovupd	%xmm3, %xmm4
+# CHECK-NEXT: [2,5]     .   D======eeER..   vmovdqa	%xmm4, %xmm5
+# CHECK-NEXT: [2,6]     .    D=======eeER   vmovdqu	%xmm5, %xmm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -97,10 +119,10 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     5.3    0.0    0.0       vxorps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: 1.     3     5.3    0.3    0.0       vmovaps	%xmm0, %xmm1
-# CHECK-NEXT: 2.     3     6.0    0.0    0.0       vmovups	%xmm1, %xmm2
-# CHECK-NEXT: 3.     3     7.0    0.0    0.0       vmovapd	%xmm2, %xmm3
-# CHECK-NEXT: 4.     3     7.7    0.0    0.0       vmovupd	%xmm3, %xmm4
-# CHECK-NEXT: 5.     3     8.3    0.0    0.0       vmovdqa	%xmm4, %xmm5
-# CHECK-NEXT: 6.     3     9.0    0.0    0.0       vmovdqu	%xmm5, %xmm0
+# CHECK-NEXT: 0.     3     0.0    0.0    5.3       vxorps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: 1.     3     1.7    1.7    3.0       vmovaps	%xmm0, %xmm1
+# CHECK-NEXT: 2.     3     2.7    0.3    2.0       vmovups	%xmm1, %xmm2
+# CHECK-NEXT: 3.     3     3.7    0.0    2.0       vmovapd	%xmm2, %xmm3
+# CHECK-NEXT: 4.     3     4.3    0.0    1.3       vmovupd	%xmm3, %xmm4
+# CHECK-NEXT: 5.     3     5.0    0.0    0.3       vmovdqa	%xmm4, %xmm5
+# CHECK-NEXT: 6.     3     6.7    0.0    0.0       vmovdqu	%xmm5, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-4.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-4.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-4.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
 
 xor %eax, %eax
 mov %eax, %ebx
@@ -9,13 +9,13 @@
 
 # CHECK:      Iterations:        3
 # CHECK-NEXT: Instructions:      15
-# CHECK-NEXT: Total Cycles:      15
+# CHECK-NEXT: Total Cycles:      11
 # CHECK-NEXT: Total uOps:        15
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               1.00
-# CHECK-NEXT: Block RThroughput: 1.3
+# CHECK-NEXT: uOps Per Cycle:    1.36
+# CHECK-NEXT: IPC:               1.36
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -27,56 +27,78 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      0     0.25                        xorl	%eax, %eax
-# CHECK-NEXT:  1      1     0.33                        movl	%eax, %ebx
-# CHECK-NEXT:  1      1     0.33                        movl	%ebx, %ecx
-# CHECK-NEXT:  1      1     0.33                        movl	%ecx, %edx
-# CHECK-NEXT:  1      1     0.33                        movl	%edx, %eax
+# CHECK-NEXT:  1      1     0.50                        movl	%eax, %ebx
+# CHECK-NEXT:  1      1     0.50                        movl	%ebx, %ecx
+# CHECK-NEXT:  1      1     0.50                        movl	%ecx, %edx
+# CHECK-NEXT:  1      1     0.50                        movl	%edx, %eax
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    18
-# CHECK-NEXT: Max number of mappings used:         15
+# CHECK-NEXT: Total number of mappings created:    12
+# CHECK-NEXT: Max number of mappings used:         11
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 12
+# CHECK-NEXT:    Max number of mappings used:      11
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.33   1.33    -     1.33    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorl	%eax, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movl	%eax, %ebx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movl	%ebx, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movl	%ecx, %edx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movl	%edx, %eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	%eax, %eax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     movl	%eax, %ebx
+# CHECK-NEXT:  -      -      -      -      -     0.67   0.33    -      -      -      -      -      -      -      -      -      -      -      -      -     movl	%ebx, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.33   0.67    -      -      -      -      -      -      -      -      -      -      -      -      -     movl	%ecx, %edx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     movl	%edx, %eax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
+# CHECK-NEXT:                     0
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DR   .    .   .   xorl	%eax, %eax
-# CHECK-NEXT: [0,1]     DeER .    .   .   movl	%eax, %ebx
-# CHECK-NEXT: [0,2]     D=eER.    .   .   movl	%ebx, %ecx
-# CHECK-NEXT: [0,3]     D==eER    .   .   movl	%ecx, %edx
-# CHECK-NEXT: [0,4]     .D==eER   .   .   movl	%edx, %eax
-# CHECK-NEXT: [1,0]     .D===ER   .   .   xorl	%eax, %eax
-# CHECK-NEXT: [1,1]     .D===eER  .   .   movl	%eax, %ebx
-# CHECK-NEXT: [1,2]     .D====eER .   .   movl	%ebx, %ecx
-# CHECK-NEXT: [1,3]     . D====eER.   .   movl	%ecx, %edx
-# CHECK-NEXT: [1,4]     . D=====eER   .   movl	%edx, %eax
-# CHECK-NEXT: [2,0]     . D======ER   .   xorl	%eax, %eax
-# CHECK-NEXT: [2,1]     . D======eER  .   movl	%eax, %ebx
-# CHECK-NEXT: [2,2]     .  D======eER .   movl	%ebx, %ecx
-# CHECK-NEXT: [2,3]     .  D=======eER.   movl	%ecx, %edx
-# CHECK-NEXT: [2,4]     .  D========eER   movl	%edx, %eax
+# CHECK:      [0,0]     DR   .    .   xorl	%eax, %eax
+# CHECK-NEXT: [0,1]     DeER .    .   movl	%eax, %ebx
+# CHECK-NEXT: [0,2]     D=eER.    .   movl	%ebx, %ecx
+# CHECK-NEXT: [0,3]     D==eER    .   movl	%ecx, %edx
+# CHECK-NEXT: [0,4]     .D==eER   .   movl	%edx, %eax
+# CHECK-NEXT: [1,0]     .D----R   .   xorl	%eax, %eax
+# CHECK-NEXT: [1,1]     .DeE--R   .   movl	%eax, %ebx
+# CHECK-NEXT: [1,2]     .D=eE-R   .   movl	%ebx, %ecx
+# CHECK-NEXT: [1,3]     . D=eE-R  .   movl	%ecx, %edx
+# CHECK-NEXT: [1,4]     . D==eER  .   movl	%edx, %eax
+# CHECK-NEXT: [2,0]     . D----R  .   xorl	%eax, %eax
+# CHECK-NEXT: [2,1]     . D==eER  .   movl	%eax, %ebx
+# CHECK-NEXT: [2,2]     .  D==eER .   movl	%ebx, %ecx
+# CHECK-NEXT: [2,3]     .  D===eER.   movl	%ecx, %edx
+# CHECK-NEXT: [2,4]     .  D====eER   movl	%edx, %eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -85,8 +107,8 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     3.7    0.0    0.0       xorl	%eax, %eax
-# CHECK-NEXT: 1.     3     4.0    0.3    0.0       movl	%eax, %ebx
-# CHECK-NEXT: 2.     3     4.7    0.0    0.0       movl	%ebx, %ecx
-# CHECK-NEXT: 3.     3     5.3    0.0    0.0       movl	%ecx, %edx
-# CHECK-NEXT: 4.     3     6.0    0.0    0.0       movl	%edx, %eax
+# CHECK-NEXT: 0.     3     0.0    0.0    2.7       xorl	%eax, %eax
+# CHECK-NEXT: 1.     3     1.7    1.7    0.7       movl	%eax, %ebx
+# CHECK-NEXT: 2.     3     2.3    0.0    0.3       movl	%ebx, %ecx
+# CHECK-NEXT: 3.     3     3.0    0.0    0.3       movl	%ecx, %edx
+# CHECK-NEXT: 4.     3     3.7    0.0    0.0       movl	%edx, %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-5.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-5.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/reg-move-elimination-5.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=3 -timeline -register-file-stats < %s | FileCheck %s
 
 xor %rax, %rax
 mov %rax, %rbx
@@ -9,13 +9,13 @@
 
 # CHECK:      Iterations:        3
 # CHECK-NEXT: Instructions:      15
-# CHECK-NEXT: Total Cycles:      15
+# CHECK-NEXT: Total Cycles:      11
 # CHECK-NEXT: Total uOps:        15
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.00
-# CHECK-NEXT: IPC:               1.00
-# CHECK-NEXT: Block RThroughput: 1.3
+# CHECK-NEXT: uOps Per Cycle:    1.36
+# CHECK-NEXT: IPC:               1.36
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -27,56 +27,78 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      0     0.25                        xorq	%rax, %rax
-# CHECK-NEXT:  1      1     0.33                        movq	%rax, %rbx
-# CHECK-NEXT:  1      1     0.33                        movq	%rbx, %rcx
-# CHECK-NEXT:  1      1     0.33                        movq	%rcx, %rdx
-# CHECK-NEXT:  1      1     0.33                        movq	%rdx, %rax
+# CHECK-NEXT:  1      1     0.50                        movq	%rax, %rbx
+# CHECK-NEXT:  1      1     0.50                        movq	%rbx, %rcx
+# CHECK-NEXT:  1      1     0.50                        movq	%rcx, %rdx
+# CHECK-NEXT:  1      1     0.50                        movq	%rdx, %rax
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    18
-# CHECK-NEXT: Max number of mappings used:         15
+# CHECK-NEXT: Total number of mappings created:    12
+# CHECK-NEXT: Max number of mappings used:         11
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 12
+# CHECK-NEXT:    Max number of mappings used:      11
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.33   1.33    -     1.33    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorq	%rax, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movq	%rax, %rbx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movq	%rbx, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movq	%rcx, %rdx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movq	%rdx, %rax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	%rax, %rax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     movq	%rax, %rbx
+# CHECK-NEXT:  -      -      -      -      -     0.67   0.33    -      -      -      -      -      -      -      -      -      -      -      -      -     movq	%rbx, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.33   0.67    -      -      -      -      -      -      -      -      -      -      -      -      -     movq	%rcx, %rdx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -     movq	%rdx, %rax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
+# CHECK-NEXT:                     0
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DR   .    .   .   xorq	%rax, %rax
-# CHECK-NEXT: [0,1]     DeER .    .   .   movq	%rax, %rbx
-# CHECK-NEXT: [0,2]     D=eER.    .   .   movq	%rbx, %rcx
-# CHECK-NEXT: [0,3]     D==eER    .   .   movq	%rcx, %rdx
-# CHECK-NEXT: [0,4]     .D==eER   .   .   movq	%rdx, %rax
-# CHECK-NEXT: [1,0]     .D===ER   .   .   xorq	%rax, %rax
-# CHECK-NEXT: [1,1]     .D===eER  .   .   movq	%rax, %rbx
-# CHECK-NEXT: [1,2]     .D====eER .   .   movq	%rbx, %rcx
-# CHECK-NEXT: [1,3]     . D====eER.   .   movq	%rcx, %rdx
-# CHECK-NEXT: [1,4]     . D=====eER   .   movq	%rdx, %rax
-# CHECK-NEXT: [2,0]     . D======ER   .   xorq	%rax, %rax
-# CHECK-NEXT: [2,1]     . D======eER  .   movq	%rax, %rbx
-# CHECK-NEXT: [2,2]     .  D======eER .   movq	%rbx, %rcx
-# CHECK-NEXT: [2,3]     .  D=======eER.   movq	%rcx, %rdx
-# CHECK-NEXT: [2,4]     .  D========eER   movq	%rdx, %rax
+# CHECK:      [0,0]     DR   .    .   xorq	%rax, %rax
+# CHECK-NEXT: [0,1]     DeER .    .   movq	%rax, %rbx
+# CHECK-NEXT: [0,2]     D=eER.    .   movq	%rbx, %rcx
+# CHECK-NEXT: [0,3]     D==eER    .   movq	%rcx, %rdx
+# CHECK-NEXT: [0,4]     .D==eER   .   movq	%rdx, %rax
+# CHECK-NEXT: [1,0]     .D----R   .   xorq	%rax, %rax
+# CHECK-NEXT: [1,1]     .DeE--R   .   movq	%rax, %rbx
+# CHECK-NEXT: [1,2]     .D=eE-R   .   movq	%rbx, %rcx
+# CHECK-NEXT: [1,3]     . D=eE-R  .   movq	%rcx, %rdx
+# CHECK-NEXT: [1,4]     . D==eER  .   movq	%rdx, %rax
+# CHECK-NEXT: [2,0]     . D----R  .   xorq	%rax, %rax
+# CHECK-NEXT: [2,1]     . D==eER  .   movq	%rax, %rbx
+# CHECK-NEXT: [2,2]     .  D==eER .   movq	%rbx, %rcx
+# CHECK-NEXT: [2,3]     .  D===eER.   movq	%rcx, %rdx
+# CHECK-NEXT: [2,4]     .  D====eER   movq	%rdx, %rax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -85,8 +107,8 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     3.7    0.0    0.0       xorq	%rax, %rax
-# CHECK-NEXT: 1.     3     4.0    0.3    0.0       movq	%rax, %rbx
-# CHECK-NEXT: 2.     3     4.7    0.0    0.0       movq	%rbx, %rcx
-# CHECK-NEXT: 3.     3     5.3    0.0    0.0       movq	%rcx, %rdx
-# CHECK-NEXT: 4.     3     6.0    0.0    0.0       movq	%rdx, %rax
+# CHECK-NEXT: 0.     3     0.0    0.0    2.7       xorq	%rax, %rax
+# CHECK-NEXT: 1.     3     1.7    1.7    0.7       movq	%rax, %rbx
+# CHECK-NEXT: 2.     3     2.3    0.0    0.3       movq	%rbx, %rcx
+# CHECK-NEXT: 3.     3     3.0    0.0    0.3       movq	%rcx, %rdx
+# CHECK-NEXT: 4.     3     3.7    0.0    0.0       movq	%rdx, %rax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-1.s
@@ -1,17 +1,17 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=5 -instruction-info=false -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=5 -instruction-info=false -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
 
 vaddps %xmm0, %xmm0, %xmm0
 vmulps %xmm0, %xmm0, %xmm0
 
 # CHECK:      Iterations:        5
 # CHECK-NEXT: Instructions:      10
-# CHECK-NEXT: Total Cycles:      43
+# CHECK-NEXT: Total Cycles:      53
 # CHECK-NEXT: Total uOps:        10
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.23
-# CHECK-NEXT: IPC:               0.23
+# CHECK-NEXT: uOps Per Cycle:    0.19
+# CHECK-NEXT: IPC:               0.19
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Dynamic Dispatch Stall Cycles:
@@ -24,47 +24,69 @@
 
 # CHECK:      Dispatch Logic - number of cycles where we saw N micro opcodes dispatched:
 # CHECK-NEXT: [# dispatched], [# cycles]
-# CHECK-NEXT:  0,              40  (93.0%)
-# CHECK-NEXT:  2,              1  (2.3%)
-# CHECK-NEXT:  4,              2  (4.7%)
+# CHECK-NEXT:  0,              50  (94.3%)
+# CHECK-NEXT:  2,              1  (1.9%)
+# CHECK-NEXT:  4,              2  (3.8%)
 
 # CHECK:      Register File statistics:
 # CHECK-NEXT: Total number of mappings created:    10
 # CHECK-NEXT: Max number of mappings used:         10
 
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 10
+# CHECK-NEXT:    Max number of mappings used:      10
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
+
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     2.00    -      -      -      -     1.00   1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -      -     vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -      -      -     vmulps	%xmm0, %xmm0, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456789
-# CHECK-NEXT: Index     0123456789          0123456789          012
+# CHECK-NEXT:                     0123456789          0123456789          012
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeER    .    .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [0,1]     D===eeeeeER    .    .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [1,0]     D========eeeER .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [1,1]     D===========eeeeeER .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [2,0]     .D===============eeeER   .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [2,1]     .D==================eeeeeER   .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [3,0]     .D=======================eeeER.    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [3,1]     .D==========================eeeeeER.    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [4,0]     . D==============================eeeER  . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [4,1]     . D=================================eeeeeER   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK:      [0,0]     DeeeeeER  .    .    .    .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [0,1]     D=====eeeeeER  .    .    .    .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [1,0]     D==========eeeeeER  .    .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [1,1]     D===============eeeeeER  .    .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [2,0]     .D===================eeeeeER  .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [2,1]     .D========================eeeeeER  .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [3,0]     .D=============================eeeeeER  .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [3,1]     .D==================================eeeeeER  .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [4,0]     . D======================================eeeeeER  . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [4,1]     . D===========================================eeeeeER   vmulps	%xmm0, %xmm0, %xmm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -73,5 +95,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     5     16.2   0.2    0.0       vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: 1.     5     19.2   0.0    0.0       vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: 0.     5     20.2   0.2    0.0       vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: 1.     5     25.2   0.0    0.0       vmulps	%xmm0, %xmm0, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-2.s
@@ -1,21 +1,21 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -register-file-size=5 -iterations=5 -instruction-info=false -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -register-file-size=5 -iterations=5 -instruction-info=false -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
 
 vaddps %xmm0, %xmm0, %xmm0
 vmulps %xmm0, %xmm0, %xmm0
 
 # CHECK:      Iterations:        5
 # CHECK-NEXT: Instructions:      10
-# CHECK-NEXT: Total Cycles:      43
+# CHECK-NEXT: Total Cycles:      53
 # CHECK-NEXT: Total uOps:        10
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.23
-# CHECK-NEXT: IPC:               0.23
+# CHECK-NEXT: uOps Per Cycle:    0.19
+# CHECK-NEXT: IPC:               0.19
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Dynamic Dispatch Stall Cycles:
-# CHECK-NEXT: RAT     - Register unavailable:                      20  (46.5%)
+# CHECK-NEXT: RAT     - Register unavailable:                      26  (49.1%)
 # CHECK-NEXT: RCU     - Retire tokens unavailable:                 0
 # CHECK-NEXT: SCHEDQ  - Scheduler full:                            0
 # CHECK-NEXT: LQ      - Load queue full:                           0
@@ -24,47 +24,69 @@
 
 # CHECK:      Dispatch Logic - number of cycles where we saw N micro opcodes dispatched:
 # CHECK-NEXT: [# dispatched], [# cycles]
-# CHECK-NEXT:  0,              36  (83.7%)
-# CHECK-NEXT:  1,              6  (14.0%)
-# CHECK-NEXT:  4,              1  (2.3%)
+# CHECK-NEXT:  0,              46  (86.8%)
+# CHECK-NEXT:  1,              6  (11.3%)
+# CHECK-NEXT:  4,              1  (1.9%)
 
 # CHECK:      Register File statistics:
 # CHECK-NEXT: Total number of mappings created:    10
 # CHECK-NEXT: Max number of mappings used:         5
 
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 10
+# CHECK-NEXT:    Max number of mappings used:      5
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
+
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     2.00    -      -      -      -     1.00   1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -      -     vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -      -      -     vmulps	%xmm0, %xmm0, %xmm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456789
-# CHECK-NEXT: Index     0123456789          0123456789          012
+# CHECK-NEXT:                     0123456789          0123456789          012
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeER    .    .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [0,1]     D===eeeeeER    .    .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [1,0]     D========eeeER .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [1,1]     D===========eeeeeER .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [2,0]     .D===============eeeER   .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [2,1]     .    D==============eeeeeER   .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [3,0]     .    .    D==============eeeER.    .    . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [3,1]     .    .    .  D==============eeeeeER.    . .   vmulps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [4,0]     .    .    .    .  D==============eeeER  . .   vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: [4,1]     .    .    .    .    .D==============eeeeeER   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK:      [0,0]     DeeeeeER  .    .    .    .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [0,1]     D=====eeeeeER  .    .    .    .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [1,0]     D==========eeeeeER  .    .    .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [1,1]     D===============eeeeeER  .    .    .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [2,0]     .D===================eeeeeER  .    .    .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [2,1]     .    . D==================eeeeeER  .    .    .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [3,0]     .    .    . D==================eeeeeER  .    .    . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [3,1]     .    .    .    . D==================eeeeeER  .    . .   vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [4,0]     .    .    .    .    . D==================eeeeeER  . .   vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: [4,1]     .    .    .    .    .    . D==================eeeeeER   vmulps	%xmm0, %xmm0, %xmm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -73,5 +95,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     5     11.2   0.2    0.0       vaddps	%xmm0, %xmm0, %xmm0
-# CHECK-NEXT: 1.     5     12.2   0.0    0.0       vmulps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: 0.     5     14.0   0.2    0.0       vaddps	%xmm0, %xmm0, %xmm0
+# CHECK-NEXT: 1.     5     15.8   0.0    0.0       vmulps	%xmm0, %xmm0, %xmm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-3.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-3.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-3.s
@@ -1,17 +1,17 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -register-file-size=5 -iterations=2 -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -register-file-size=5 -iterations=2 -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
 
 idiv %eax
 
 # CHECK:      Iterations:        2
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      55
-# CHECK-NEXT: Total uOps:        2
+# CHECK-NEXT: Total Cycles:      42
+# CHECK-NEXT: Total uOps:        4
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.04
-# CHECK-NEXT: IPC:               0.04
-# CHECK-NEXT: Block RThroughput: 10.0
+# CHECK-NEXT: uOps Per Cycle:    0.10
+# CHECK-NEXT: IPC:               0.05
+# CHECK-NEXT: Block RThroughput: 25.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -22,10 +22,10 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      25    10.00                 U     idivl	%eax
+# CHECK-NEXT:  2      14    25.00                 U     idivl	%eax
 
 # CHECK:      Dynamic Dispatch Stall Cycles:
-# CHECK-NEXT: RAT     - Register unavailable:                      27  (49.1%)
+# CHECK-NEXT: RAT     - Register unavailable:                      16  (38.1%)
 # CHECK-NEXT: RCU     - Retire tokens unavailable:                 0
 # CHECK-NEXT: SCHEDQ  - Scheduler full:                            0
 # CHECK-NEXT: LQ      - Load queue full:                           0
@@ -34,37 +34,59 @@
 
 # CHECK:      Dispatch Logic - number of cycles where we saw N micro opcodes dispatched:
 # CHECK-NEXT: [# dispatched], [# cycles]
-# CHECK-NEXT:  0,              53  (96.4%)
-# CHECK-NEXT:  1,              2  (3.6%)
+# CHECK-NEXT:  0,              40  (95.2%)
+# CHECK-NEXT:  2,              2  (4.8%)
 
 # CHECK:      Register File statistics:
 # CHECK-NEXT: Total number of mappings created:    6
 # CHECK-NEXT: Max number of mappings used:         3
 
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 6
+# CHECK-NEXT:    Max number of mappings used:      3
+
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -     25.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     idivl	%eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -     25.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivl	%eax
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456789          01234
-# CHECK-NEXT: Index     0123456789          0123456789          0123456789
+# CHECK-NEXT:                     0123456789          0123456789
+# CHECK-NEXT: Index     0123456789          0123456789          01
 
-# CHECK:      [0,0]     DeeeeeeeeeeeeeeeeeeeeeeeeeER  .    .    .    .    .   .   idivl	%eax
-# CHECK-NEXT: [1,0]     .    .    .    .    .    . DeeeeeeeeeeeeeeeeeeeeeeeeeER   idivl	%eax
+# CHECK:      [0,0]     DeeeeeeeeeeeeeeER   .    .    .    .    ..   idivl	%eax
+# CHECK-NEXT: [1,0]     .    .    .    .D=========eeeeeeeeeeeeeeER   idivl	%eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -73,4 +95,4 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     1.0    1.0    0.0       idivl	%eax
+# CHECK-NEXT: 0.     2     5.5    5.5    0.0       idivl	%eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-4.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-4.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-4.s
@@ -1,17 +1,17 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=22 -dispatch-stats -register-file-stats -resource-pressure=false -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=22 -dispatch-stats -register-file-stats -resource-pressure=false -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 idiv %eax
 
 # CHECK:      Iterations:        22
 # CHECK-NEXT: Instructions:      22
-# CHECK-NEXT: Total Cycles:      553
-# CHECK-NEXT: Total uOps:        22
+# CHECK-NEXT: Total Cycles:      542
+# CHECK-NEXT: Total uOps:        44
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.04
+# CHECK-NEXT: uOps Per Cycle:    0.08
 # CHECK-NEXT: IPC:               0.04
-# CHECK-NEXT: Block RThroughput: 10.0
+# CHECK-NEXT: Block RThroughput: 25.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -22,7 +22,7 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      25    10.00                 U     idivl	%eax
+# CHECK-NEXT:  2      14    25.00                 U     idivl	%eax
 
 # CHECK:      Dynamic Dispatch Stall Cycles:
 # CHECK-NEXT: RAT     - Register unavailable:                      0
@@ -34,21 +34,30 @@
 
 # CHECK:      Dispatch Logic - number of cycles where we saw N micro opcodes dispatched:
 # CHECK-NEXT: [# dispatched], [# cycles]
-# CHECK-NEXT:  0,              547  (98.9%)
-# CHECK-NEXT:  2,              1  (0.2%)
-# CHECK-NEXT:  4,              5  (0.9%)
+# CHECK-NEXT:  0,              531  (98.0%)
+# CHECK-NEXT:  4,              11  (2.0%)
 
 # CHECK:      Register File statistics:
 # CHECK-NEXT: Total number of mappings created:    66
 # CHECK-NEXT: Max number of mappings used:         66
 
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 66
+# CHECK-NEXT:    Max number of mappings used:      66
+
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456789          0123456789          01234567
-# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456789
+# CHECK-NEXT:                     0123456789          0123456789          0123456789
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456
 
-# CHECK:      [0,0]     DeeeeeeeeeeeeeeeeeeeeeeeeeER  .    .    .    .    .    .    .    .    .    . .   idivl	%eax
-# CHECK-NEXT: [1,0]     D=========================eeeeeeeeeeeeeeeeeeeeeeeeeER  .    .    .    .    . .   idivl	%eax
-# CHECK-NEXT: [2,0]     D==================================================eeeeeeeeeeeeeeeeeeeeeeeeeER   idivl	%eax
+# CHECK:      [0,0]     DeeeeeeeeeeeeeeER   .    .    .    .    .    .    .    .    .    ..   idivl	%eax
+# CHECK-NEXT: [1,0]     D=========================eeeeeeeeeeeeeeER   .    .    .    .    ..   idivl	%eax
+# CHECK-NEXT: [2,0]     .D=================================================eeeeeeeeeeeeeeER   idivl	%eax
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -57,4 +66,4 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     26.0   0.3    0.0       idivl	%eax
+# CHECK-NEXT: 0.     3     25.7   7.7    0.0       idivl	%eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-5.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-5.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/register-files-5.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=false -instruction-info=false -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=false -instruction-info=false -dispatch-stats -register-file-stats -timeline < %s | FileCheck %s
 
   vdivps %ymm0, %ymm0, %ymm1
   vaddps %ymm0, %ymm0, %ymm2
@@ -37,13 +37,13 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      33
-# CHECK-NEXT: Total Cycles:      37
-# CHECK-NEXT: Total uOps:        35
+# CHECK-NEXT: Total Cycles:      70
+# CHECK-NEXT: Total uOps:        66
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.95
-# CHECK-NEXT: IPC:               0.89
-# CHECK-NEXT: Block RThroughput: 32.0
+# CHECK-NEXT: uOps Per Cycle:    0.94
+# CHECK-NEXT: IPC:               0.47
+# CHECK-NEXT: Block RThroughput: 64.0
 
 # CHECK:      Dynamic Dispatch Stall Cycles:
 # CHECK-NEXT: RAT     - Register unavailable:                      0
@@ -55,51 +55,61 @@
 
 # CHECK:      Dispatch Logic - number of cycles where we saw N micro opcodes dispatched:
 # CHECK-NEXT: [# dispatched], [# cycles]
-# CHECK-NEXT:  0,              28  (75.7%)
-# CHECK-NEXT:  3,              1  (2.7%)
-# CHECK-NEXT:  4,              8  (21.6%)
+# CHECK-NEXT:  0,              53  (75.7%)
+# CHECK-NEXT:  2,              1  (1.4%)
+# CHECK-NEXT:  4,              16  (22.9%)
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    33
-# CHECK-NEXT: Max number of mappings used:         33
+# CHECK-NEXT: Total number of mappings created:    66
+# CHECK-NEXT: Max number of mappings used:         54
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 66
+# CHECK-NEXT:    Max number of mappings used:      54
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          0123456
-# CHECK-NEXT: Index     0123456789          0123456789
+# CHECK-NEXT:                     0123456789          0123456789          0123456789
+# CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER   ..   vdivps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [0,1]     DeeeE--------------------------R   ..   vaddps	%ymm0, %ymm0, %ymm2
-# CHECK-NEXT: [0,2]     .DeeeE-------------------------R   ..   vaddps	%ymm0, %ymm0, %ymm3
-# CHECK-NEXT: [0,3]     .D=eeeE------------------------R   ..   vaddps	%ymm0, %ymm0, %ymm4
-# CHECK-NEXT: [0,4]     .D==eeeE-----------------------R   ..   vaddps	%ymm0, %ymm0, %ymm5
-# CHECK-NEXT: [0,5]     .D===eeeE----------------------R   ..   vaddps	%ymm0, %ymm0, %ymm6
-# CHECK-NEXT: [0,6]     . D===eeeE---------------------R   ..   vaddps	%ymm0, %ymm0, %ymm7
-# CHECK-NEXT: [0,7]     . D=====eeeE-------------------R   ..   vaddps	%ymm0, %ymm0, %ymm8
-# CHECK-NEXT: [0,8]     . D======eeeE------------------R   ..   vaddps	%ymm0, %ymm0, %ymm9
-# CHECK-NEXT: [0,9]     . D=======eeeE-----------------R   ..   vaddps	%ymm0, %ymm0, %ymm10
-# CHECK-NEXT: [0,10]    .  D=======eeeE----------------R   ..   vaddps	%ymm0, %ymm0, %ymm11
-# CHECK-NEXT: [0,11]    .  D========eeeE---------------R   ..   vaddps	%ymm0, %ymm0, %ymm12
-# CHECK-NEXT: [0,12]    .  D=========eeeE--------------R   ..   vaddps	%ymm0, %ymm0, %ymm13
-# CHECK-NEXT: [0,13]    .  D===========eeeE------------R   ..   vaddps	%ymm0, %ymm0, %ymm14
-# CHECK-NEXT: [0,14]    .   D===========eeeE-----------R   ..   vaddps	%ymm0, %ymm0, %ymm15
-# CHECK-NEXT: [0,15]    .   D==eeeE--------------------R   ..   vaddps	%ymm2, %ymm0, %ymm0
-# CHECK-NEXT: [0,16]    .   D=========eeeE-------------R   ..   vaddps	%ymm2, %ymm0, %ymm3
-# CHECK-NEXT: [0,17]    .   D============eeeE----------R   ..   vaddps	%ymm2, %ymm0, %ymm4
-# CHECK-NEXT: [0,18]    .    D============eeeE---------R   ..   vaddps	%ymm2, %ymm0, %ymm5
-# CHECK-NEXT: [0,19]    .    D=============eeeE--------R   ..   vaddps	%ymm2, %ymm0, %ymm6
-# CHECK-NEXT: [0,20]    .    D==============eeeE-------R   ..   vaddps	%ymm2, %ymm0, %ymm7
-# CHECK-NEXT: [0,21]    .    D===============eeeE------R   ..   vaddps	%ymm2, %ymm0, %ymm8
-# CHECK-NEXT: [0,22]    .    .D===============eeeE-----R   ..   vaddps	%ymm2, %ymm0, %ymm9
-# CHECK-NEXT: [0,23]    .    .D================eeeE----R   ..   vaddps	%ymm2, %ymm0, %ymm10
-# CHECK-NEXT: [0,24]    .    .D=================eeeE---R   ..   vaddps	%ymm2, %ymm0, %ymm11
-# CHECK-NEXT: [0,25]    .    .D==================eeeE--R   ..   vaddps	%ymm2, %ymm0, %ymm12
-# CHECK-NEXT: [0,26]    .    . D==================eeeE-R   ..   vaddps	%ymm2, %ymm0, %ymm13
-# CHECK-NEXT: [0,27]    .    . D===================eeeER   ..   vaddps	%ymm2, %ymm0, %ymm14
-# CHECK-NEXT: [0,28]    .    . D====================eeeER  ..   vaddps	%ymm2, %ymm0, %ymm15
-# CHECK-NEXT: [0,29]    .    . D=====================eeeER ..   vaddps	%ymm3, %ymm0, %ymm2
-# CHECK-NEXT: [0,30]    .    .  D=====================eeeER..   vaddps	%ymm3, %ymm0, %ymm4
-# CHECK-NEXT: [0,31]    .    .  D======================eeeER.   vaddps	%ymm3, %ymm0, %ymm5
-# CHECK-NEXT: [0,32]    .    .  D=======================eeeER   vaddps	%ymm3, %ymm0, %ymm6
+# CHECK:      [0,0]     DeeeeeeeeeER   .    .    .    .    .    .    .    .    .    .    .   .   vdivps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [0,1]     DeeeeeE----R   .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm2
+# CHECK-NEXT: [0,2]     .D=eeeeeE--R   .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm3
+# CHECK-NEXT: [0,3]     .D===eeeeeER   .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm4
+# CHECK-NEXT: [0,4]     . D====eeeeeER .    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm5
+# CHECK-NEXT: [0,5]     . D======eeeeeER    .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm6
+# CHECK-NEXT: [0,6]     .  D=======eeeeeER  .    .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm7
+# CHECK-NEXT: [0,7]     .  D===========eeeeeER   .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm8
+# CHECK-NEXT: [0,8]     .   D============eeeeeER .    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm9
+# CHECK-NEXT: [0,9]     .   D==============eeeeeER    .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm10
+# CHECK-NEXT: [0,10]    .    D===============eeeeeER  .    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm11
+# CHECK-NEXT: [0,11]    .    D=================eeeeeER.    .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm12
+# CHECK-NEXT: [0,12]    .    .D==================eeeeeER   .    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm13
+# CHECK-NEXT: [0,13]    .    .D======================eeeeeER    .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm14
+# CHECK-NEXT: [0,14]    .    . D=======================eeeeeER  .    .    .    .    .    .   .   vaddps	%ymm0, %ymm0, %ymm15
+# CHECK-NEXT: [0,15]    .    . D=====eeeeeE------------------R  .    .    .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm0
+# CHECK-NEXT: [0,16]    .    .  D==================eeeeeE----R  .    .    .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm3
+# CHECK-NEXT: [0,17]    .    .  D========================eeeeeER.    .    .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm4
+# CHECK-NEXT: [0,18]    .    .   D=========================eeeeeER   .    .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm5
+# CHECK-NEXT: [0,19]    .    .   D===========================eeeeeER .    .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm6
+# CHECK-NEXT: [0,20]    .    .    D============================eeeeeER    .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm7
+# CHECK-NEXT: [0,21]    .    .    D==============================eeeeeER  .    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm8
+# CHECK-NEXT: [0,22]    .    .    .D===============================eeeeeER.    .    .    .   .   vaddps	%ymm2, %ymm0, %ymm9
+# CHECK-NEXT: [0,23]    .    .    .D=================================eeeeeER   .    .    .   .   vaddps	%ymm2, %ymm0, %ymm10
+# CHECK-NEXT: [0,24]    .    .    . D==================================eeeeeER .    .    .   .   vaddps	%ymm2, %ymm0, %ymm11
+# CHECK-NEXT: [0,25]    .    .    . D====================================eeeeeER    .    .   .   vaddps	%ymm2, %ymm0, %ymm12
+# CHECK-NEXT: [0,26]    .    .    .  D=====================================eeeeeER  .    .   .   vaddps	%ymm2, %ymm0, %ymm13
+# CHECK-NEXT: [0,27]    .    .    .  D=======================================eeeeeER.    .   .   vaddps	%ymm2, %ymm0, %ymm14
+# CHECK-NEXT: [0,28]    .    .    .   D========================================eeeeeER   .   .   vaddps	%ymm2, %ymm0, %ymm15
+# CHECK-NEXT: [0,29]    .    .    .   D==========================================eeeeeER .   .   vaddps	%ymm3, %ymm0, %ymm2
+# CHECK-NEXT: [0,30]    .    .    .    D===========================================eeeeeER   .   vaddps	%ymm3, %ymm0, %ymm4
+# CHECK-NEXT: [0,31]    .    .    .    D=============================================eeeeeER .   vaddps	%ymm3, %ymm0, %ymm5
+# CHECK-NEXT: [0,32]    .    .    .    .D==============================================eeeeeER   vaddps	%ymm3, %ymm0, %ymm6
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -109,35 +119,35 @@
 
 # CHECK:            [0]    [1]    [2]    [3]
 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       vdivps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: 1.     1     1.0    1.0    26.0      vaddps	%ymm0, %ymm0, %ymm2
-# CHECK-NEXT: 2.     1     1.0    1.0    25.0      vaddps	%ymm0, %ymm0, %ymm3
-# CHECK-NEXT: 3.     1     2.0    2.0    24.0      vaddps	%ymm0, %ymm0, %ymm4
-# CHECK-NEXT: 4.     1     3.0    3.0    23.0      vaddps	%ymm0, %ymm0, %ymm5
-# CHECK-NEXT: 5.     1     4.0    4.0    22.0      vaddps	%ymm0, %ymm0, %ymm6
-# CHECK-NEXT: 6.     1     4.0    4.0    21.0      vaddps	%ymm0, %ymm0, %ymm7
-# CHECK-NEXT: 7.     1     6.0    6.0    19.0      vaddps	%ymm0, %ymm0, %ymm8
-# CHECK-NEXT: 8.     1     7.0    7.0    18.0      vaddps	%ymm0, %ymm0, %ymm9
-# CHECK-NEXT: 9.     1     8.0    8.0    17.0      vaddps	%ymm0, %ymm0, %ymm10
-# CHECK-NEXT: 10.    1     8.0    8.0    16.0      vaddps	%ymm0, %ymm0, %ymm11
-# CHECK-NEXT: 11.    1     9.0    9.0    15.0      vaddps	%ymm0, %ymm0, %ymm12
-# CHECK-NEXT: 12.    1     10.0   10.0   14.0      vaddps	%ymm0, %ymm0, %ymm13
-# CHECK-NEXT: 13.    1     12.0   12.0   12.0      vaddps	%ymm0, %ymm0, %ymm14
-# CHECK-NEXT: 14.    1     12.0   12.0   11.0      vaddps	%ymm0, %ymm0, %ymm15
-# CHECK-NEXT: 15.    1     3.0    3.0    20.0      vaddps	%ymm2, %ymm0, %ymm0
-# CHECK-NEXT: 16.    1     10.0   4.0    13.0      vaddps	%ymm2, %ymm0, %ymm3
-# CHECK-NEXT: 17.    1     13.0   7.0    10.0      vaddps	%ymm2, %ymm0, %ymm4
-# CHECK-NEXT: 18.    1     13.0   8.0    9.0       vaddps	%ymm2, %ymm0, %ymm5
-# CHECK-NEXT: 19.    1     14.0   9.0    8.0       vaddps	%ymm2, %ymm0, %ymm6
-# CHECK-NEXT: 20.    1     15.0   10.0   7.0       vaddps	%ymm2, %ymm0, %ymm7
-# CHECK-NEXT: 21.    1     16.0   11.0   6.0       vaddps	%ymm2, %ymm0, %ymm8
-# CHECK-NEXT: 22.    1     16.0   12.0   5.0       vaddps	%ymm2, %ymm0, %ymm9
-# CHECK-NEXT: 23.    1     17.0   13.0   4.0       vaddps	%ymm2, %ymm0, %ymm10
-# CHECK-NEXT: 24.    1     18.0   14.0   3.0       vaddps	%ymm2, %ymm0, %ymm11
-# CHECK-NEXT: 25.    1     19.0   15.0   2.0       vaddps	%ymm2, %ymm0, %ymm12
-# CHECK-NEXT: 26.    1     19.0   16.0   1.0       vaddps	%ymm2, %ymm0, %ymm13
-# CHECK-NEXT: 27.    1     20.0   17.0   0.0       vaddps	%ymm2, %ymm0, %ymm14
-# CHECK-NEXT: 28.    1     21.0   18.0   0.0       vaddps	%ymm2, %ymm0, %ymm15
-# CHECK-NEXT: 29.    1     22.0   12.0   0.0       vaddps	%ymm3, %ymm0, %ymm2
-# CHECK-NEXT: 30.    1     22.0   13.0   0.0       vaddps	%ymm3, %ymm0, %ymm4
-# CHECK-NEXT: 31.    1     23.0   14.0   0.0       vaddps	%ymm3, %ymm0, %ymm5
-# CHECK-NEXT: 32.    1     24.0   15.0   0.0       vaddps	%ymm3, %ymm0, %ymm6
+# CHECK-NEXT: 1.     1     1.0    1.0    4.0       vaddps	%ymm0, %ymm0, %ymm2
+# CHECK-NEXT: 2.     1     2.0    2.0    2.0       vaddps	%ymm0, %ymm0, %ymm3
+# CHECK-NEXT: 3.     1     4.0    4.0    0.0       vaddps	%ymm0, %ymm0, %ymm4
+# CHECK-NEXT: 4.     1     5.0    5.0    0.0       vaddps	%ymm0, %ymm0, %ymm5
+# CHECK-NEXT: 5.     1     7.0    7.0    0.0       vaddps	%ymm0, %ymm0, %ymm6
+# CHECK-NEXT: 6.     1     8.0    8.0    0.0       vaddps	%ymm0, %ymm0, %ymm7
+# CHECK-NEXT: 7.     1     12.0   12.0   0.0       vaddps	%ymm0, %ymm0, %ymm8
+# CHECK-NEXT: 8.     1     13.0   13.0   0.0       vaddps	%ymm0, %ymm0, %ymm9
+# CHECK-NEXT: 9.     1     15.0   15.0   0.0       vaddps	%ymm0, %ymm0, %ymm10
+# CHECK-NEXT: 10.    1     16.0   16.0   0.0       vaddps	%ymm0, %ymm0, %ymm11
+# CHECK-NEXT: 11.    1     18.0   18.0   0.0       vaddps	%ymm0, %ymm0, %ymm12
+# CHECK-NEXT: 12.    1     19.0   19.0   0.0       vaddps	%ymm0, %ymm0, %ymm13
+# CHECK-NEXT: 13.    1     23.0   23.0   0.0       vaddps	%ymm0, %ymm0, %ymm14
+# CHECK-NEXT: 14.    1     24.0   24.0   0.0       vaddps	%ymm0, %ymm0, %ymm15
+# CHECK-NEXT: 15.    1     6.0    6.0    18.0      vaddps	%ymm2, %ymm0, %ymm0
+# CHECK-NEXT: 16.    1     19.0   9.0    4.0       vaddps	%ymm2, %ymm0, %ymm3
+# CHECK-NEXT: 17.    1     25.0   15.0   0.0       vaddps	%ymm2, %ymm0, %ymm4
+# CHECK-NEXT: 18.    1     26.0   17.0   0.0       vaddps	%ymm2, %ymm0, %ymm5
+# CHECK-NEXT: 19.    1     28.0   19.0   0.0       vaddps	%ymm2, %ymm0, %ymm6
+# CHECK-NEXT: 20.    1     29.0   21.0   0.0       vaddps	%ymm2, %ymm0, %ymm7
+# CHECK-NEXT: 21.    1     31.0   23.0   0.0       vaddps	%ymm2, %ymm0, %ymm8
+# CHECK-NEXT: 22.    1     32.0   25.0   0.0       vaddps	%ymm2, %ymm0, %ymm9
+# CHECK-NEXT: 23.    1     34.0   27.0   0.0       vaddps	%ymm2, %ymm0, %ymm10
+# CHECK-NEXT: 24.    1     35.0   29.0   0.0       vaddps	%ymm2, %ymm0, %ymm11
+# CHECK-NEXT: 25.    1     37.0   31.0   0.0       vaddps	%ymm2, %ymm0, %ymm12
+# CHECK-NEXT: 26.    1     38.0   33.0   0.0       vaddps	%ymm2, %ymm0, %ymm13
+# CHECK-NEXT: 27.    1     40.0   35.0   0.0       vaddps	%ymm2, %ymm0, %ymm14
+# CHECK-NEXT: 28.    1     41.0   37.0   0.0       vaddps	%ymm2, %ymm0, %ymm15
+# CHECK-NEXT: 29.    1     43.0   25.0   0.0       vaddps	%ymm3, %ymm0, %ymm2
+# CHECK-NEXT: 30.    1     44.0   27.0   0.0       vaddps	%ymm3, %ymm0, %ymm4
+# CHECK-NEXT: 31.    1     46.0   29.0   0.0       vaddps	%ymm3, %ymm0, %ymm5
+# CHECK-NEXT: 32.    1     47.0   31.0   0.0       vaddps	%ymm3, %ymm0, %ymm6
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-3dnow.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-3dnow.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-3dnow.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 femms
 
@@ -87,122 +87,134 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  31     31    10.33   *      *      U     femms
-# CHECK-NEXT:  1      3     1.00                        pavgusb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pavgusb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pf2id	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pf2id	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pf2iw	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pf2iw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfacc	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfacc	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfadd	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfadd	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfcmpeq	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfcmpeq	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfcmpge	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfcmpge	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfcmpgt	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfcmpgt	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfmax	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfmax	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfmin	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfmin	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfmul	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfmul	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfnacc	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfnacc	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfpnacc	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfpnacc	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfrcp	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfrcp	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfrcpit1	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfrcpit1	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfrcpit2	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfrcpit2	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfrsqit1	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfrsqit1	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfrsqrt	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfrsqrt	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfsub	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfsub	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pfsubr	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pfsubr	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pi2fd	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pi2fd	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pi2fw	%mm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   pi2fw	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmulhrw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmulhrw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50    *      *      U     femms
+# CHECK-NEXT:  1      2     0.50                        pavgusb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pavgusb	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pf2id	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pf2id	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pf2iw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pf2iw	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfacc	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfacc	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfadd	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfadd	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfcmpeq	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfcmpeq	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfcmpge	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfcmpge	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfcmpgt	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfcmpgt	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfmax	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfmax	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfmin	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfmin	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfmul	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfmul	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfnacc	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfnacc	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfpnacc	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfpnacc	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfrcp	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfrcp	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfrcpit1	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfrcpit1	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfrcpit2	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfrcpit2	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfrsqit1	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfrsqit1	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfrsqrt	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfrsqrt	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfsub	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfsub	(%rax), %mm2
+# CHECK-NEXT:  1      5     1.00                        pfsubr	%mm0, %mm2
+# CHECK-NEXT:  1      10    1.00    *                   pfsubr	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pi2fd	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pi2fd	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pi2fw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pi2fw	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmulhrw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhrw	(%rax), %mm2
 # CHECK-NEXT:  1      5     0.50    *      *            prefetch	(%rax)
 # CHECK-NEXT:  1      5     0.50    *      *            prefetchw	(%rax)
-# CHECK-NEXT:  1      1     1.00                        pswapd	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   pswapd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pswapd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pswapd	(%rax), %mm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     12.33  54.33   -     12.33  13.00  13.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 13.00  13.00   -      -      -      -      -      -     17.50  17.50  2.00   2.00   2.00   8.00   38.50  10.50   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     10.33  10.33   -     10.33   -      -     femms
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pavgusb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pavgusb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pf2id	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pf2id	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pf2iw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pf2iw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfacc	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfacc	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfadd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfadd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfcmpeq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfcmpeq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfcmpge	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfcmpge	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfcmpgt	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfcmpgt	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfmax	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfmax	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfmin	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfmin	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfmul	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfmul	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfnacc	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfnacc	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfpnacc	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfpnacc	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfrcp	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfrcp	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfrcpit1	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfrcpit1	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfrcpit2	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfrcpit2	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfrsqit1	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfrsqit1	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfrsqrt	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfrsqrt	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfsub	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfsub	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pfsubr	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pfsubr	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pi2fd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pi2fd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pi2fw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pi2fw	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhrw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhrw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetch	(%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetchw	(%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     pswapd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   pswapd	(%rax), %mm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     femms
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgusb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgusb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pf2id	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pf2id	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pf2iw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pf2iw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfacc	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfacc	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfadd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfadd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfcmpeq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfcmpeq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfcmpge	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfcmpge	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfcmpgt	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfcmpgt	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfmax	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfmax	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfmin	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfmin	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfmul	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfmul	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfnacc	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfnacc	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfpnacc	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfpnacc	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrcp	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrcp	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrcpit1	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrcpit1	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrcpit2	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrcpit2	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrsqit1	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrsqit1	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrsqrt	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfrsqrt	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfsub	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfsub	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfsubr	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pfsubr	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pi2fd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pi2fd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pi2fw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pi2fw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhrw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhrw	(%rax), %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetch	(%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetchw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pswapd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pswapd	(%rax), %mm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-adx.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-adx.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-adx.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 adcx        %ebx, %ecx
 adcx        (%rbx), %ecx
@@ -20,36 +20,48 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      2     0.67                        adcxl	%ebx, %ecx
-# CHECK-NEXT:  3      7     0.67    *                   adcxl	(%rbx), %ecx
-# CHECK-NEXT:  2      2     0.67                        adcxq	%rbx, %rcx
-# CHECK-NEXT:  3      7     0.67    *                   adcxq	(%rbx), %rcx
-# CHECK-NEXT:  2      2     0.67                        adoxl	%ebx, %ecx
-# CHECK-NEXT:  3      7     0.67    *                   adoxl	(%rbx), %ecx
-# CHECK-NEXT:  2      2     0.67                        adoxq	%rbx, %rcx
-# CHECK-NEXT:  3      7     0.67    *                   adoxq	(%rbx), %rcx
+# CHECK-NEXT:  1      1     1.00                        adcxl	%ebx, %ecx
+# CHECK-NEXT:  1      5     1.00    *                   adcxl	(%rbx), %ecx
+# CHECK-NEXT:  1      1     1.00                        adcxq	%rbx, %rcx
+# CHECK-NEXT:  1      5     1.00    *                   adcxq	(%rbx), %rcx
+# CHECK-NEXT:  1      1     1.00                        adoxl	%ebx, %ecx
+# CHECK-NEXT:  1      5     1.00    *                   adoxl	(%rbx), %ecx
+# CHECK-NEXT:  1      1     1.00                        adoxq	%rbx, %rcx
+# CHECK-NEXT:  1      5     1.00    *                   adoxq	(%rbx), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     6.67   2.67    -     6.67   2.00   2.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 2.00   2.00    -      -      -     8.00   8.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcxl	%ebx, %ecx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adcxl	(%rbx), %ecx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcxq	%rbx, %rcx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adcxq	(%rbx), %rcx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adoxl	%ebx, %ecx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adoxl	(%rbx), %ecx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adoxq	%rbx, %rcx
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adoxq	(%rbx), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcxl	%ebx, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcxl	(%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcxq	%rbx, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcxq	(%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adoxl	%ebx, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adoxl	(%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adoxq	%rbx, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adoxq	(%rbx), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-aes.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-aes.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-aes.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 aesdec          %xmm0, %xmm2
 aesdec          (%rax), %xmm2
@@ -28,44 +28,56 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      7     1.00                        aesdec	%xmm0, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   aesdec	(%rax), %xmm2
-# CHECK-NEXT:  2      7     1.00                        aesdeclast	%xmm0, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   aesdeclast	(%rax), %xmm2
-# CHECK-NEXT:  2      7     1.00                        aesenc	%xmm0, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   aesenc	(%rax), %xmm2
-# CHECK-NEXT:  2      7     1.00                        aesenclast	%xmm0, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   aesenclast	(%rax), %xmm2
-# CHECK-NEXT:  2      12    2.00                        aesimc	%xmm0, %xmm2
-# CHECK-NEXT:  3      18    2.00    *                   aesimc	(%rax), %xmm2
-# CHECK-NEXT:  1      8     3.67                        aeskeygenassist	$22, %xmm0, %xmm2
-# CHECK-NEXT:  1      8     3.33    *                   aeskeygenassist	$22, (%rax), %xmm2
+# CHECK-NEXT:  2      9     1.00                        aesdec	%xmm0, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   aesdec	(%rax), %xmm2
+# CHECK-NEXT:  2      9     1.00                        aesdeclast	%xmm0, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   aesdeclast	(%rax), %xmm2
+# CHECK-NEXT:  2      9     1.00                        aesenc	%xmm0, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   aesenc	(%rax), %xmm2
+# CHECK-NEXT:  2      9     1.00                        aesenclast	%xmm0, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   aesenclast	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        aesimc	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   aesimc	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        aeskeygenassist	$22, %xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   aeskeygenassist	$22, (%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     9.67   9.67    -     21.67  3.00   3.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 3.00   3.00    -      -      -      -      -      -      -      -      -      -     12.00   -     12.00   -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     aesdec	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   aesdec	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     aesdeclast	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   aesdeclast	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     aesenc	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   aesenc	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     aesenclast	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   aesenclast	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     2.00    -      -     aesimc	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     2.00   0.50   0.50   aesimc	(%rax), %xmm2
-# CHECK-NEXT:  -      -     3.67   3.67    -     3.67    -      -     aeskeygenassist	$22, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     3.33   3.33    -     3.33   0.50   0.50   aeskeygenassist	$22, (%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesdec	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesdec	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesdeclast	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesdeclast	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesenc	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesenc	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesenclast	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesenclast	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesimc	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aesimc	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aeskeygenassist	$22, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     aeskeygenassist	$22, (%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-avx1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-avx1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-avx1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 vaddpd            %xmm0, %xmm1, %xmm2
 vaddpd            (%rax), %xmm1, %xmm2
@@ -1030,1402 +1030,1414 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vaddpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vaddpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vaddps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vaddps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vaddsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vaddsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vaddss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddsubpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vaddsubpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddsubpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vaddsubpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vaddsubps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vaddsubps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vaddsubps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vaddsubps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  2      7     1.00                        vaesdec	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   vaesdec	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00                        vaesdeclast	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   vaesdeclast	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00                        vaesenc	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   vaesenc	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00                        vaesenclast	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      13    1.00    *                   vaesenclast	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      12    2.00                        vaesimc	%xmm0, %xmm2
-# CHECK-NEXT:  3      18    2.00    *                   vaesimc	(%rax), %xmm2
-# CHECK-NEXT:  1      8     3.67                        vaeskeygenassist	$22, %xmm0, %xmm2
-# CHECK-NEXT:  1      8     3.33    *                   vaeskeygenassist	$22, (%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vandnpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vandnpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vandnpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vandnpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vandnps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vandnps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vandnps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vandnps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vandpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vandpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vandpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vandpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vandps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vandps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vandps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vandps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     0.50                        vblendpd	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vblendpd	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vblendpd	$11, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     0.50    *                   vblendpd	$11, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     0.50                        vblendps	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vblendps	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vblendps	$11, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     0.50    *                   vblendps	$11, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  2      2     1.00                        vblendvpd	%xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vblendvpd	%xmm3, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vblendvpd	%ymm3, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  3      9     1.00    *                   vblendvpd	%ymm3, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  2      2     1.00                        vblendvps	%xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vblendvps	%xmm3, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vblendvps	%ymm3, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  3      9     1.00    *                   vblendvps	%ymm3, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  2      7     1.00    *                   vbroadcastf128	(%rax), %ymm2
-# CHECK-NEXT:  1      7     0.50    *                   vbroadcastsd	(%rax), %ymm2
-# CHECK-NEXT:  1      6     0.50    *                   vbroadcastss	(%rax), %xmm2
-# CHECK-NEXT:  1      7     0.50    *                   vbroadcastss	(%rax), %ymm2
-# CHECK-NEXT:  1      3     1.00                        vcmppd	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcmppd	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcmppd	$0, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vcmppd	$0, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vcmpps	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcmpps	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcmpps	$0, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vcmpps	$0, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vcmpsd	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcmpsd	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcmpss	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcmpss	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vcomisd	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   vcomisd	(%rax), %xmm1
-# CHECK-NEXT:  2      2     1.00                        vcomiss	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   vcomiss	(%rax), %xmm1
-# CHECK-NEXT:  2      4     1.00                        vcvtdq2pd	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtdq2pd	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        vcvtdq2pd	%xmm0, %ymm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtdq2pd	(%rax), %ymm2
-# CHECK-NEXT:  1      3     1.00                        vcvtdq2ps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcvtdq2ps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvtdq2ps	%ymm0, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vcvtdq2ps	(%rax), %ymm2
-# CHECK-NEXT:  2      4     1.00                        vcvtpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtpd2dqx	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        vcvtpd2dq	%ymm0, %xmm2
-# CHECK-NEXT:  3      11    1.00    *                   vcvtpd2dqy	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        vcvtpd2ps	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtpd2psx	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        vcvtpd2ps	%ymm0, %xmm2
-# CHECK-NEXT:  3      11    1.00    *                   vcvtpd2psy	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvtps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcvtps2dq	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvtps2dq	%ymm0, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vcvtps2dq	(%rax), %ymm2
-# CHECK-NEXT:  2      2     1.00                        vcvtps2pd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vcvtps2pd	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        vcvtps2pd	%xmm0, %ymm2
-# CHECK-NEXT:  2      7     1.00    *                   vcvtps2pd	(%rax), %ymm2
-# CHECK-NEXT:  2      5     1.00                        vcvtsd2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        vcvtsd2si	%xmm0, %rcx
-# CHECK-NEXT:  3      10    1.00    *                   vcvtsd2si	(%rax), %ecx
-# CHECK-NEXT:  3      10    1.00    *                   vcvtsd2si	(%rax), %rcx
-# CHECK-NEXT:  2      4     1.00                        vcvtsd2ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtsd2ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaddpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vaddpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vaddpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaddps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vaddps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      5     1.00                        vaddsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaddsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaddss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddsubpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaddsubpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vaddsubpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vaddsubpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      5     1.00                        vaddsubps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaddsubps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vaddsubps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vaddsubps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  2      9     1.00                        vaesdec	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   vaesdec	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      9     1.00                        vaesdeclast	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   vaesdeclast	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      9     1.00                        vaesenc	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   vaesenc	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      9     1.00                        vaesenclast	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  2      14    1.00    *                   vaesenclast	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaesimc	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaesimc	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaeskeygenassist	$22, %xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vaeskeygenassist	$22, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vandnpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vandnpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vandnpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vandnpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vandnps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vandnps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vandnps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vandnps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vandpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vandpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vandpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vandpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vandps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vandps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vandps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vandps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vblendpd	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vblendpd	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vblendpd	$11, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vblendpd	$11, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vblendps	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vblendps	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vblendps	$11, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vblendps	$11, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     2.00                        vblendvpd	%xmm3, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     2.00    *                   vblendvpd	%xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     3.00                        vblendvpd	%ymm3, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     3.00    *                   vblendvpd	%ymm3, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     2.00                        vblendvps	%xmm3, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     2.00    *                   vblendvps	%xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     3.00                        vblendvps	%ymm3, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     3.00    *                   vblendvps	%ymm3, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  2      7     0.50    *                   vbroadcastf128	(%rax), %ymm2
+# CHECK-NEXT:  2      6     2.00    *                   vbroadcastsd	(%rax), %ymm2
+# CHECK-NEXT:  1      7     0.50    *                   vbroadcastss	(%rax), %xmm2
+# CHECK-NEXT:  2      6     2.00    *                   vbroadcastss	(%rax), %ymm2
+# CHECK-NEXT:  1      2     1.00                        vcmppd	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vcmppd	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     2.00                        vcmppd	$0, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     2.00    *                   vcmppd	$0, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     1.00                        vcmpps	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vcmpps	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     2.00                        vcmpps	$0, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     2.00    *                   vcmpps	$0, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     1.00                        vcmpsd	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vcmpsd	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00                        vcmpss	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vcmpss	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      1     1.00                        vcomisd	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vcomisd	(%rax), %xmm1
+# CHECK-NEXT:  2      1     1.00                        vcomiss	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vcomiss	(%rax), %xmm1
+# CHECK-NEXT:  2      8     1.00                        vcvtdq2pd	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   vcvtdq2pd	(%rax), %xmm2
+# CHECK-NEXT:  4      8     2.00                        vcvtdq2pd	%xmm0, %ymm2
+# CHECK-NEXT:  5      13    2.00    *                   vcvtdq2pd	(%rax), %ymm2
+# CHECK-NEXT:  1      4     1.00                        vcvtdq2ps	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vcvtdq2ps	(%rax), %xmm2
+# CHECK-NEXT:  2      4     2.00                        vcvtdq2ps	%ymm0, %ymm2
+# CHECK-NEXT:  2      9     2.00    *                   vcvtdq2ps	(%rax), %ymm2
+# CHECK-NEXT:  2      8     1.00                        vcvtpd2dq	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   vcvtpd2dqx	(%rax), %xmm2
+# CHECK-NEXT:  4      8     2.00                        vcvtpd2dq	%ymm0, %xmm2
+# CHECK-NEXT:  4      13    2.00    *                   vcvtpd2dqy	(%rax), %xmm2
+# CHECK-NEXT:  2      8     1.00                        vcvtpd2ps	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   vcvtpd2psx	(%rax), %xmm2
+# CHECK-NEXT:  4      8     2.00                        vcvtpd2ps	%ymm0, %xmm2
+# CHECK-NEXT:  4      13    2.00    *                   vcvtpd2psy	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        vcvtps2dq	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vcvtps2dq	(%rax), %xmm2
+# CHECK-NEXT:  2      4     2.00                        vcvtps2dq	%ymm0, %ymm2
+# CHECK-NEXT:  2      9     2.00    *                   vcvtps2dq	(%rax), %ymm2
+# CHECK-NEXT:  2      8     1.00                        vcvtps2pd	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   vcvtps2pd	(%rax), %xmm2
+# CHECK-NEXT:  4      8     2.00                        vcvtps2pd	%xmm0, %ymm2
+# CHECK-NEXT:  5      13    2.00    *                   vcvtps2pd	(%rax), %ymm2
+# CHECK-NEXT:  2      13    1.00                        vcvtsd2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        vcvtsd2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   vcvtsd2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   vcvtsd2si	(%rax), %rcx
+# CHECK-NEXT:  1      4     1.00                        vcvtsd2ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vcvtsd2ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  2      4     1.00                        vcvtsi2sdl	%ecx, %xmm0, %xmm2
 # CHECK-NEXT:  2      4     1.00                        vcvtsi2sdq	%rcx, %xmm0, %xmm2
 # CHECK-NEXT:  2      9     1.00    *                   vcvtsi2sdl	(%rax), %xmm0, %xmm2
 # CHECK-NEXT:  2      9     1.00    *                   vcvtsi2sdq	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vcvtsi2ssl	%ecx, %xmm0, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vcvtsi2ssq	%rcx, %xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtsi2ssl	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvtsi2ssq	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vcvtss2sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vcvtss2sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      5     1.00                        vcvtss2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        vcvtss2si	%xmm0, %rcx
-# CHECK-NEXT:  3      10    1.00    *                   vcvtss2si	(%rax), %ecx
-# CHECK-NEXT:  3      10    1.00    *                   vcvtss2si	(%rax), %rcx
-# CHECK-NEXT:  2      4     1.00                        vcvttpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   vcvttpd2dqx	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        vcvttpd2dq	%ymm0, %xmm2
-# CHECK-NEXT:  3      11    1.00    *                   vcvttpd2dqy	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vcvttps2dq	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvttps2dq	%ymm0, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vcvttps2dq	(%rax), %ymm2
-# CHECK-NEXT:  2      5     1.00                        vcvttsd2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        vcvttsd2si	%xmm0, %rcx
-# CHECK-NEXT:  3      10    1.00    *                   vcvttsd2si	(%rax), %ecx
-# CHECK-NEXT:  3      10    1.00    *                   vcvttsd2si	(%rax), %rcx
-# CHECK-NEXT:  2      5     1.00                        vcvttss2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        vcvttss2si	%xmm0, %rcx
-# CHECK-NEXT:  3      10    1.00    *                   vcvttss2si	(%rax), %ecx
-# CHECK-NEXT:  3      10    1.00    *                   vcvttss2si	(%rax), %rcx
-# CHECK-NEXT:  1      22    22.00                       vdivpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      28    22.00   *                   vdivpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      45    44.00                       vdivpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  4      52    44.00   *                   vdivpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      14    14.00                       vdivps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   vdivps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      29    28.00                       vdivps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  4      36    28.00   *                   vdivps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      22    22.00                       vdivsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      28    22.00   *                   vdivsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      14    14.00                       vdivss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   vdivss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      9     1.00                        vdppd	$22, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      15    1.00    *                   vdppd	$22, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  4      12    2.00                        vdpps	$22, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  5      18    2.00    *                   vdpps	$22, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  4      12    2.00                        vdpps	$22, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  5      19    2.00    *                   vdpps	$22, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vextractf128	$1, %ymm0, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vextractf128	$1, %ymm0, (%rax)
-# CHECK-NEXT:  2      3     1.00                        vextractps	$1, %xmm0, %ecx
-# CHECK-NEXT:  3      5     1.00           *            vextractps	$1, %xmm0, (%rax)
-# CHECK-NEXT:  3      5     2.00                        vhaddpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   vhaddpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vhaddpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  4      12    2.00    *                   vhaddpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  3      5     2.00                        vhaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   vhaddps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vhaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  4      12    2.00    *                   vhaddps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  3      5     2.00                        vhsubpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   vhsubpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vhsubpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  4      12    2.00    *                   vhsubpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  3      5     2.00                        vhsubps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   vhsubps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      5     2.00                        vhsubps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  4      12    2.00    *                   vhsubps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vinsertf128	$1, %xmm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      4     1.00                        vcvtsi2ssl	%ecx, %xmm0, %xmm2
+# CHECK-NEXT:  2      4     1.00                        vcvtsi2ssq	%rcx, %xmm0, %xmm2
+# CHECK-NEXT:  2      9     1.00    *                   vcvtsi2ssl	(%rax), %xmm0, %xmm2
+# CHECK-NEXT:  2      9     1.00    *                   vcvtsi2ssq	(%rax), %xmm0, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vcvtss2sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vcvtss2sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      13    1.00                        vcvtss2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        vcvtss2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   vcvtss2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   vcvtss2si	(%rax), %rcx
+# CHECK-NEXT:  2      8     1.00                        vcvttpd2dq	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   vcvttpd2dqx	(%rax), %xmm2
+# CHECK-NEXT:  4      8     2.00                        vcvttpd2dq	%ymm0, %xmm2
+# CHECK-NEXT:  4      13    2.00    *                   vcvttpd2dqy	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vcvttps2dq	(%rax), %xmm2
+# CHECK-NEXT:  2      4     2.00                        vcvttps2dq	%ymm0, %ymm2
+# CHECK-NEXT:  2      9     2.00    *                   vcvttps2dq	(%rax), %ymm2
+# CHECK-NEXT:  2      13    1.00                        vcvttsd2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        vcvttsd2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   vcvttsd2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   vcvttsd2si	(%rax), %rcx
+# CHECK-NEXT:  2      13    1.00                        vcvttss2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        vcvttss2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   vcvttss2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   vcvttss2si	(%rax), %rcx
+# CHECK-NEXT:  1      9     9.50                        vdivpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   vdivpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      9     19.00                       vdivpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      14    19.00   *                   vdivpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      9     9.50                        vdivps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   vdivps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      9     19.00                       vdivps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      14    19.00   *                   vdivps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      9     9.50                        vdivsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   vdivsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      9     9.50                        vdivss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   vdivss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  15     15    1.50                        vdppd	$22, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  17     20    1.50    *                   vdppd	$22, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  17     25    1.50                        vdpps	$22, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  18     30    1.50    *                   vdpps	$22, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  25     27    3.00                        vdpps	$22, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  29     32    3.00    *                   vdpps	$22, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vextractf128	$1, %ymm0, %xmm2
+# CHECK-NEXT:  2      7     0.50           *            vextractf128	$1, %ymm0, (%rax)
+# CHECK-NEXT:  2      13    1.00                        vextractps	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            vextractps	$1, %xmm0, (%rax)
+# CHECK-NEXT:  3      11    1.00                        vhaddpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   vhaddpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  8      11    2.00                        vhaddpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  10     16    2.00    *                   vhaddpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  3      11    1.00                        vhaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   vhaddps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  8      11    2.00                        vhaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  10     16    2.00    *                   vhaddps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  3      11    1.00                        vhsubpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   vhsubpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  8      11    2.00                        vhsubpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  10     16    2.00    *                   vhsubpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  3      11    1.00                        vhsubps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   vhsubps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  8      11    2.00                        vhsubps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  10     16    2.00    *                   vhsubps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  2      2     0.50                        vinsertf128	$1, %xmm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      7     0.50    *                   vinsertf128	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vinsertps	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vinsertps	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   vlddqu	(%rax), %xmm2
-# CHECK-NEXT:  1      7     0.50    *                   vlddqu	(%rax), %ymm2
-# CHECK-NEXT:  4      5     1.00    *      *      U     vldmxcsr	(%rax)
+# CHECK-NEXT:  1      2     0.50                        vinsertps	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vinsertps	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vlddqu	(%rax), %xmm2
+# CHECK-NEXT:  2      5     0.50    *                   vlddqu	(%rax), %ymm2
+# CHECK-NEXT:  1      5     0.50    *      *      U     vldmxcsr	(%rax)
 # CHECK-NEXT:  1      1     1.00    *      *      U     vmaskmovdqu	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   vmaskmovpd	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  3      9     1.00    *                   vmaskmovpd	(%rax), %ymm0, %ymm2
-# CHECK-NEXT:  3      5     1.00    *      *            vmaskmovpd	%xmm0, %xmm1, (%rax)
-# CHECK-NEXT:  3      5     1.00    *      *            vmaskmovpd	%ymm0, %ymm1, (%rax)
-# CHECK-NEXT:  3      8     1.00    *                   vmaskmovps	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  3      9     1.00    *                   vmaskmovps	(%rax), %ymm0, %ymm2
-# CHECK-NEXT:  3      5     1.00    *      *            vmaskmovps	%xmm0, %xmm1, (%rax)
-# CHECK-NEXT:  3      5     1.00    *      *            vmaskmovps	%ymm0, %ymm1, (%rax)
-# CHECK-NEXT:  1      3     1.00                        vmaxpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vmaxpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vmaxpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vmaxpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vmaxps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vmaxps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vmaxps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vmaxps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vmaxsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vmaxsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vmaxss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vmaxss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vminpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vminpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vminpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vminpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vminps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vminps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vminps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vminps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vminsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vminsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vminss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vminss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovapd	%xmm0, %xmm2
+# CHECK-NEXT:  1      6     1.00    *                   vmaskmovpd	(%rax), %xmm0, %xmm2
+# CHECK-NEXT:  2      6     2.00    *                   vmaskmovpd	(%rax), %ymm0, %ymm2
+# CHECK-NEXT:  18     6     2.00    *      *            vmaskmovpd	%xmm0, %xmm1, (%rax)
+# CHECK-NEXT:  34     6     2.00    *      *            vmaskmovpd	%ymm0, %ymm1, (%rax)
+# CHECK-NEXT:  1      6     1.00    *                   vmaskmovps	(%rax), %xmm0, %xmm2
+# CHECK-NEXT:  2      6     2.00    *                   vmaskmovps	(%rax), %ymm0, %ymm2
+# CHECK-NEXT:  18     6     2.00    *      *            vmaskmovps	%xmm0, %xmm1, (%rax)
+# CHECK-NEXT:  34     6     2.00    *      *            vmaskmovps	%ymm0, %ymm1, (%rax)
+# CHECK-NEXT:  1      2     1.00                        vmaxpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vmaxpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     2.00                        vmaxpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     2.00    *                   vmaxpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     1.00                        vmaxps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vmaxps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     2.00                        vmaxps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     2.00    *                   vmaxps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     1.00                        vmaxsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vmaxsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00                        vmaxss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vmaxss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00                        vminpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vminpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     2.00                        vminpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     2.00    *                   vminpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     1.00                        vminps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vminps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     2.00                        vminps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     2.00    *                   vminps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     1.00                        vminsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vminsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00                        vminss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   vminss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      1     0.50                        vmovapd	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            vmovapd	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovapd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovapd	%ymm0, %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovapd	%ymm0, (%rax)
-# CHECK-NEXT:  1      7     0.50    *                   vmovapd	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vmovaps	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovapd	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovapd	%ymm0, %ymm2
+# CHECK-NEXT:  4      1     1.00           *            vmovapd	%ymm0, (%rax)
+# CHECK-NEXT:  2      5     0.50    *                   vmovapd	(%rax), %ymm2
+# CHECK-NEXT:  1      1     0.50                        vmovaps	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            vmovaps	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovaps	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovaps	%ymm0, %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovaps	%ymm0, (%rax)
-# CHECK-NEXT:  1      7     0.50    *                   vmovaps	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vmovd	%eax, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   vmovd	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        vmovd	%xmm0, %ecx
-# CHECK-NEXT:  1      1     1.00           *            vmovd	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00                        vmovddup	%xmm0, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   vmovddup	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovddup	%ymm0, %ymm2
-# CHECK-NEXT:  1      7     0.50    *                   vmovddup	(%rax), %ymm2
-# CHECK-NEXT:  1      1     0.33                        vmovdqa	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovaps	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovaps	%ymm0, %ymm2
+# CHECK-NEXT:  4      1     1.00           *            vmovaps	%ymm0, (%rax)
+# CHECK-NEXT:  2      5     0.50    *                   vmovaps	(%rax), %ymm2
+# CHECK-NEXT:  2      10    0.50                        vmovd	%eax, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovd	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00                        vmovd	%xmm0, %ecx
+# CHECK-NEXT:  1      2     1.00           *            vmovd	%xmm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        vmovddup	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vmovddup	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovddup	%ymm0, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vmovddup	(%rax), %ymm2
+# CHECK-NEXT:  1      2     0.50                        vmovdqa	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            vmovdqa	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovdqa	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vmovdqa	%ymm0, %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovdqa	%ymm0, (%rax)
-# CHECK-NEXT:  1      7     0.50    *                   vmovdqa	(%rax), %ymm2
-# CHECK-NEXT:  1      1     0.33                        vmovdqu	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovdqa	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovdqa	%ymm0, %ymm2
+# CHECK-NEXT:  4      1     1.00           *            vmovdqa	%ymm0, (%rax)
+# CHECK-NEXT:  2      5     0.50    *                   vmovdqa	(%rax), %ymm2
+# CHECK-NEXT:  1      2     0.50                        vmovdqu	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            vmovdqu	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovdqu	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vmovdqu	%ymm0, %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovdqu	%ymm0, (%rax)
-# CHECK-NEXT:  1      7     0.50    *                   vmovdqu	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vmovhlps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovlhps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vmovhpd	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   vmovhpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vmovhps	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   vmovhps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vmovlpd	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   vmovlpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vmovlps	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   vmovlps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      2     1.00                        vmovmskpd	%xmm0, %ecx
-# CHECK-NEXT:  1      2     1.00                        vmovmskpd	%ymm0, %ecx
-# CHECK-NEXT:  1      2     1.00                        vmovmskps	%xmm0, %ecx
-# CHECK-NEXT:  1      2     1.00                        vmovmskps	%ymm0, %ecx
-# CHECK-NEXT:  1      1     1.00           *            vmovntdq	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            vmovntdq	%ymm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovntdqa	(%rax), %xmm2
-# CHECK-NEXT:  1      7     0.50    *                   vmovntdqa	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovntpd	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            vmovntpd	%ymm0, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            vmovntps	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            vmovntps	%ymm0, (%rax)
-# CHECK-NEXT:  1      1     0.33                        vmovq	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovq	%rax, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   vmovq	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        vmovq	%xmm0, %rcx
-# CHECK-NEXT:  1      1     1.00           *            vmovq	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00                        vmovsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vmovsd	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovshdup	%xmm0, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   vmovshdup	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovshdup	%ymm0, %ymm2
-# CHECK-NEXT:  1      7     0.50    *                   vmovshdup	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vmovsldup	%xmm0, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   vmovsldup	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovsldup	%ymm0, %ymm2
-# CHECK-NEXT:  1      7     0.50    *                   vmovsldup	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vmovss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            vmovss	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovss	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovupd	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovdqu	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovdqu	%ymm0, %ymm2
+# CHECK-NEXT:  8      1     1.00           *            vmovdqu	%ymm0, (%rax)
+# CHECK-NEXT:  2      5     0.50    *                   vmovdqu	(%rax), %ymm2
+# CHECK-NEXT:  1      2     0.50                        vmovhlps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vmovlhps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00           *            vmovhpd	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   vmovhpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00           *            vmovhps	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   vmovhps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00           *            vmovlpd	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   vmovlpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00           *            vmovlps	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   vmovlps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      10    1.00                        vmovmskpd	%xmm0, %ecx
+# CHECK-NEXT:  2      10    1.00                        vmovmskpd	%ymm0, %ecx
+# CHECK-NEXT:  2      10    1.00                        vmovmskps	%xmm0, %ecx
+# CHECK-NEXT:  2      10    1.00                        vmovmskps	%ymm0, %ecx
+# CHECK-NEXT:  1      2     1.00           *            vmovntdq	%xmm0, (%rax)
+# CHECK-NEXT:  4      2     2.00           *            vmovntdq	%ymm0, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   vmovntdqa	(%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovntdqa	(%rax), %ymm2
+# CHECK-NEXT:  1      3     1.00           *            vmovntpd	%xmm0, (%rax)
+# CHECK-NEXT:  4      3     2.00           *            vmovntpd	%ymm0, (%rax)
+# CHECK-NEXT:  1      3     1.00           *            vmovntps	%xmm0, (%rax)
+# CHECK-NEXT:  4      3     2.00           *            vmovntps	%ymm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        vmovq	%xmm0, %xmm2
+# CHECK-NEXT:  2      10    0.50                        vmovq	%rax, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovq	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00                        vmovq	%xmm0, %rcx
+# CHECK-NEXT:  1      2     1.00           *            vmovq	%xmm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        vmovsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00           *            vmovsd	%xmm0, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   vmovsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vmovshdup	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vmovshdup	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovshdup	%ymm0, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vmovshdup	(%rax), %ymm2
+# CHECK-NEXT:  1      2     0.50                        vmovsldup	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vmovsldup	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovsldup	%ymm0, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vmovsldup	(%rax), %ymm2
+# CHECK-NEXT:  1      2     0.50                        vmovss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     1.00           *            vmovss	%xmm0, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   vmovss	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50                        vmovupd	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            vmovupd	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovupd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovupd	%ymm0, %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovupd	%ymm0, (%rax)
-# CHECK-NEXT:  1      7     0.50    *                   vmovupd	(%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vmovups	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovupd	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovupd	%ymm0, %ymm2
+# CHECK-NEXT:  8      1     1.00           *            vmovupd	%ymm0, (%rax)
+# CHECK-NEXT:  2      5     0.50    *                   vmovupd	(%rax), %ymm2
+# CHECK-NEXT:  1      1     0.50                        vmovups	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            vmovups	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   vmovups	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vmovups	%ymm0, %ymm2
-# CHECK-NEXT:  1      1     1.00           *            vmovups	%ymm0, (%rax)
-# CHECK-NEXT:  1      7     0.50    *                   vmovups	(%rax), %ymm2
-# CHECK-NEXT:  3      7     1.00                        vmpsadbw	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      13    1.00    *                   vmpsadbw	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   vmovups	(%rax), %xmm2
+# CHECK-NEXT:  2      2     1.00                        vmovups	%ymm0, %ymm2
+# CHECK-NEXT:  8      1     1.00           *            vmovups	%ymm0, (%rax)
+# CHECK-NEXT:  2      5     0.50    *                   vmovups	(%rax), %ymm2
+# CHECK-NEXT:  9      9     2.00                        vmpsadbw	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  9      14    2.00    *                   vmpsadbw	$1, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     1.00                        vmulpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vmulpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vmulpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      12    1.00    *                   vmulpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    1.00    *                   vmulpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vmulpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vmulpd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     1.00                        vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vmulps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      12    1.00    *                   vmulps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    1.00    *                   vmulps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vmulps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     1.00                        vmulsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vmulsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vmulsd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     1.00                        vmulss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vmulss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vorpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vorpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vorpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vorpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vorps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vorps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vorps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vorps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     0.50                        vpabsb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpabsb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpabsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpabsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpabsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpabsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpackssdw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpackssdw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpacksswb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpacksswb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpackusdw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpackusdw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpackuswb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpackuswb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddusb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddusb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddusw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddusw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpaddw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpaddw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpalignr	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpalignr	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.33                        vpand	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpand	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.33                        vpandn	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpandn	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpavgb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpavgb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpavgw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpavgw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpblendvb	%xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpblendvb	%xmm3, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpblendw	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpblendw	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      14    6.00                        vpclmulqdq	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  1      14    5.67    *                   vpclmulqdq	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpeqb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpeqd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpeqq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpeqw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpeqw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      4     2.67                        vpcmpestri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  1      4     2.33    *                   vpcmpestri	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      11    2.67                        vpcmpestrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  1      11    2.33    *                   vpcmpestrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpgtb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpgtb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpgtd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpgtd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpcmpgtq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpcmpgtq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpcmpgtw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpcmpgtw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      11    3.00                        vpcmpistri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  4      17    3.00    *                   vpcmpistri	$1, (%rax), %xmm2
-# CHECK-NEXT:  3      11    3.00                        vpcmpistrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  4      17    3.00    *                   vpcmpistrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vperm2f128	$1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vperm2f128	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vpermilpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vpermilpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpermilpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vpermilpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpermilpd	$1, %ymm0, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vpermilpd	$1, (%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vpermilpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vpermilpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vpermilps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vpermilps	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpermilps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vpermilps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpermilps	$1, %ymm0, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vpermilps	$1, (%rax), %ymm2
-# CHECK-NEXT:  1      1     1.00                        vpermilps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vpermilps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  2      3     1.00                        vpextrb	$1, %xmm0, %ecx
-# CHECK-NEXT:  3      5     1.00           *            vpextrb	$1, %xmm0, (%rax)
-# CHECK-NEXT:  2      3     1.00                        vpextrd	$1, %xmm0, %ecx
-# CHECK-NEXT:  4      5     1.00           *            vpextrd	$1, %xmm0, (%rax)
-# CHECK-NEXT:  2      3     1.00                        vpextrq	$1, %xmm0, %rcx
-# CHECK-NEXT:  4      5     1.00           *            vpextrq	$1, %xmm0, (%rax)
-# CHECK-NEXT:  2      3     1.00                        vpextrw	$1, %xmm0, %ecx
-# CHECK-NEXT:  3      5     1.00           *            vpextrw	$1, %xmm0, (%rax)
-# CHECK-NEXT:  3      3     1.50                        vphaddd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   vphaddd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      3     1.50                        vphaddsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   vphaddsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      3     1.50                        vphaddw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   vphaddw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vphminposuw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vphminposuw	(%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        vphsubd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   vphsubd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      3     1.50                        vphsubsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   vphsubsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  3      3     1.50                        vphsubw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   vphsubw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpinsrb	$1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpinsrb	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpinsrd	$1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpinsrd	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpinsrq	$1, %rax, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpinsrq	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpinsrw	$1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpinsrw	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmaddubsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmaddubsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmaddwd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmaddwd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmaxsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmaxsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmaxsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmaxsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmaxsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmaxsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmaxub	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmaxub	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmaxud	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmaxud	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmaxuw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmaxuw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpminsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpminsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpminsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpminsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpminsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpminsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpminub	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpminub	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpminud	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpminud	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpminuw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpminuw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      2     1.00                        vpmovmskb	%xmm0, %ecx
-# CHECK-NEXT:  1      1     0.50                        vpmovsxbd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovsxbd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovsxbq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovsxbq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovsxbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovsxbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovsxdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovsxdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovsxwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovsxwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovsxwq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovsxwq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovzxbd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovzxbd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovzxbq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovzxbq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovzxbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovzxbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovzxdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovzxdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovzxwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovzxwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpmovzxwq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpmovzxwq	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmuldq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmuldq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmulhrsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmulhrsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmulhuw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmulhuw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmulhw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmulhw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmulld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmulld	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmullw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmullw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpmuludq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpmuludq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.33                        vpor	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpor	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vpsadbw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vpsadbw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpshufb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpshufb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpshufd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpshufd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpshufhw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpshufhw	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpshuflw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpshuflw	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsignb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsignb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsignd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsignd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsignw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsignw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpslld	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpslld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpslld	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpslldq	$1, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsllq	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsllq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsllq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsllw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsllw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsllw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsrad	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsrad	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsrad	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsraw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsraw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsraw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsrld	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsrld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsrld	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsrldq	$1, %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsrlq	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsrlq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsrlq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vpsrlw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vpsrlw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   vpsrlw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubusb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubusb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubusw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubusw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpsubw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        vptest	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   vptest	(%rax), %xmm1
-# CHECK-NEXT:  2      2     1.00                        vptest	%ymm0, %ymm1
-# CHECK-NEXT:  3      9     1.00    *                   vptest	(%rax), %ymm1
-# CHECK-NEXT:  1      1     0.50                        vpunpckhbw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpckhbw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpckhdq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpckhdq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpckhqdq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpckhqdq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpckhwd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpckhwd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpcklbw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpcklbw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpckldq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpckldq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpcklqdq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpcklqdq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpunpcklwd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpunpcklwd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     0.33                        vpxor	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   vpxor	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vmulss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vorpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vorpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vorpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vorpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vorps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vorps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vorps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vorps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vpabsb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpabsb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpabsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpabsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpabsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpabsw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpackssdw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpackssdw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpacksswb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpacksswb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpackusdw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpackusdw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpackuswb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpackuswb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddusb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddusb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddusw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddusw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpaddw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpaddw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpalignr	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpalignr	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpand	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpandn	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpandn	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpavgb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpavgb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpavgw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpavgw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     2.00                        vpblendvb	%xmm3, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     2.00    *                   vpblendvb	%xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpblendw	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpblendw	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  6      13    1.00                        vpclmulqdq	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  6      17    1.00    *                   vpclmulqdq	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpeqb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpeqd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpeqq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpeqw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpeqw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  27     15    4.00                        vpcmpestri	$1, %xmm0, %xmm2
+# CHECK-NEXT:  28     20    4.50    *                   vpcmpestri	$1, (%rax), %xmm2
+# CHECK-NEXT:  27     10    4.00                        vpcmpestrm	$1, %xmm0, %xmm2
+# CHECK-NEXT:  28     15    4.50    *                   vpcmpestrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpgtb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpgtd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpgtq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpcmpgtw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  7      14    1.00                        vpcmpistri	$1, %xmm0, %xmm2
+# CHECK-NEXT:  8      19    1.00    *                   vpcmpistri	$1, (%rax), %xmm2
+# CHECK-NEXT:  7      6     1.00                        vpcmpistrm	$1, %xmm0, %xmm2
+# CHECK-NEXT:  9      11    1.00    *                   vpcmpistrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  8      4     0.50                        vperm2f128	$1, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  10     8     0.50    *                   vperm2f128	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vpermilpd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpermilpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      3     2.00                        vpermilpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     2.00    *                   vpermilpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vpermilpd	$1, %ymm0, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vpermilpd	$1, (%rax), %ymm2
+# CHECK-NEXT:  2      3     3.00                        vpermilpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      8     3.00    *                   vpermilpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vpermilps	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpermilps	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      3     2.00                        vpermilps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     2.00    *                   vpermilps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vpermilps	$1, %ymm0, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vpermilps	$1, (%rax), %ymm2
+# CHECK-NEXT:  2      3     3.00                        vpermilps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      8     3.00    *                   vpermilps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  2      13    1.00                        vpextrb	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            vpextrb	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      13    1.00                        vpextrd	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            vpextrd	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      13    1.00                        vpextrq	$1, %xmm0, %rcx
+# CHECK-NEXT:  2      13    1.00           *            vpextrq	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      13    1.00                        vpextrw	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            vpextrw	$1, %xmm0, (%rax)
+# CHECK-NEXT:  3      5     0.50                        vphaddd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   vphaddd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  3      5     0.50                        vphaddsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   vphaddsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  3      5     0.50                        vphaddw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   vphaddw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      4     1.00                        vphminposuw	%xmm0, %xmm2
+# CHECK-NEXT:  2      9     1.00    *                   vphminposuw	(%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        vphsubd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   vphsubd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  3      5     0.50                        vphsubsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   vphsubsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  3      5     0.50                        vphsubw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   vphsubw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     0.50                        vpinsrb	$1, %eax, %xmm1, %xmm2
+# CHECK-NEXT:  2      6     0.50    *                   vpinsrb	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     0.50                        vpinsrd	$1, %eax, %xmm1, %xmm2
+# CHECK-NEXT:  2      6     0.50    *                   vpinsrd	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     0.50                        vpinsrq	$1, %rax, %xmm1, %xmm2
+# CHECK-NEXT:  2      6     0.50    *                   vpinsrq	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     0.50                        vpinsrw	$1, %eax, %xmm1, %xmm2
+# CHECK-NEXT:  2      6     0.50    *                   vpinsrw	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmaddubsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmaddubsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmaddwd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmaddwd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmaxsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmaxsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmaxsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmaxsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmaxsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmaxsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmaxub	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmaxub	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmaxud	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmaxud	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmaxuw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmaxuw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpminsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpminsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpminsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpminsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpminsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpminsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpminub	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpminub	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpminud	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpminud	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpminuw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpminuw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      13    1.00                        vpmovmskb	%xmm0, %ecx
+# CHECK-NEXT:  1      2     0.50                        vpmovsxbd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovsxbd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovsxbq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovsxbq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovsxbw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovsxbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovsxdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovsxdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovsxwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovsxwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovsxwq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovsxwq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovzxbd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovzxbd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovzxbq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovzxbq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovzxbw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovzxbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovzxdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovzxdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovzxwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovzxwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpmovzxwq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpmovzxwq	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmuldq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmuldq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmulhrsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmulhrsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmulhuw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmulhuw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmulhw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmulhw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     2.00                        vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    2.00    *                   vpmulld	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmullw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmullw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vpmuludq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vpmuludq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpor	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpor	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      4     0.50                        vpsadbw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  2      9     0.50    *                   vpsadbw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      3     2.00                        vpshufb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     2.00    *                   vpshufb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpshufd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpshufd	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpshufhw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpshufhw	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpshuflw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpshuflw	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsignb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsignb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsignd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsignd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsignw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsignw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpslld	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpslld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpslld	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpslldq	$1, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsllq	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsllq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsllq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsllw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsllw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsllw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsrad	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsrad	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsrad	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsraw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsraw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsraw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsrld	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsrld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsrld	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsrldq	$1, %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsrlq	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsrlq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsrlq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsrlw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        vpsrlw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   vpsrlw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubusb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubusb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubusw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubusw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpsubw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      1     1.00                        vptest	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vptest	(%rax), %xmm1
+# CHECK-NEXT:  4      1     1.00                        vptest	%ymm0, %ymm1
+# CHECK-NEXT:  6      6     1.00    *                   vptest	(%rax), %ymm1
+# CHECK-NEXT:  1      2     0.50                        vpunpckhbw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpckhbw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpckhdq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpckhdq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpckhqdq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpckhqdq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpckhwd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpckhwd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpcklbw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpcklbw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpckldq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpckldq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpcklqdq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpcklqdq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpunpcklwd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpunpcklwd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpxor	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vpxor	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     1.00                        vrcpps	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vrcpps	(%rax), %xmm2
-# CHECK-NEXT:  3      7     2.00                        vrcpps	%ymm0, %ymm2
-# CHECK-NEXT:  4      14    2.00    *                   vrcpps	(%rax), %ymm2
+# CHECK-NEXT:  1      10    1.00    *                   vrcpps	(%rax), %xmm2
+# CHECK-NEXT:  2      5     2.00                        vrcpps	%ymm0, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vrcpps	(%rax), %ymm2
 # CHECK-NEXT:  1      5     1.00                        vrcpss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vrcpss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vroundpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vroundpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vroundpd	$1, %ymm0, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vroundpd	$1, (%rax), %ymm2
-# CHECK-NEXT:  1      3     1.00                        vroundps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vroundps	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vroundps	$1, %ymm0, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vroundps	$1, (%rax), %ymm2
-# CHECK-NEXT:  1      3     1.00                        vroundsd	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vroundsd	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vroundss	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vroundss	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vrcpss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vroundpd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vroundpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  2      4     2.00                        vroundpd	$1, %ymm0, %ymm2
+# CHECK-NEXT:  2      9     2.00    *                   vroundpd	$1, (%rax), %ymm2
+# CHECK-NEXT:  1      4     1.00                        vroundps	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vroundps	$1, (%rax), %xmm2
+# CHECK-NEXT:  2      4     2.00                        vroundps	$1, %ymm0, %ymm2
+# CHECK-NEXT:  2      9     2.00    *                   vroundps	$1, (%rax), %ymm2
+# CHECK-NEXT:  1      4     1.00                        vroundsd	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vroundsd	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      4     1.00                        vroundss	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   vroundss	$1, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     1.00                        vrsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vrsqrtps	(%rax), %xmm2
-# CHECK-NEXT:  3      7     2.00                        vrsqrtps	%ymm0, %ymm2
-# CHECK-NEXT:  4      14    2.00    *                   vrsqrtps	(%rax), %ymm2
+# CHECK-NEXT:  1      10    1.00    *                   vrsqrtps	(%rax), %xmm2
+# CHECK-NEXT:  2      5     2.00                        vrsqrtps	%ymm0, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vrsqrtps	(%rax), %ymm2
 # CHECK-NEXT:  1      5     1.00                        vrsqrtss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   vrsqrtss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vshufpd	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vshufpd	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vshufpd	$1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vshufpd	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vshufps	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vshufps	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vshufps	$1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vshufps	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      21    21.00                       vsqrtpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      27    21.00   *                   vsqrtpd	(%rax), %xmm2
-# CHECK-NEXT:  3      45    44.00                       vsqrtpd	%ymm0, %ymm2
-# CHECK-NEXT:  4      52    44.00   *                   vsqrtpd	(%rax), %ymm2
-# CHECK-NEXT:  1      14    14.00                       vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   vsqrtps	(%rax), %xmm2
-# CHECK-NEXT:  3      29    28.00                       vsqrtps	%ymm0, %ymm2
-# CHECK-NEXT:  4      36    28.00   *                   vsqrtps	(%rax), %ymm2
-# CHECK-NEXT:  1      21    21.00                       vsqrtsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      27    21.00   *                   vsqrtsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      14    14.00                       vsqrtss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   vsqrtss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  4      5     1.00    *      *      U     vstmxcsr	(%rax)
-# CHECK-NEXT:  1      3     1.00                        vsubpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vsubpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vsubpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vsubpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vsubps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vsubps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vsubps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    1.00    *                   vsubps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vsubsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vsubsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      3     1.00                        vsubss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   vsubss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vtestpd	%xmm0, %xmm1
-# CHECK-NEXT:  2      7     1.00    *                   vtestpd	(%rax), %xmm1
-# CHECK-NEXT:  1      1     1.00                        vtestpd	%ymm0, %ymm1
-# CHECK-NEXT:  2      8     1.00    *                   vtestpd	(%rax), %ymm1
-# CHECK-NEXT:  1      1     1.00                        vtestps	%xmm0, %xmm1
-# CHECK-NEXT:  2      7     1.00    *                   vtestps	(%rax), %xmm1
-# CHECK-NEXT:  1      1     1.00                        vtestps	%ymm0, %ymm1
-# CHECK-NEXT:  2      8     1.00    *                   vtestps	(%rax), %ymm1
-# CHECK-NEXT:  2      2     1.00                        vucomisd	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   vucomisd	(%rax), %xmm1
-# CHECK-NEXT:  2      2     1.00                        vucomiss	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   vucomiss	(%rax), %xmm1
-# CHECK-NEXT:  1      1     1.00                        vunpckhpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vunpckhpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vunpckhpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vunpckhpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vunpckhps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vunpckhps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vunpckhps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vunpckhps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vunpcklpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vunpcklpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vunpcklpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vunpcklpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vunpcklps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vunpcklps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vunpcklps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vunpcklps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vxorpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vxorpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vxorpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vxorpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vxorps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   vxorps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vxorps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vxorps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  20     9     2.00    *      *      U     vzeroall
-# CHECK-NEXT:  1      100   0.33    *      *      U     vzeroupper
+# CHECK-NEXT:  1      10    1.00    *                   vrsqrtss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vshufpd	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vshufpd	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vshufpd	$1, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vshufpd	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vshufps	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vshufps	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vshufps	$1, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vshufps	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      9     13.50                       vsqrtpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    13.50   *                   vsqrtpd	(%rax), %xmm2
+# CHECK-NEXT:  2      9     27.00                       vsqrtpd	%ymm0, %ymm2
+# CHECK-NEXT:  2      14    27.00   *                   vsqrtpd	(%rax), %ymm2
+# CHECK-NEXT:  1      9     10.50                       vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    10.50   *                   vsqrtps	(%rax), %xmm2
+# CHECK-NEXT:  2      9     21.00                       vsqrtps	%ymm0, %ymm2
+# CHECK-NEXT:  2      14    21.00   *                   vsqrtps	(%rax), %ymm2
+# CHECK-NEXT:  1      9     13.50                       vsqrtsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      14    13.50   *                   vsqrtsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      9     10.50                       vsqrtss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      14    10.50   *                   vsqrtss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      1     0.50    *      *      U     vstmxcsr	(%rax)
+# CHECK-NEXT:  1      5     1.00                        vsubpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vsubpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vsubpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vsubpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      5     1.00                        vsubps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vsubps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vsubps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    2.00    *                   vsubps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      5     1.00                        vsubsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vsubsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vsubss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   vsubss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      1     1.00                        vtestpd	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vtestpd	(%rax), %xmm1
+# CHECK-NEXT:  4      1     1.00                        vtestpd	%ymm0, %ymm1
+# CHECK-NEXT:  6      6     1.00    *                   vtestpd	(%rax), %ymm1
+# CHECK-NEXT:  2      1     1.00                        vtestps	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vtestps	(%rax), %xmm1
+# CHECK-NEXT:  4      1     1.00                        vtestps	%ymm0, %ymm1
+# CHECK-NEXT:  6      6     1.00    *                   vtestps	(%rax), %ymm1
+# CHECK-NEXT:  2      1     1.00                        vucomisd	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vucomisd	(%rax), %xmm1
+# CHECK-NEXT:  2      1     1.00                        vucomiss	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   vucomiss	(%rax), %xmm1
+# CHECK-NEXT:  1      2     0.50                        vunpckhpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vunpckhpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vunpckhpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vunpckhpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vunpckhps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vunpckhps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vunpckhps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vunpckhps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vunpcklpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vunpcklpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vunpcklpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vunpcklpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vunpcklps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vunpcklps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vunpcklps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vunpcklps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vxorpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vxorpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vxorpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vxorpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      2     0.50                        vxorps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   vxorps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      2     1.00                        vxorps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      7     1.00    *                   vxorps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  32     90    8.00    *      *      U     vzeroall
+# CHECK-NEXT:  16     46    4.00    *      *      U     vzeroupper
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -     572.00 246.83 317.33 39.00  365.83 179.50 179.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 229.00 229.00  -      -      -     56.00   -      -     588.00 588.00 127.50 127.50 38.00  107.00 402.50 429.50  -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddsubpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddsubpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddsubpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddsubpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddsubps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddsubps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddsubps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vaddsubps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     vaesdec	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   vaesdec	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     vaesdeclast	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   vaesdeclast	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     vaesenc	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   vaesenc	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     vaesenclast	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33   0.50   0.50   vaesenclast	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     2.00    -      -     vaesimc	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     2.00   0.50   0.50   vaesimc	(%rax), %xmm2
-# CHECK-NEXT:  -      -     3.67   3.67    -     3.67    -      -     vaeskeygenassist	$22, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     3.33   3.33    -     3.33   0.50   0.50   vaeskeygenassist	$22, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandnpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandnpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandnps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandnps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vandps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     vblendpd	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   vblendpd	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     vblendpd	$11, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   vblendpd	$11, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     vblendps	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   vblendps	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     vblendps	$11, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   vblendps	$11, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vblendvpd	%xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vblendvpd	%xmm3, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vblendvpd	%ymm3, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vblendvpd	%ymm3, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vblendvps	%xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vblendvps	%xmm3, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vblendvps	%ymm3, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vblendvps	%ymm3, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vbroadcastf128	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vbroadcastsd	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vbroadcastss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vbroadcastss	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcmppd	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcmppd	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcmppd	$0, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcmppd	$0, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcmpps	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcmpps	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcmpps	$0, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcmpps	$0, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcmpsd	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcmpsd	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcmpss	$0, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcmpss	$0, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcomisd	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcomisd	(%rax), %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcomiss	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcomiss	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtdq2pd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtdq2pd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtdq2pd	%xmm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtdq2pd	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtdq2ps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtdq2ps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtdq2ps	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtdq2ps	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtpd2dqx	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtpd2dq	%ymm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtpd2dqy	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtpd2ps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtpd2psx	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtpd2ps	%ymm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtpd2psy	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtps2dq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtps2dq	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtps2dq	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vcvtps2pd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vcvtps2pd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vcvtps2pd	%xmm0, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vcvtps2pd	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvtsd2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvtsd2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvtsd2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvtsd2si	(%rax), %rcx
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtsd2ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtsd2ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtsi2sdl	%ecx, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvtsi2sdq	%rcx, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtsi2sdl	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtsi2sdq	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vcvtsi2ssl	%ecx, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vcvtsi2ssq	%rcx, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtsi2ssl	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvtsi2ssq	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vcvtss2sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vcvtss2sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvtss2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvtss2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvtss2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvtss2si	(%rax), %rcx
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvttpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvttpd2dqx	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vcvttpd2dq	%ymm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vcvttpd2dqy	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvttps2dq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvttps2dq	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvttps2dq	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvttsd2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvttsd2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvttsd2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvttsd2si	(%rax), %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvttss2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vcvttss2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvttss2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vcvttss2si	(%rax), %rcx
-# CHECK-NEXT:  -     22.00  1.00    -      -      -      -      -     vdivpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     22.00  1.00    -      -      -     0.50   0.50   vdivpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -     44.00  2.50    -      -     0.50    -      -     vdivpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -     44.00  2.50    -      -     0.50   0.50   0.50   vdivpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     vdivps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   vdivps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -     28.00  2.50    -      -     0.50    -      -     vdivps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -     28.00  2.50    -      -     0.50   0.50   0.50   vdivps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -     22.00  1.00    -      -      -      -      -     vdivsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     22.00  1.00    -      -      -     0.50   0.50   vdivsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     vdivss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   vdivss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     vdppd	$22, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   0.50   0.50   vdppd	$22, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   2.00    -     1.00    -      -     vdpps	$22, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   2.00    -     1.00   0.50   0.50   vdpps	$22, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   2.00    -     1.00    -      -     vdpps	$22, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00   2.00    -     1.00   0.50   0.50   vdpps	$22, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vextractf128	$1, %ymm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vextractf128	$1, %ymm0, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vextractps	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00   1.00   0.50   0.50   vextractps	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhaddpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhaddpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhaddpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhaddpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhaddps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhaddps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhaddps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhsubpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhsubpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhsubpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhsubpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhsubps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhsubps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     vhsubps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   vhsubps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vinsertf128	$1, %xmm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   vinsertf128	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vinsertps	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vinsertps	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vlddqu	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vlddqu	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   0.50   0.50   vldmxcsr	(%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmaskmovdqu	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vmaskmovpd	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vmaskmovpd	(%rax), %ymm0, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50   1.00    -     0.50   0.50   vmaskmovpd	%xmm0, %xmm1, (%rax)
-# CHECK-NEXT:  -      -     0.50   0.50   1.00    -     0.50   0.50   vmaskmovpd	%ymm0, %ymm1, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vmaskmovps	(%rax), %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vmaskmovps	(%rax), %ymm0, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50   1.00    -     0.50   0.50   vmaskmovps	%xmm0, %xmm1, (%rax)
-# CHECK-NEXT:  -      -     0.50   0.50   1.00    -     0.50   0.50   vmaskmovps	%ymm0, %ymm1, (%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmaxpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vmaxpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmaxpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vmaxpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmaxps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vmaxps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmaxps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vmaxps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmaxsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vmaxsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vmaxss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vmaxss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vminpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vminpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vminpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vminpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vminps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vminps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vminps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vminps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vminsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vminsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vminss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vminss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovapd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovapd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovapd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovapd	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovapd	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovapd	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovaps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovaps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovaps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovaps	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovaps	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovaps	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovd	%eax, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovd	%xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovddup	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovddup	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovddup	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovddup	(%rax), %ymm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vmovdqa	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovdqa	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovdqa	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     vmovdqa	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovdqa	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovdqa	(%rax), %ymm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vmovdqu	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovdqu	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovdqu	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     vmovdqu	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovdqu	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovdqu	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovhlps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovlhps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovhpd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vmovhpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovhps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vmovhps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovlpd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vmovlpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovlps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vmovlps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovmskpd	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovmskpd	%ymm0, %ecx
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovmskps	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovmskps	%ymm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovntdq	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovntdq	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovntdqa	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovntdqa	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovntpd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovntpd	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovntps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovntps	%ymm0, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vmovq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovq	%rax, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmovq	%xmm0, %rcx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovq	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovsd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovshdup	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovshdup	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovshdup	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovshdup	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovsldup	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovsldup	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovsldup	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovsldup	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovss	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovupd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovupd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovupd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovupd	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovupd	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovupd	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovups	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovups	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovups	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vmovups	%ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   vmovups	%ymm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vmovups	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     vmpsadbw	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   0.50   0.50   vmpsadbw	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vmulpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vmulpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vmulps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vmulps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vmulsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vmulss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vorpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vorpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vorpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vorpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vorps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vorps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vorps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vorps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpabsb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpabsb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpabsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpabsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpabsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpabsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpackssdw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpackssdw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpacksswb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpacksswb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpackusdw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpackusdw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpackuswb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpackuswb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddusb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddusb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddusw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddusw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpaddw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpaddw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpalignr	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpalignr	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vpand	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   vpand	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vpandn	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   vpandn	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpavgb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpavgb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpavgw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpavgw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     vpblendvb	%xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   vpblendvb	%xmm3, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpblendw	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpblendw	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     6.00   6.00    -     6.00    -      -     vpclmulqdq	$11, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     5.67   5.67    -     5.67   0.50   0.50   vpclmulqdq	$11, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpeqb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpeqb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpeqd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpeqd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpeqq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpeqq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpeqw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpeqw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     2.67   2.67    -     2.67    -      -     vpcmpestri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     2.33   2.33    -     2.33   0.50   0.50   vpcmpestri	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     2.67   2.67    -     2.67    -      -     vpcmpestrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     2.33   2.33    -     2.33   0.50   0.50   vpcmpestrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpgtb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpgtb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpgtd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpgtd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpcmpgtq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpcmpgtq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmpgtw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmpgtw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -      -      -     vpcmpistri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -     0.50   0.50   vpcmpistri	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -      -      -     vpcmpistrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -     0.50   0.50   vpcmpistrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vperm2f128	$1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vperm2f128	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilpd	$1, %ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilpd	$1, (%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilps	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilps	$1, %ymm0, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilps	$1, (%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermilps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermilps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpextrb	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -     0.50   1.00   0.50   0.50   0.50   vpextrb	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpextrd	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   0.50   1.00   0.50   0.50   0.50   vpextrd	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpextrq	$1, %xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   0.50   1.00   0.50   0.50   0.50   vpextrq	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpextrw	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -     0.50   1.00   0.50   0.50   0.50   vpextrw	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vphminposuw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vphminposuw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphsubd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphsubd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphsubsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphsubsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphsubw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphsubw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     vpinsrb	$1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpinsrb	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     vpinsrd	$1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpinsrd	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     vpinsrq	$1, %rax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpinsrq	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     vpinsrw	$1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpinsrw	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmaddubsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmaddubsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmaddwd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmaddwd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmaxsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmaxsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmaxsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmaxsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmaxsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmaxsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmaxub	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmaxub	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmaxud	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmaxud	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmaxuw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmaxuw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpminsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpminsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpminsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpminsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpminsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpminsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpminub	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpminub	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpminud	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpminud	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpminuw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpminuw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmovmskb	%xmm0, %ecx
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovsxbd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovsxbd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovsxbq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovsxbq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovsxbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovsxbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovsxdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovsxdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovsxwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovsxwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovsxwq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovsxwq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovzxbd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovzxbd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovzxbq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovzxbq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovzxbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovzxbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovzxdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovzxdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovzxwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovzxwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpmovzxwq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpmovzxwq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmuldq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmuldq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmulhrsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmulhrsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmulhuw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmulhuw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmulhw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmulhw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmulld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmulld	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmullw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmullw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmuludq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmuludq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vpor	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   vpor	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsadbw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpsadbw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpshufb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpshufb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpshufd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpshufd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpshufhw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpshufhw	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpshuflw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpshuflw	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsignb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsignb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsignd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsignd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsignw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsignw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpslld	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpslld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpslld	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpslldq	$1, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsllq	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsllq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsllq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsllw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsllw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsllw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsrad	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsrad	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsrad	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsraw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsraw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsraw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsrld	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsrld	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsrld	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsrldq	$1, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsrlq	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsrlq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsrlq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpsrlw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     vpsrlw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   vpsrlw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubsb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubsb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubsw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubsw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubusb	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubusb	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubusw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubusw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpsubw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpsubw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vptest	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vptest	(%rax), %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     vptest	%ymm0, %ymm1
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   vptest	(%rax), %ymm1
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpckhbw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpckhbw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpckhdq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpckhdq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpckhqdq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpckhqdq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpckhwd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpckhwd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpcklbw	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpcklbw	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpckldq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpckldq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpcklqdq	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpcklqdq	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpunpcklwd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpunpcklwd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vpxor	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   vpxor	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vrcpps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vrcpps	(%rax), %xmm2
-# CHECK-NEXT:  -      -     2.50    -      -     0.50    -      -     vrcpps	%ymm0, %ymm2
-# CHECK-NEXT:  -      -     2.50    -      -     0.50   0.50   0.50   vrcpps	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vrcpss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vrcpss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vroundpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vroundpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vroundpd	$1, %ymm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vroundpd	$1, (%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vroundps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vroundps	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vroundps	$1, %ymm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vroundps	$1, (%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vroundsd	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vroundsd	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vroundss	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vroundss	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vrsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vrsqrtps	(%rax), %xmm2
-# CHECK-NEXT:  -      -     2.50    -      -     0.50    -      -     vrsqrtps	%ymm0, %ymm2
-# CHECK-NEXT:  -      -     2.50    -      -     0.50   0.50   0.50   vrsqrtps	(%rax), %ymm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vrsqrtss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vrsqrtss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vshufpd	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vshufpd	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vshufpd	$1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vshufpd	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vshufps	$1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vshufps	$1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vshufps	$1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vshufps	$1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -      -      -     vsqrtpd	%xmm0, %xmm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -     0.50   0.50   vsqrtpd	(%rax), %xmm2
-# CHECK-NEXT:  -     44.00  2.50    -      -     0.50    -      -     vsqrtpd	%ymm0, %ymm2
-# CHECK-NEXT:  -     44.00  2.50    -      -     0.50   0.50   0.50   vsqrtpd	(%rax), %ymm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     vsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   vsqrtps	(%rax), %xmm2
-# CHECK-NEXT:  -     28.00  2.50    -      -     0.50    -      -     vsqrtps	%ymm0, %ymm2
-# CHECK-NEXT:  -     28.00  2.50    -      -     0.50   0.50   0.50   vsqrtps	(%rax), %ymm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -      -      -     vsqrtsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -     0.50   0.50   vsqrtsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     vsqrtss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   vsqrtss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   0.50   0.50   vstmxcsr	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vsubpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vsubpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vsubpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vsubpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vsubps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vsubps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vsubps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vsubps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vsubsd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vsubsd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vsubss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vsubss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vtestpd	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vtestpd	(%rax), %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vtestpd	%ymm0, %ymm1
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vtestpd	(%rax), %ymm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vtestps	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vtestps	(%rax), %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vtestps	%ymm0, %ymm1
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vtestps	(%rax), %ymm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vucomisd	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vucomisd	(%rax), %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     vucomiss	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   vucomiss	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpckhpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpckhpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpckhpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpckhpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpckhps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpckhps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpckhps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpckhps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpcklpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpcklpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpcklpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpcklpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpcklps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpcklps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vunpcklps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vunpcklps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vxorpd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vxorpd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vxorpd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vxorpd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vxorps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vxorps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vxorps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vxorps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     2.00    -      -     vzeroall
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     vzeroupper
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddsubpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddsubpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddsubpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddsubpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddsubps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vaddsubps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddsubps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddsubps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesdec	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesdec	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesdeclast	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesdeclast	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesenc	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesenc	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesenclast	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesenclast	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesimc	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaesimc	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaeskeygenassist	$22, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vaeskeygenassist	$22, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandnpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandnpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandnpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandnpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandnps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandnps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandnps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandnps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vandps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vandps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vblendpd	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vblendpd	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendpd	$11, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendpd	$11, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vblendps	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vblendps	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendps	$11, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendps	$11, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vblendvpd	%xmm3, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vblendvpd	%xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendvpd	%ymm3, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendvpd	%ymm3, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vblendvps	%xmm3, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vblendvps	%xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendvps	%ymm3, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vblendvps	%ymm3, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vbroadcastf128	(%rax), %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vbroadcastsd	(%rax), %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vbroadcastss	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vbroadcastss	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmppd	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmppd	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vcmppd	$0, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vcmppd	$0, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmpps	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmpps	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vcmpps	$0, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vcmpps	$0, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmpsd	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmpsd	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmpss	$0, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcmpss	$0, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcomisd	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcomisd	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcomiss	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vcomiss	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtdq2pd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtdq2pd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtdq2pd	%xmm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtdq2pd	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtdq2ps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtdq2ps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtdq2ps	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtdq2ps	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtpd2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtpd2dqx	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvtpd2dq	%ymm0, %xmm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvtpd2dqy	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtpd2ps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtpd2psx	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvtpd2ps	%ymm0, %xmm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvtpd2psy	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtps2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtps2dq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtps2dq	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtps2dq	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtps2pd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtps2pd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtps2pd	%xmm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtps2pd	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtsd2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtsd2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtsd2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtsd2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsd2ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsd2ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2sdl	%ecx, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2sdq	%rcx, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2sdl	(%rax), %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2sdq	(%rax), %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2ssl	%ecx, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2ssq	%rcx, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2ssl	(%rax), %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtsi2ssq	(%rax), %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtss2sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtss2sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtss2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtss2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtss2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvtss2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvttpd2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvttpd2dqx	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvttpd2dq	%ymm0, %xmm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvttpd2dqy	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvttps2dq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvttps2dq	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvttps2dq	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttsd2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttsd2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttsd2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttsd2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttss2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttss2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttss2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     vcvttss2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     19.00  19.00   -      -      -      -      -     2.00    -      -      -      -     vdivpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     19.00  19.00   -      -      -      -      -     2.00    -      -      -      -     vdivpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     19.00  19.00   -      -      -      -      -     2.00    -      -      -      -     vdivps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     19.00  19.00   -      -      -      -      -     2.00    -      -      -      -     vdivps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     vdivss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     vdppd	$22, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     vdppd	$22, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     vdpps	$22, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     vdpps	$22, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -      -     2.00    -      -      -      -     vdpps	$22, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -      -     2.00    -      -      -      -     vdpps	$22, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vextractf128	$1, %ymm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vextractf128	$1, %ymm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vextractps	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vextractps	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhaddpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhaddpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhaddpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhaddpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhaddps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhaddps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhaddps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhsubpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhsubpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhsubpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhsubpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhsubps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vhsubps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhsubps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vhsubps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vinsertf128	$1, %xmm0, %ymm1, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vinsertf128	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vinsertps	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vinsertps	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vlddqu	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vlddqu	(%rax), %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vldmxcsr	(%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmaskmovdqu	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     1.00   1.00    -      -      -      -     0.50   0.50    -      -      -      -     vmaskmovpd	(%rax), %xmm0, %xmm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vmaskmovpd	(%rax), %ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vmaskmovpd	%xmm0, %xmm1, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vmaskmovpd	%ymm0, %ymm1, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     1.00   1.00    -      -      -      -     0.50   0.50    -      -      -      -     vmaskmovps	(%rax), %xmm0, %xmm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vmaskmovps	(%rax), %ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vmaskmovps	%xmm0, %xmm1, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vmaskmovps	%ymm0, %ymm1, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vmaxpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vmaxpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vmaxps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vmaxps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmaxss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vminpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vminpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vminps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vminps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vminss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovapd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovapd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovapd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovapd	%ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovapd	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovapd	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovaps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovaps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovaps	%ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovaps	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovaps	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovd	%eax, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmovd	%xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovd	%xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovddup	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovddup	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovddup	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovddup	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovdqa	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovdqa	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovdqa	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00   1.00    -      -     1.00   1.00    -      -      -      -     vmovdqa	%ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovdqa	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovdqa	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovdqu	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovdqu	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovdqu	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00   1.00    -      -     1.00   1.00    -      -      -      -     vmovdqu	%ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovdqu	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovdqu	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovhlps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovlhps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovhpd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovhpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovhps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovhps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovlpd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovlpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovlps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovlps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmovmskpd	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmovmskpd	%ymm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmovmskps	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmovmskps	%ymm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovntdq	%xmm0, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     2.00    -     2.00    -      -      -      -     vmovntdq	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovntdqa	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovntdqa	(%rax), %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovntpd	%xmm0, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     2.00    -     2.00    -      -      -      -     vmovntpd	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovntps	%xmm0, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     2.00    -     2.00    -      -      -      -     vmovntps	%ymm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovq	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovq	%rax, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vmovq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vmovq	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovq	%xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovsd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovshdup	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovshdup	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovshdup	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovshdup	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovsldup	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovsldup	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovsldup	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovsldup	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovss	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovupd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovupd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovupd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovupd	%ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovupd	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovupd	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovups	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovups	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovups	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vmovups	%ymm0, %ymm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vmovups	%ymm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vmovups	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     2.00    -     1.00    -      -      -      -      -     vmpsadbw	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     2.00    -     1.00    -      -      -      -      -     vmpsadbw	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vmulpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vmulpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vmulps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vmulss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vorpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vorpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vorpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vorpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vorps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vorps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vorps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vorps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpabsb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpabsb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpabsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpabsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpabsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpabsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpackssdw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpackssdw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpacksswb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpacksswb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpackusdw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpackusdw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpackuswb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpackuswb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddusb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddusb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddusw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddusw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpaddw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpalignr	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpalignr	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpand	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpand	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpandn	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpandn	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpavgb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpavgb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpavgw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpavgw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpblendvb	%xmm3, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpblendvb	%xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpblendw	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpblendw	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpclmulqdq	$11, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpclmulqdq	$11, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpeqw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT: 4.00   4.00    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     vpcmpestri	$1, %xmm0, %xmm2
+# CHECK-NEXT: 4.50   4.50    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     vpcmpestri	$1, (%rax), %xmm2
+# CHECK-NEXT: 4.00   4.00    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     vpcmpestrm	$1, %xmm0, %xmm2
+# CHECK-NEXT: 4.50   4.50    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     vpcmpestrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmpgtw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     vpcmpistri	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     vpcmpistri	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     vpcmpistrm	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     vpcmpistrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vperm2f128	$1, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vperm2f128	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vpermilpd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vpermilpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermilpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermilpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilpd	$1, %ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilpd	$1, (%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vpermilps	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vpermilps	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermilps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermilps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilps	$1, %ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilps	$1, (%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermilps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vpextrb	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vpextrb	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vpextrd	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vpextrd	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vpextrq	$1, %xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vpextrq	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vpextrw	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vpextrw	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     1.00    -      -      -      -      -     vphminposuw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     1.00    -      -      -      -      -     vphminposuw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrb	$1, %eax, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrb	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrd	$1, %eax, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrd	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrq	$1, %rax, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrq	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrw	$1, %eax, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpinsrw	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmaddubsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmaddubsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmaddwd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmaddwd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxub	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxub	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxud	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxud	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxuw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmaxuw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminub	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminub	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminud	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminud	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminuw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpminuw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vpmovmskb	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxbd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxbd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxbq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxbq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxwq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovsxwq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxbd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxbd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxbq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxbq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxwq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpmovzxwq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmuldq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmuldq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmulhrsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmulhrsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmulhuw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmulhuw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmulhw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmulhw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmulld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmulld	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmullw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmullw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmuludq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmuludq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpor	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpor	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsadbw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsadbw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpshufb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpshufb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshufd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshufd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshufhw	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshufhw	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshuflw	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshuflw	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsignb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsignb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsignd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsignd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsignw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsignw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpslld	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpslld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpslld	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpslldq	$1, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsllq	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsllq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsllq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsllw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsllw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsllw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrad	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrad	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrad	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsraw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsraw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsraw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrld	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrld	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrld	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrldq	$1, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrlq	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrlq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrlq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrlw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrlw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsrlw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubsb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubsb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubsw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubsw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubusb	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubusb	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubusw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubusw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpsubw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vptest	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vptest	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vptest	%ymm0, %ymm1
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vptest	(%rax), %ymm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhbw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhbw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhdq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhdq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhqdq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhqdq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhwd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckhwd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpcklbw	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpcklbw	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckldq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpckldq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpcklqdq	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpcklqdq	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpcklwd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpunpcklwd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpxor	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpxor	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrcpps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrcpps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vrcpps	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vrcpps	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrcpss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrcpss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundpd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vroundpd	$1, %ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vroundpd	$1, (%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundps	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundps	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vroundps	$1, %ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vroundps	$1, (%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundsd	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundsd	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundss	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vroundss	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrsqrtps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrsqrtps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vrsqrtps	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vrsqrtps	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrsqrtss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     vrsqrtss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vshufpd	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vshufpd	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vshufpd	$1, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vshufpd	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vshufps	$1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vshufps	$1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vshufps	$1, %ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vshufps	$1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     27.00  27.00   -      -      -      -      -     2.00    -      -      -      -     vsqrtpd	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     27.00  27.00   -      -      -      -      -     2.00    -      -      -      -     vsqrtpd	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     21.00  21.00   -      -      -      -      -     2.00    -      -      -      -     vsqrtps	%ymm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     21.00  21.00   -      -      -      -      -     2.00    -      -      -      -     vsqrtps	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     vsqrtss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vstmxcsr	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vsubpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vsubpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vsubps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vsubps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubsd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubsd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vsubss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vtestpd	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vtestpd	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vtestpd	%ymm0, %ymm1
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vtestpd	(%rax), %ymm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vtestps	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vtestps	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vtestps	%ymm0, %ymm1
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vtestps	(%rax), %ymm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vucomisd	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vucomisd	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vucomiss	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     vucomiss	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpckhpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpckhpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpckhpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpckhpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpckhps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpckhps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpckhps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpckhps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpcklpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpcklpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpcklpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpcklpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpcklps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vunpcklps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpcklps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vunpcklps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vxorpd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vxorpd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vxorpd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vxorpd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vxorps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     vxorps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vxorps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00   1.00    -      -      -      -     vxorps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vzeroall
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vzeroupper
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-bmi1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-bmi1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-bmi1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 andn        %eax, %ebx, %ecx
 andn        (%rax), %ebx, %ecx
@@ -46,68 +46,80 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        andnl	%eax, %ebx, %ecx
-# CHECK-NEXT:  2      6     0.50    *                   andnl	(%rax), %ebx, %ecx
-# CHECK-NEXT:  1      1     0.33                        andnq	%rax, %rbx, %rcx
-# CHECK-NEXT:  2      6     0.50    *                   andnq	(%rax), %rbx, %rcx
-# CHECK-NEXT:  2      2     1.00                        bextrl	%eax, %ebx, %ecx
-# CHECK-NEXT:  3      7     1.00    *                   bextrl	%eax, (%rbx), %ecx
-# CHECK-NEXT:  2      2     1.00                        bextrq	%rax, %rbx, %rcx
-# CHECK-NEXT:  3      7     1.00    *                   bextrq	%rax, (%rbx), %rcx
-# CHECK-NEXT:  1      1     0.33                        blsil	%eax, %ecx
+# CHECK-NEXT:  1      1     0.50                        andnl	%eax, %ebx, %ecx
+# CHECK-NEXT:  1      5     0.50    *                   andnl	(%rax), %ebx, %ecx
+# CHECK-NEXT:  1      1     0.50                        andnq	%rax, %rbx, %rcx
+# CHECK-NEXT:  1      5     0.50    *                   andnq	(%rax), %rbx, %rcx
+# CHECK-NEXT:  2      2     0.50                        bextrl	%eax, %ebx, %ecx
+# CHECK-NEXT:  2      6     0.50    *                   bextrl	%eax, (%rbx), %ecx
+# CHECK-NEXT:  2      2     0.50                        bextrq	%rax, %rbx, %rcx
+# CHECK-NEXT:  2      6     0.50    *                   bextrq	%rax, (%rbx), %rcx
+# CHECK-NEXT:  2      2     0.50                        blsil	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blsil	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blsiq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blsiq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blsiq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blsmskl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blsmskl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blsmskl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blsmskq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blsmskq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blsmskq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blsrl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blsrl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blsrl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blsrq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blsrq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blsrq	(%rax), %rcx
-# CHECK-NEXT:  1      3     1.00                        tzcntl	%eax, %ecx
-# CHECK-NEXT:  2      8     1.00    *                   tzcntl	(%rax), %ecx
-# CHECK-NEXT:  1      3     1.00                        tzcntq	%rax, %rcx
-# CHECK-NEXT:  2      8     1.00    *                   tzcntq	(%rax), %rcx
+# CHECK-NEXT:  2      2     1.00                        tzcntl	%eax, %ecx
+# CHECK-NEXT:  2      6     1.00    *                   tzcntl	(%rax), %ecx
+# CHECK-NEXT:  2      2     1.00                        tzcntq	%rax, %rcx
+# CHECK-NEXT:  2      6     1.00    *                   tzcntq	(%rax), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     7.33   13.33   -     7.33   6.00   6.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 6.00   6.00    -      -      -     14.00  14.00   -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andnl	%eax, %ebx, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   andnl	(%rax), %ebx, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andnq	%rax, %rbx, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   andnq	(%rax), %rbx, %rcx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50    -      -     bextrl	%eax, %ebx, %ecx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50   0.50   0.50   bextrl	%eax, (%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50    -      -     bextrq	%rax, %rbx, %rcx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50   0.50   0.50   bextrq	%rax, (%rbx), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsil	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsil	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsiq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsiq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsmskl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsmskl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsmskq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsmskq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsrl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsrl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsrq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsrq	(%rax), %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     tzcntl	%eax, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   tzcntl	(%rax), %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     tzcntq	%rax, %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   tzcntq	(%rax), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andnl	%eax, %ebx, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andnl	(%rax), %ebx, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andnq	%rax, %rbx, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andnq	(%rax), %rbx, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrl	%eax, %ebx, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrl	%eax, (%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrq	%rax, %rbx, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrq	%rax, (%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsil	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsil	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsiq	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsiq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsmskl	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsmskl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsmskq	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsmskq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsrl	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsrl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsrq	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsrq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     tzcntl	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     tzcntl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     tzcntq	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     tzcntq	(%rax), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-clflushopt.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-clflushopt.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-clflushopt.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 clflushopt (%rax)
 
@@ -12,22 +12,34 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  4      5     1.00    *      *      U     clflushopt	(%rax)
+# CHECK-NEXT:  1      5     0.50    *      *      U     clflushopt	(%rax)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     0.50   0.50   1.00   1.00   0.50   0.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50   0.50   1.00   1.00   0.50   0.50   clflushopt	(%rax)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     clflushopt	(%rax)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-cmov.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-cmov.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-cmov.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 cmovow    %si, %di
 cmovnow   %si, %di
@@ -112,212 +112,224 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      2     0.67                        cmovow	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovnow	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovbw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovaew	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovew	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovnew	%si, %di
-# CHECK-NEXT:  3      3     1.00                        cmovbew	%si, %di
-# CHECK-NEXT:  3      3     1.00                        cmovaw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovsw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovnsw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovpw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovnpw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovlw	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovgew	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovlew	%si, %di
-# CHECK-NEXT:  2      2     0.67                        cmovgw	%si, %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovow	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovnow	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovbw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovaew	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovew	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovnew	(%rax), %di
-# CHECK-NEXT:  4      8     1.00    *                   cmovbew	(%rax), %di
-# CHECK-NEXT:  4      8     1.00    *                   cmovaw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovsw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovnsw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovpw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovnpw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovlw	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovgew	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovlew	(%rax), %di
-# CHECK-NEXT:  3      7     0.67    *                   cmovgw	(%rax), %di
-# CHECK-NEXT:  2      2     0.67                        cmovol	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovnol	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovbl	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovael	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovel	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovnel	%esi, %edi
-# CHECK-NEXT:  3      3     1.00                        cmovbel	%esi, %edi
-# CHECK-NEXT:  3      3     1.00                        cmoval	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovsl	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovnsl	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovpl	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovnpl	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovll	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovgel	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovlel	%esi, %edi
-# CHECK-NEXT:  2      2     0.67                        cmovgl	%esi, %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovol	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnol	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovbl	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovael	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovel	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnel	(%rax), %edi
-# CHECK-NEXT:  4      8     1.00    *                   cmovbel	(%rax), %edi
-# CHECK-NEXT:  4      8     1.00    *                   cmoval	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovsl	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnsl	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovpl	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnpl	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovll	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovgel	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovlel	(%rax), %edi
-# CHECK-NEXT:  3      7     0.67    *                   cmovgl	(%rax), %edi
-# CHECK-NEXT:  2      2     0.67                        cmovoq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovnoq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovbq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovaeq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmoveq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovneq	%rsi, %rdi
-# CHECK-NEXT:  3      3     1.00                        cmovbeq	%rsi, %rdi
-# CHECK-NEXT:  3      3     1.00                        cmovaq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovsq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovnsq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovpq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovnpq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovlq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovgeq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovleq	%rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        cmovgq	%rsi, %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovoq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnoq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovbq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovaeq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmoveq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovneq	(%rax), %rdi
-# CHECK-NEXT:  4      8     1.00    *                   cmovbeq	(%rax), %rdi
-# CHECK-NEXT:  4      8     1.00    *                   cmovaq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovsq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnsq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovpq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovnpq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovlq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovgeq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovleq	(%rax), %rdi
-# CHECK-NEXT:  3      7     0.67    *                   cmovgq	(%rax), %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovow	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovnow	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovbw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovaew	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovew	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovnew	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovbew	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovaw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovsw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovnsw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovpw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovnpw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovlw	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovgew	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovlew	%si, %di
+# CHECK-NEXT:  1      1     0.50                        cmovgw	%si, %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovow	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovnow	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovbw	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovaew	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovew	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovnew	(%rax), %di
+# CHECK-NEXT:  2      5     0.50    *                   cmovbew	(%rax), %di
+# CHECK-NEXT:  2      5     0.50    *                   cmovaw	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovsw	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovnsw	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovpw	(%rax), %di
+# CHECK-NEXT:  1      5     0.50    *                   cmovnpw	(%rax), %di
+# CHECK-NEXT:  2      5     0.50    *                   cmovlw	(%rax), %di
+# CHECK-NEXT:  2      5     0.50    *                   cmovgew	(%rax), %di
+# CHECK-NEXT:  2      5     0.50    *                   cmovlew	(%rax), %di
+# CHECK-NEXT:  2      5     0.50    *                   cmovgw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        cmovol	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovnol	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovbl	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovael	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovel	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovnel	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovbel	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmoval	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovsl	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovnsl	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovpl	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovnpl	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovll	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovgel	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovlel	%esi, %edi
+# CHECK-NEXT:  1      1     0.50                        cmovgl	%esi, %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovol	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnol	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovbl	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovael	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovel	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnel	(%rax), %edi
+# CHECK-NEXT:  2      5     0.50    *                   cmovbel	(%rax), %edi
+# CHECK-NEXT:  2      5     0.50    *                   cmoval	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovsl	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnsl	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovpl	(%rax), %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnpl	(%rax), %edi
+# CHECK-NEXT:  2      5     0.50    *                   cmovll	(%rax), %edi
+# CHECK-NEXT:  2      5     0.50    *                   cmovgel	(%rax), %edi
+# CHECK-NEXT:  2      5     0.50    *                   cmovlel	(%rax), %edi
+# CHECK-NEXT:  2      5     0.50    *                   cmovgl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        cmovoq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovnoq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovbq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovaeq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmoveq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovneq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovbeq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovaq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovsq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovnsq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovpq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovnpq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovlq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovgeq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovleq	%rsi, %rdi
+# CHECK-NEXT:  1      1     0.50                        cmovgq	%rsi, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovoq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnoq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovbq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovaeq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmoveq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovneq	(%rax), %rdi
+# CHECK-NEXT:  2      5     0.50    *                   cmovbeq	(%rax), %rdi
+# CHECK-NEXT:  2      5     0.50    *                   cmovaq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovsq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnsq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovpq	(%rax), %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmovnpq	(%rax), %rdi
+# CHECK-NEXT:  2      5     0.50    *                   cmovlq	(%rax), %rdi
+# CHECK-NEXT:  2      5     0.50    *                   cmovgeq	(%rax), %rdi
+# CHECK-NEXT:  2      5     0.50    *                   cmovleq	(%rax), %rdi
+# CHECK-NEXT:  2      5     0.50    *                   cmovgq	(%rax), %rdi
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     86.00  32.00   -     86.00  24.00  24.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 24.00  24.00   -      -      -     48.00  48.00   -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovow	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnow	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovbw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovaew	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovew	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnew	%si, %di
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33    -      -     cmovbew	%si, %di
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33    -      -     cmovaw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovsw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnsw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovpw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnpw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovlw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovgew	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovlew	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovgw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovow	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnow	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovbw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovaew	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovew	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnew	(%rax), %di
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33   0.50   0.50   cmovbew	(%rax), %di
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33   0.50   0.50   cmovaw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovsw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnsw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovpw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnpw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovlw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovgew	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovlew	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovgw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovol	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnol	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovbl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovael	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovel	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnel	%esi, %edi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33    -      -     cmovbel	%esi, %edi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33    -      -     cmoval	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovsl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnsl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovpl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnpl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovll	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovgel	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovlel	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovgl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovol	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnol	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovbl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovael	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovel	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnel	(%rax), %edi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33   0.50   0.50   cmovbel	(%rax), %edi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33   0.50   0.50   cmoval	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovsl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnsl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovpl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnpl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovll	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovgel	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovlel	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovgl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovoq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnoq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovbq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovaeq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmoveq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovneq	%rsi, %rdi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33    -      -     cmovbeq	%rsi, %rdi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33    -      -     cmovaq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovsq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnsq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovpq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovnpq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovlq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovgeq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovleq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     cmovgq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovoq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnoq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovbq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovaeq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmoveq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovneq	(%rax), %rdi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33   0.50   0.50   cmovbeq	(%rax), %rdi
-# CHECK-NEXT:  -      -     1.33   0.33    -     1.33   0.50   0.50   cmovaq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovsq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnsq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovpq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovnpq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovlq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovgeq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovleq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   cmovgq	(%rax), %rdi
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovow	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnow	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaew	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovew	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnew	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbew	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovsw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnsw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovpw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnpw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgew	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlew	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgw	%si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovow	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnow	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaew	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovew	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnew	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbew	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovsw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnsw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovpw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnpw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgew	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlew	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovol	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnol	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovael	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovel	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnel	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbel	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmoval	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovsl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnsl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovpl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnpl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovll	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgel	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlel	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgl	%esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovol	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnol	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovael	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovel	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnel	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbel	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmoval	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovsl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnsl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovpl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnpl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovll	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgel	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlel	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovoq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnoq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaeq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmoveq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovneq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbeq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovsq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnsq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovpq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnpq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgeq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovleq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgq	%rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovoq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnoq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaeq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmoveq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovneq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovbeq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovaq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovsq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnsq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovpq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovnpq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovlq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgeq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovleq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmovgq	(%rax), %rdi
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-cmpxchg.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-cmpxchg.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-cmpxchg.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 cmpxchg8b  (%rax)
 cmpxchg16b (%rax)
@@ -13,24 +13,36 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  3      6     1.00    *      *            cmpxchg8b	(%rax)
-# CHECK-NEXT:  3      6     1.00    *      *            cmpxchg16b	(%rax)
+# CHECK-NEXT:  18     3     1.00    *      *            cmpxchg8b	(%rax)
+# CHECK-NEXT:  22     3     1.00    *      *            cmpxchg16b	(%rax)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     0.67   0.67   2.00   0.67   2.00   2.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -     2.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   cmpxchg8b	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   cmpxchg16b	(%rax)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchg8b	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchg16b	(%rax)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-f16c.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-f16c.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-f16c.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 vcvtph2ps   %xmm0, %xmm2
 vcvtph2ps   (%rax), %xmm2
@@ -22,36 +22,48 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vcvtph2ps	%xmm0, %xmm2
-# CHECK-NEXT:  2      8     1.00    *                   vcvtph2ps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        vcvtph2ps	%xmm0, %ymm2
-# CHECK-NEXT:  2      8     1.00    *                   vcvtph2ps	(%rax), %ymm2
-# CHECK-NEXT:  1      3     1.00                        vcvtps2ph	$0, %xmm0, %xmm2
-# CHECK-NEXT:  1      4     1.00           *            vcvtps2ph	$0, %xmm0, (%rax)
-# CHECK-NEXT:  1      3     1.00                        vcvtps2ph	$0, %ymm0, %xmm2
-# CHECK-NEXT:  1      4     1.00           *            vcvtps2ph	$0, %ymm0, (%rax)
+# CHECK-NEXT:  2      8     1.00                        vcvtph2ps	%xmm0, %xmm2
+# CHECK-NEXT:  3      13    1.00    *                   vcvtph2ps	(%rax), %xmm2
+# CHECK-NEXT:  4      8     2.00                        vcvtph2ps	%xmm0, %ymm2
+# CHECK-NEXT:  7      13    2.00    *                   vcvtph2ps	(%rax), %ymm2
+# CHECK-NEXT:  2      8     1.00                        vcvtps2ph	$0, %xmm0, %xmm2
+# CHECK-NEXT:  3      4     1.00           *            vcvtps2ph	$0, %xmm0, (%rax)
+# CHECK-NEXT:  4      8     2.00                        vcvtps2ph	$0, %ymm0, %xmm2
+# CHECK-NEXT:  4      4     2.00           *            vcvtps2ph	$0, %ymm0, (%rax)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     8.00   2.00    -     2.00   2.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 2.50   2.50    -      -      -      -      -      -     1.00   1.00    -      -      -     8.00    -     12.00   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtph2ps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtph2ps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtph2ps	%xmm0, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vcvtph2ps	(%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtps2ph	$0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     0.50   0.50   vcvtps2ph	$0, %xmm0, (%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vcvtps2ph	$0, %ymm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     0.50   0.50   vcvtps2ph	$0, %ymm0, (%rax)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtph2ps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtph2ps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtph2ps	%xmm0, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vcvtph2ps	(%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtps2ph	$0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vcvtps2ph	$0, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvtps2ph	$0, %ymm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -     1.00    -     2.00    -      -      -      -     vcvtps2ph	$0, %ymm0, (%rax)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-fma.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-fma.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-fma.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 vfmadd132pd %xmm0, %xmm1, %xmm2
 vfmadd132pd (%rax), %xmm1, %xmm2
@@ -299,403 +299,415 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      5     0.50                        vfmadd132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmadd132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmadd132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmadd132pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmadd213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmadd213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmadd213pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmadd231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmadd231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmadd231pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmadd132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmadd132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmadd132ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmadd213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmadd213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmadd213ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmadd231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmadd231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmadd231ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd132sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd213sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd231sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd132ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd213ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmadd231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmadd231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmadd231ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmaddsub132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmaddsub132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsub132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmaddsub132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsub132pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmaddsub213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmaddsub213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsub213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmaddsub213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsub213pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmaddsub231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmaddsub231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsub231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmaddsub231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsub231pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmaddsub132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmaddsub132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsub132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmaddsub132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsub132ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmaddsub213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmaddsub213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsub213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmaddsub213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsub213ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmaddsub231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmaddsub231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsub231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsub231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmaddsub231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsub231ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsub132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsub132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsub132pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsub213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsub213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsub213pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsub231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsub231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsub231pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsub132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsub132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsub132ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsub213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsub213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsub213ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsub231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsub231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsub231ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub132sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub213sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub231sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub132ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub213ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmsub231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsub231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsub231ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfmsubadd132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsubadd132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubadd132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsubadd132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubadd132pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsubadd213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsubadd213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubadd213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsubadd213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubadd213pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsubadd231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsubadd231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubadd231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsubadd231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubadd231pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsubadd132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsubadd132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubadd132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsubadd132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubadd132ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsubadd213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsubadd213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubadd213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsubadd213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubadd213ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfmsubadd231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfmsubadd231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubadd231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubadd231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfmsubadd231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubadd231ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmadd132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmadd132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmadd132pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmadd213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmadd213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmadd213pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmadd231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmadd231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmadd231pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmadd132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmadd132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmadd132ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmadd213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmadd213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmadd213ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmadd231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmadd231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmadd231ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd132sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd213sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd231sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd132ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd213ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmadd231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmadd231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmadd231ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmsub132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmsub132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsub132pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmsub213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmsub213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsub213pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmsub231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmsub231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsub231pd	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmsub132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmsub132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsub132ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmsub213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmsub213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsub213ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  1      5     0.50                        vfnmsub231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  2      5     0.50                        vfnmsub231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsub231ps	(%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub132sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub213sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub231sd	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub132ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub213ss	(%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      5     0.50                        vfnmsub231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsub231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsub231ss	(%rax), %xmm1, %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     96.00  96.00   -      -     48.00  48.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 66.00  66.00   -      -      -      -      -      -     96.00  96.00   -      -      -      -     48.00  48.00  48.00  48.00   -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd132pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd213pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd231pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd132ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd213ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd231ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd132sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd213sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd231sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd132ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd213ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmadd231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmadd231ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub132pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub213pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub231pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub132ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub213ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsub231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsub231ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub132pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub213pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub231pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub132ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub213ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub231ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub132sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub213sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub231sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub132ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub213ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsub231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsub231ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd132pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd213pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd231pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd132ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd213ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubadd231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubadd231ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd132pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd213pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd231pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd132ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd213ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd231ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd132sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd213sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd231sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd132ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd213ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmadd231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmadd231ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub132pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub132pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub132pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub132pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub213pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub213pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub213pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub213pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub231pd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub231pd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub231pd	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub231pd	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub132ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub132ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub132ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub132ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub213ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub213ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub213ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub213ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub231ps	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub231ps	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub231ps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub231ps	(%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub132sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub132sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub213sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub213sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub231sd	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub231sd	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub132ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub132ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub213ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub213ss	(%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsub231ss	%xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsub231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmadd231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsub231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsub231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubadd231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmadd231ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231pd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231pd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231pd	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231pd	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231ps	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231ps	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231ps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231ps	(%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231sd	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231sd	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub132ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub213ss	(%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231ss	%xmm0, %xmm1, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsub231ss	(%rax), %xmm1, %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-fma4.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-fma4.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-fma4.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 vfmaddpd    %xmm0, %xmm1, %xmm2, %xmm3
 vfmaddpd    (%rax), %xmm1, %xmm2, %xmm3
@@ -139,211 +139,223 @@
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      5     0.50                        vfmaddpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddpd	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmaddps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmaddps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmaddps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddps	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmaddsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsd	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfmaddss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddss	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfmaddsubpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmaddsubpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsubpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsubpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmaddsubpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsubpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsubpd	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmaddsubps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmaddsubps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmaddsubps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsubps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmaddsubps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmaddsubps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsubps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmaddsubps	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmsubaddpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmsubaddpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubaddpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubaddpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmsubaddpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubaddpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubaddpd	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmsubaddps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmsubaddps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubaddps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubaddps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubaddps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmsubaddps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubaddps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubaddps	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmsubpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubpd	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmsubps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfmsubps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfmsubps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfmsubps	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfmsubsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubsd	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfmsubss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfmsubss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfmsubss	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfnmaddpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfnmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfnmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmaddpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmaddpd	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfnmaddps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfnmaddps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfnmaddps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmaddps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmaddps	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfnmaddsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddsd	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfnmaddss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmaddss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmaddss	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfnmsubpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfnmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfnmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsubpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsubpd	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfnmsubps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  1      5     0.50                        vfnmsubps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  2      5     0.50                        vfnmsubps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsubps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      10    1.00    *                   vfnmsubps	%ymm0, (%rax), %ymm2, %ymm3
 # CHECK-NEXT:  1      5     0.50                        vfnmsubsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubsd	%xmm0, (%rax), %xmm2, %xmm3
 # CHECK-NEXT:  1      5     0.50                        vfnmsubss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      10    0.50    *                   vfnmsubss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    0.50    *                   vfnmsubss	%xmm0, (%rax), %xmm2, %xmm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     48.00  48.00   -      -     32.00  32.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 44.00  44.00   -      -      -      -      -      -     48.00  48.00   -      -      -      -     24.00  24.00  24.00  24.00   -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddpd	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddps	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddss	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsubpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsubpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubpd	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsubps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmaddsubps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmaddsubps	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubaddpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubaddpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddpd	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubaddps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubaddps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubaddps	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubpd	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubps	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubsd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfmsubss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfmsubss	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmaddpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddpd	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmaddps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmaddps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddps	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmaddsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddsd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmaddss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmaddss	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsubpd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubpd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubpd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubpd	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubpd	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsubps	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubps	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubps	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsubps	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubps	(%rax), %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubps	%ymm0, (%rax), %ymm2, %ymm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsubsd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubsd	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubsd	%xmm0, (%rax), %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     vfnmsubss	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubss	(%rax), %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50   vfnmsubss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddpd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddps	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddss	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubpd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubps	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmaddsubps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddpd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddps	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubaddps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubpd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubps	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubsd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubss	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfmsubss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddpd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddps	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddsd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddss	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmaddss	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubpd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubpd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubpd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubpd	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubpd	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubpd	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubps	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubps	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubps	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubps	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubps	(%rax), %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubps	%ymm0, (%rax), %ymm2, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubsd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubsd	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubsd	%xmm0, (%rax), %xmm2, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubss	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubss	(%rax), %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.25   0.25   0.25   0.25    -      -     vfnmsubss	%xmm0, (%rax), %xmm2, %xmm3
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-lea.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-lea.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-lea.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 lea 0(), %cx
 lea 0(), %ecx
@@ -148,290 +148,302 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.50                        leaw	0, %cx
-# CHECK-NEXT:  1      1     0.50                        leal	0, %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	0, %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%eax), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%eax), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%eax), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%rax), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(,%ebx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(,%ebx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(,%ebx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(,%rbx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(,%rbx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(,%rbx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%eax,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%eax,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%eax,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%rax,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%rax,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%rax,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%eax,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%eax,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%eax,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%rax,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%rax,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%rax,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%eax,%ebx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%eax,%ebx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%eax,%ebx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	(%rax,%rbx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	(%rax,%rbx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	(%rax,%rbx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16, %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16, %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16, %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%eax), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%eax), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%eax), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%rax), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%rax), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%rax), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(,%ebx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(,%ebx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(,%ebx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(,%rbx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(,%rbx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(,%rbx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%eax,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%eax,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%eax,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%rax,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%rax,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%rax,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%eax,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%eax,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%eax,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%rax,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%rax,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%rax,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%eax,%ebx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%eax,%ebx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%eax,%ebx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	-16(%rax,%rbx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	-16(%rax,%rbx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	-16(%rax,%rbx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024, %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024, %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024, %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%eax), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%eax), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%eax), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%rax), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%rax), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%rax), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(,%ebx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(,%ebx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(,%ebx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(,%rbx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(,%rbx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(,%rbx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%eax,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%eax,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%eax,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%rax,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%rax,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%rax,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%eax,%ebx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%eax,%ebx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%eax,%ebx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%rax,%rbx), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%rax,%rbx), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%rax,%rbx), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%eax,%ebx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%eax,%ebx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%eax,%ebx,2), %rcx
-# CHECK-NEXT:  1      1     0.50                        leaw	1024(%rax,%rbx,2), %cx
-# CHECK-NEXT:  1      1     0.50                        leal	1024(%rax,%rbx,2), %ecx
-# CHECK-NEXT:  1      1     0.50                        leaq	1024(%rax,%rbx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	0, %cx
+# CHECK-NEXT:  2      1     0.50                        leal	0, %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	0, %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%eax), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%eax), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%eax), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%rax), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%rax), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%rax), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(,%ebx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(,%ebx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(,%ebx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(,%rbx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(,%rbx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(,%rbx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%eax,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%eax,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%eax,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%rax,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%rax,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%rax,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%eax,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%eax,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%eax,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%rax,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%rax,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%rax,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%eax,%ebx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%eax,%ebx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%eax,%ebx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	(%rax,%rbx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	(%rax,%rbx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	(%rax,%rbx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16, %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16, %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16, %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%eax), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%eax), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%eax), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%rax), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%rax), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%rax), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(,%ebx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(,%ebx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(,%ebx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(,%rbx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(,%rbx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(,%rbx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%eax,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%eax,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%eax,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%rax,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%rax,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%rax,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%eax,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%eax,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%eax,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%rax,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%rax,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%rax,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%eax,%ebx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%eax,%ebx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%eax,%ebx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	-16(%rax,%rbx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	-16(%rax,%rbx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	-16(%rax,%rbx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024, %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024, %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024, %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%eax), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%eax), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%eax), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%rax), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%rax), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%rax), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(,%ebx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(,%ebx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(,%ebx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(,%rbx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(,%rbx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(,%rbx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%eax,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%eax,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%eax,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%rax,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%rax,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%rax,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%eax,%ebx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%eax,%ebx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%eax,%ebx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%rax,%rbx), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%rax,%rbx), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%rax,%rbx), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%eax,%ebx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%eax,%ebx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%eax,%ebx,2), %rcx
+# CHECK-NEXT:  2      1     0.50                        leaw	1024(%rax,%rbx,2), %cx
+# CHECK-NEXT:  2      1     0.50                        leal	1024(%rax,%rbx,2), %ecx
+# CHECK-NEXT:  2      1     0.50                        leaq	1024(%rax,%rbx,2), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     67.50  67.50   -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     67.50  67.50   -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	0, %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	0, %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	0, %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%eax), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%eax), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%eax), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%rax), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(,%ebx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(,%ebx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(,%ebx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(,%rbx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(,%rbx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(,%rbx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%eax,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%eax,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%eax,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%rax,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%rax,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%rax,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%eax,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%eax,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%eax,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%rax,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%rax,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%rax,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%eax,%ebx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%eax,%ebx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%eax,%ebx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	(%rax,%rbx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	(%rax,%rbx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	(%rax,%rbx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16, %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16, %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16, %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%eax), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%eax), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%eax), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%rax), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%rax), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%rax), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(,%ebx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(,%ebx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(,%ebx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(,%rbx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(,%rbx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(,%rbx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%eax,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%eax,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%eax,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%rax,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%rax,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%rax,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%eax,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%eax,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%eax,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%rax,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%rax,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%rax,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%eax,%ebx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%eax,%ebx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%eax,%ebx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	-16(%rax,%rbx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	-16(%rax,%rbx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	-16(%rax,%rbx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024, %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024, %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024, %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%eax), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%eax), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%eax), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%rax), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%rax), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%rax), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(,%ebx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(,%ebx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(,%ebx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(,%rbx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(,%rbx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(,%rbx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%eax,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%eax,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%eax,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%rax,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%rax,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%rax,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%eax,%ebx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%eax,%ebx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%eax,%ebx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%rax,%rbx), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%rax,%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%rax,%rbx), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%eax,%ebx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%eax,%ebx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%eax,%ebx,2), %rcx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaw	1024(%rax,%rbx,2), %cx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leal	1024(%rax,%rbx,2), %ecx
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	1024(%rax,%rbx,2), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	0, %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	0, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%eax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%eax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%eax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%rax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(,%ebx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(,%ebx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(,%ebx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(,%rbx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(,%rbx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(,%rbx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%eax,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%eax,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%eax,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%rax,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%rax,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%rax,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%eax,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%eax,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%eax,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%rax,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%rax,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%rax,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%eax,%ebx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%eax,%ebx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%eax,%ebx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	(%rax,%rbx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	(%rax,%rbx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	(%rax,%rbx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16, %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%eax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%eax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%eax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%rax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(,%ebx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(,%ebx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(,%ebx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(,%rbx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(,%rbx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(,%rbx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%eax,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%eax,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%eax,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%rax,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%rax,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%rax,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%eax,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%eax,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%eax,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%rax,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%rax,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%rax,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%eax,%ebx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%eax,%ebx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%eax,%ebx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	-16(%rax,%rbx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	-16(%rax,%rbx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	-16(%rax,%rbx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024, %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%eax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%eax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%eax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%rax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(,%ebx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(,%ebx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(,%ebx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(,%rbx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(,%rbx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(,%rbx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%eax,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%eax,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%eax,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%rax,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%rax,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%rax,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%eax,%ebx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%eax,%ebx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%eax,%ebx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%rax,%rbx), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%rax,%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%rax,%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%eax,%ebx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%eax,%ebx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%eax,%ebx,2), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaw	1024(%rax,%rbx,2), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leal	1024(%rax,%rbx,2), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	1024(%rax,%rbx,2), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-lzcnt.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-lzcnt.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-lzcnt.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 lzcntw      %cx, %cx
 lzcntw      (%rax), %cx
@@ -19,32 +19,44 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        lzcntw	%cx, %cx
-# CHECK-NEXT:  2      8     1.00    *                   lzcntw	(%rax), %cx
-# CHECK-NEXT:  1      3     1.00                        lzcntl	%eax, %ecx
-# CHECK-NEXT:  2      8     1.00    *                   lzcntl	(%rax), %ecx
-# CHECK-NEXT:  1      3     1.00                        lzcntq	%rax, %rcx
-# CHECK-NEXT:  2      8     1.00    *                   lzcntq	(%rax), %rcx
+# CHECK-NEXT:  2      2     0.50                        lzcntw	%cx, %cx
+# CHECK-NEXT:  2      6     0.50    *                   lzcntw	(%rax), %cx
+# CHECK-NEXT:  2      2     0.50                        lzcntl	%eax, %ecx
+# CHECK-NEXT:  2      6     0.50    *                   lzcntl	(%rax), %ecx
+# CHECK-NEXT:  2      2     0.50                        lzcntq	%rax, %rcx
+# CHECK-NEXT:  2      6     0.50    *                   lzcntq	(%rax), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     6.00    -      -     1.50   1.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 1.50   1.50    -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     lzcntw	%cx, %cx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   lzcntw	(%rax), %cx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     lzcntl	%eax, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   lzcntl	(%rax), %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     lzcntq	%rax, %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   lzcntq	(%rax), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntw	%cx, %cx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntw	(%rax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntl	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntq	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lzcntq	(%rax), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-mmx.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-mmx.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-mmx.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 emms
 
@@ -164,230 +164,242 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  31     31    10.33   *      *      U     emms
-# CHECK-NEXT:  1      1     1.00                        movd	%eax, %mm2
+# CHECK-NEXT:  1      2     0.50    *      *      U     emms
+# CHECK-NEXT:  2      10    0.50                        movd	%eax, %mm2
 # CHECK-NEXT:  1      5     0.50    *                   movd	(%rax), %mm2
-# CHECK-NEXT:  1      2     1.00                        movd	%mm0, %ecx
-# CHECK-NEXT:  1      1     1.00           *      U     movd	%mm0, (%rax)
-# CHECK-NEXT:  1      1     1.00                        movq	%rax, %mm2
+# CHECK-NEXT:  1      10    1.00                        movd	%mm0, %ecx
+# CHECK-NEXT:  1      2     1.00           *      U     movd	%mm0, (%rax)
+# CHECK-NEXT:  2      10    0.50                        movq	%rax, %mm2
 # CHECK-NEXT:  1      5     0.50    *                   movq	(%rax), %mm2
-# CHECK-NEXT:  1      2     1.00                        movq	%mm0, %rcx
-# CHECK-NEXT:  1      1     1.00           *            movq	%mm0, (%rax)
-# CHECK-NEXT:  1      1     1.00                        packsswb	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   packsswb	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        packssdw	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   packssdw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        packuswb	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   packuswb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddd	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddd	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddsb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddsb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddsw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddsw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddusb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddusb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddusw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddusw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        paddw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   paddw	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.33                        pand	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pand	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.33                        pandn	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pandn	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpeqb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pcmpeqb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpeqd	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pcmpeqd	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpeqw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pcmpeqw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpgtb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pcmpgtb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpgtd	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pcmpgtd	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpgtw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pcmpgtw	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmaddwd	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmaddwd	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmulhw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmulhw	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmullw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmullw	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.33                        por	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   por	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        pslld	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        pslld	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   pslld	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psllq	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psllq	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psllq	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psllw	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psllw	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psllw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psrad	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psrad	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psrad	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psraw	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psraw	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psraw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psrld	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psrld	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psrld	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psrlq	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psrlq	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psrlq	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        psrlw	$1, %mm2
-# CHECK-NEXT:  1      1     1.00                        psrlw	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   psrlw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubd	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubd	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubsb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubsb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubsw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubsw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubusb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubusb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubusw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubusw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        psubw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        punpckhbw	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   punpckhbw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        punpckhdq	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   punpckhdq	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        punpckhwd	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   punpckhwd	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        punpcklbw	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   punpcklbw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        punpckldq	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   punpckldq	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        punpcklwd	%mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   punpcklwd	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.33                        pxor	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pxor	(%rax), %mm2
+# CHECK-NEXT:  1      10    1.00                        movq	%mm0, %rcx
+# CHECK-NEXT:  1      2     1.00           *            movq	%mm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        packsswb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   packsswb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        packssdw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   packssdw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        packuswb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   packuswb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddsb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddsb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddsw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddsw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddusb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddusb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddusw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddusw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pand	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pand	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pandn	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pandn	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtw	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmaddwd	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmaddwd	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmulhw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhw	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmullw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmullw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        por	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   por	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pslld	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        pslld	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   pslld	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psllq	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psllq	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psllq	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psllw	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psllw	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psllw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psrad	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psrad	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psrad	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psraw	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psraw	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psraw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psrld	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psrld	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psrld	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psrlq	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psrlq	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psrlq	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psrlw	$1, %mm2
+# CHECK-NEXT:  1      3     0.50                        psrlw	%mm0, %mm2
+# CHECK-NEXT:  1      8     0.50    *                   psrlw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubsb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubsb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubsw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubsw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubusb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubusb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubusw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubusw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        punpckhbw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhbw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        punpckhdq	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhdq	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        punpckhwd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhwd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        punpcklbw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   punpcklbw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        punpckldq	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckldq	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        punpcklwd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   punpcklwd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pxor	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pxor	(%rax), %mm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     21.00  53.00  2.00   57.00  24.00  24.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 24.00  24.00   -      -      -     2.00    -      -     2.50   2.50   46.00  46.00  6.00   2.00   55.50  49.50   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     10.33  10.33   -     10.33   -      -     emms
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movd	%eax, %mm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movd	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movd	%mm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movd	%mm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movq	%rax, %mm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movq	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movq	%mm0, %rcx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movq	%mm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     packsswb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   packsswb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     packssdw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   packssdw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     packuswb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   packuswb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddsb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddsb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddusb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddusb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddusw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddusw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     paddw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   paddw	(%rax), %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     pand	%mm0, %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   pand	(%rax), %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     pandn	%mm0, %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   pandn	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pcmpeqb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pcmpeqd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pcmpeqw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpgtb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pcmpgtb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpgtd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pcmpgtd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpgtw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pcmpgtw	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmaddwd	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmaddwd	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhw	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmullw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmullw	(%rax), %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     por	%mm0, %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   por	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     pslld	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     pslld	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   pslld	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psllq	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psllq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psllq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psllw	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psllw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psllw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrad	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrad	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psrad	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psraw	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psraw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psraw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrld	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrld	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psrld	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrlq	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrlq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psrlq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrlw	$1, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psrlw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   psrlw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubsb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubsb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubusb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubusb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubusw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubusw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     punpckhbw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   punpckhbw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     punpckhdq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   punpckhdq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     punpckhwd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   punpckhwd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     punpcklbw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   punpcklbw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     punpckldq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   punpckldq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     punpcklwd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   punpcklwd	(%rax), %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     pxor	%mm0, %mm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   pxor	(%rax), %mm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     emms
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movd	%eax, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     movd	%mm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movd	%mm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movq	%rax, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     movq	%mm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movq	%mm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packsswb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packsswb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packssdw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packssdw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packuswb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packuswb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pand	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pand	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pandn	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pandn	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddwd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddwd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmullw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmullw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     por	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     por	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslld	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslld	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslld	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllq	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllw	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrad	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrad	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrad	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psraw	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psraw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psraw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrld	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrld	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrld	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlq	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlw	$1, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhbw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhbw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhdq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhdq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhwd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhwd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklbw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklbw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckldq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckldq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklwd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklwd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pxor	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pxor	(%rax), %mm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-movbe.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-movbe.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-movbe.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 movbe  %cx, (%rax)
 movbe  (%rax), %cx
@@ -19,32 +19,44 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     1.00           *            movbew	%cx, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   movbew	(%rax), %cx
-# CHECK-NEXT:  1      1     1.00           *            movbel	%ecx, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   movbel	(%rax), %ecx
-# CHECK-NEXT:  1      1     1.00           *            movbeq	%rcx, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   movbeq	(%rax), %rcx
+# CHECK-NEXT:  1      1     0.50           *            movbew	%cx, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   movbew	(%rax), %cx
+# CHECK-NEXT:  1      1     0.50           *            movbel	%ecx, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   movbel	(%rax), %ecx
+# CHECK-NEXT:  1      1     0.50           *            movbeq	%rcx, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   movbeq	(%rax), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00   3.00   1.00   3.00   3.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 3.00   3.00    -      -      -     1.50   1.50    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movbew	%cx, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   movbew	(%rax), %cx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movbel	%ecx, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   movbel	(%rax), %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movbeq	%rcx, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   movbeq	(%rax), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     movbew	%cx, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movbew	(%rax), %cx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     movbel	%ecx, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movbel	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     movbeq	%rcx, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movbeq	(%rax), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-pclmul.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-pclmul.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-pclmul.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 pclmulqdq     $11, %xmm0, %xmm2
 pclmulqdq     $11, (%rax), %xmm2
@@ -13,24 +13,36 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      14    6.00                        pclmulqdq	$11, %xmm0, %xmm2
-# CHECK-NEXT:  1      14    5.67    *                   pclmulqdq	$11, (%rax), %xmm2
+# CHECK-NEXT:  5      12    1.00                        pclmulqdq	$11, %xmm0, %xmm2
+# CHECK-NEXT:  6      17    1.00    *                   pclmulqdq	$11, (%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     11.67  11.67   -     11.67  0.50   0.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     2.00    -     2.00    -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     6.00   6.00    -     6.00    -      -     pclmulqdq	$11, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     5.67   5.67    -     5.67   0.50   0.50   pclmulqdq	$11, (%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pclmulqdq	$11, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pclmulqdq	$11, (%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-popcnt.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-popcnt.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-popcnt.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 popcntw     %cx, %cx
 popcntw     (%rax), %cx
@@ -19,32 +19,44 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        popcntw	%cx, %cx
-# CHECK-NEXT:  2      9     1.00    *                   popcntw	(%rax), %cx
-# CHECK-NEXT:  1      3     1.00                        popcntl	%eax, %ecx
-# CHECK-NEXT:  2      9     1.00    *                   popcntl	(%rax), %ecx
-# CHECK-NEXT:  1      3     1.00                        popcntq	%rax, %rcx
-# CHECK-NEXT:  2      9     1.00    *                   popcntq	(%rax), %rcx
+# CHECK-NEXT:  1      4     0.50                        popcntw	%cx, %cx
+# CHECK-NEXT:  1      8     0.50    *                   popcntw	(%rax), %cx
+# CHECK-NEXT:  1      4     0.50                        popcntl	%eax, %ecx
+# CHECK-NEXT:  1      8     0.50    *                   popcntl	(%rax), %ecx
+# CHECK-NEXT:  1      4     0.50                        popcntq	%rax, %rcx
+# CHECK-NEXT:  1      8     0.50    *                   popcntq	(%rax), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     6.00    -      -     1.50   1.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 1.50   1.50    -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     popcntw	%cx, %cx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   popcntw	(%rax), %cx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     popcntl	%eax, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   popcntl	(%rax), %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     popcntq	%rax, %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   popcntq	(%rax), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     popcntw	%cx, %cx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     popcntw	(%rax), %cx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     popcntl	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     popcntl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     popcntq	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     popcntq	(%rax), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-prefetchw.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-prefetchw.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-prefetchw.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 prefetch    (%rax)
 prefetchw   (%rax)
@@ -17,20 +17,32 @@
 # CHECK-NEXT:  1      5     0.50    *      *            prefetchw	(%rax)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetch	(%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetchw	(%rax)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetch	(%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetchw	(%rax)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 addps       %xmm0, %xmm2
 addps       (%rax), %xmm2
@@ -194,268 +194,280 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        addps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   addps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        addss	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   addss	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        andnps	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   andnps	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        andps	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   andps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        cmpps	$0, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cmpps	$0, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        cmpss	$0, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cmpss	$0, (%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        comiss	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   comiss	(%rax), %xmm1
-# CHECK-NEXT:  1      3     1.00                        cvtpi2ps	%mm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cvtpi2ps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        cvtps2pi	%xmm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   cvtps2pi	(%rax), %mm2
-# CHECK-NEXT:  3      5     2.00                        cvtsi2ssl	%ecx, %xmm2
-# CHECK-NEXT:  3      5     2.00                        cvtsi2ssq	%rcx, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtsi2ssl	(%rax), %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtsi2ssl	(%rax), %xmm2
-# CHECK-NEXT:  2      5     1.00                        cvtss2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        cvtss2si	%xmm0, %rcx
-# CHECK-NEXT:  3      9     1.00    *                   cvtss2si	(%rax), %ecx
-# CHECK-NEXT:  3      9     1.00    *                   cvtss2si	(%rax), %rcx
-# CHECK-NEXT:  1      3     1.00                        cvttps2pi	%xmm0, %mm2
-# CHECK-NEXT:  2      9     1.00    *                   cvttps2pi	(%rax), %mm2
-# CHECK-NEXT:  2      5     1.00                        cvttss2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        cvttss2si	%xmm0, %rcx
-# CHECK-NEXT:  3      9     1.00    *                   cvttss2si	(%rax), %ecx
-# CHECK-NEXT:  3      9     1.00    *                   cvttss2si	(%rax), %rcx
-# CHECK-NEXT:  1      14    14.00                       divps	%xmm0, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   divps	(%rax), %xmm2
-# CHECK-NEXT:  1      14    14.00                       divss	%xmm0, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   divss	(%rax), %xmm2
-# CHECK-NEXT:  4      5     1.00    *      *      U     ldmxcsr	(%rax)
-# CHECK-NEXT:  1      1     1.00    *      *      U     maskmovq	%mm0, %mm1
-# CHECK-NEXT:  1      3     1.00                        maxps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   maxps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        maxss	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   maxss	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        minps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   minps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        minss	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   minss	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movaps	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     1.00                        addps	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   addps	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        addss	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   addss	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        andnps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   andnps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        andps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   andps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        cmpps	$0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   cmpps	$0, (%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        cmpss	$0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   cmpss	$0, (%rax), %xmm2
+# CHECK-NEXT:  2      1     1.00                        comiss	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   comiss	(%rax), %xmm1
+# CHECK-NEXT:  2      4     1.00                        cvtpi2ps	%mm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   cvtpi2ps	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        cvtps2pi	%xmm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   cvtps2pi	(%rax), %mm2
+# CHECK-NEXT:  2      4     1.00                        cvtsi2ssl	%ecx, %xmm2
+# CHECK-NEXT:  2      13    1.00                        cvtsi2ssq	%rcx, %xmm2
+# CHECK-NEXT:  2      9     1.00    *                   cvtsi2ssl	(%rax), %xmm2
+# CHECK-NEXT:  2      9     1.00    *                   cvtsi2ssl	(%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        cvtss2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        cvtss2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   cvtss2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   cvtss2si	(%rax), %rcx
+# CHECK-NEXT:  1      4     1.00                        cvttps2pi	%xmm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   cvttps2pi	(%rax), %mm2
+# CHECK-NEXT:  2      13    1.00                        cvttss2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        cvttss2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   cvttss2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   cvttss2si	(%rax), %rcx
+# CHECK-NEXT:  1      9     9.50                        divps	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   divps	(%rax), %xmm2
+# CHECK-NEXT:  1      9     9.50                        divss	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   divss	(%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *      *      U     ldmxcsr	(%rax)
+# CHECK-NEXT:  1      2     0.50    *      *      U     maskmovq	%mm0, %mm1
+# CHECK-NEXT:  1      2     1.00                        maxps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   maxps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        maxss	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   maxss	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        minps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   minps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        minss	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   minss	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50                        movaps	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            movaps	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movaps	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movhlps	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00                        movlhps	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            movhps	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   movhps	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00           *            movlps	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   movlps	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        movmskps	%xmm0, %ecx
-# CHECK-NEXT:  1      1     1.00           *            movntps	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00    *      *      U     movntq	%mm0, (%rax)
-# CHECK-NEXT:  1      1     1.00                        movss	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            movss	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movss	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movups	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movaps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        movhlps	%xmm0, %xmm2
+# CHECK-NEXT:  1      2     0.50                        movlhps	%xmm0, %xmm2
+# CHECK-NEXT:  2      2     1.00           *            movhps	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   movhps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00           *            movlps	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   movlps	(%rax), %xmm2
+# CHECK-NEXT:  2      10    1.00                        movmskps	%xmm0, %ecx
+# CHECK-NEXT:  1      3     1.00           *            movntps	%xmm0, (%rax)
+# CHECK-NEXT:  1      2     1.00    *      *      U     movntq	%mm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        movss	%xmm0, %xmm2
+# CHECK-NEXT:  1      2     1.00           *            movss	%xmm0, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   movss	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50                        movups	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            movups	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movups	(%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movups	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        mulps	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   mulps	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   mulps	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        mulss	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   mulss	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        orps	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   orps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        pavgb	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pavgb	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pavgw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pavgw	(%rax), %mm2
-# CHECK-NEXT:  2      3     1.00                        pextrw	$1, %mm0, %ecx
-# CHECK-NEXT:  2      2     1.00                        pinsrw	$1, %eax, %mm2
-# CHECK-NEXT:  2      7     0.50    *                   pinsrw	$1, (%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pmaxsw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pmaxsw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pmaxub	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pmaxub	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pminsw	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pminsw	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        pminub	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   pminub	(%rax), %mm2
-# CHECK-NEXT:  1      2     1.00                        pmovmskb	%xmm0, %ecx
-# CHECK-NEXT:  1      5     1.00                        pmulhuw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmulhuw	(%rax), %mm2
+# CHECK-NEXT:  1      10    1.00    *                   mulss	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        orps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   orps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pavgb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pavgb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pavgw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pavgw	(%rax), %mm2
+# CHECK-NEXT:  2      13    1.00                        pextrw	$1, %mm0, %ecx
+# CHECK-NEXT:  2      2     0.50                        pinsrw	$1, %eax, %mm2
+# CHECK-NEXT:  2      6     0.50    *                   pinsrw	$1, (%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pmaxsw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxsw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pmaxub	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxub	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pminsw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pminsw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pminub	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pminub	(%rax), %mm2
+# CHECK-NEXT:  2      13    1.00                        pmovmskb	%xmm0, %ecx
+# CHECK-NEXT:  1      4     1.00                        pmulhuw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhuw	(%rax), %mm2
 # CHECK-NEXT:  1      5     0.50    *      *            prefetcht0	(%rax)
 # CHECK-NEXT:  1      5     0.50    *      *            prefetcht1	(%rax)
 # CHECK-NEXT:  1      5     0.50    *      *            prefetcht2	(%rax)
 # CHECK-NEXT:  1      5     0.50    *      *            prefetchnta	(%rax)
-# CHECK-NEXT:  1      5     1.00                        psadbw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   psadbw	(%rax), %mm2
-# CHECK-NEXT:  1      1     1.00                        pshufw	$1, %mm0, %mm2
-# CHECK-NEXT:  2      6     1.00    *                   pshufw	$1, (%rax), %mm2
+# CHECK-NEXT:  2      4     0.50                        psadbw	%mm0, %mm2
+# CHECK-NEXT:  2      9     0.50    *                   psadbw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pshufw	$1, %mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pshufw	$1, (%rax), %mm2
 # CHECK-NEXT:  1      5     1.00                        rcpps	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   rcpps	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   rcpps	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        rcpss	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   rcpss	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   rcpss	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        rsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   rsqrtps	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   rsqrtps	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        rsqrtss	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   rsqrtss	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00    *      *      U     sfence
-# CHECK-NEXT:  1      1     1.00                        shufps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   shufps	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      14    14.00                       sqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   sqrtps	(%rax), %xmm2
-# CHECK-NEXT:  1      14    14.00                       sqrtss	%xmm0, %xmm2
-# CHECK-NEXT:  2      20    14.00   *                   sqrtss	(%rax), %xmm2
-# CHECK-NEXT:  4      5     1.00    *      *      U     stmxcsr	(%rax)
-# CHECK-NEXT:  1      3     1.00                        subps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   subps	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        subss	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   subss	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        ucomiss	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   ucomiss	(%rax), %xmm1
-# CHECK-NEXT:  1      1     1.00                        unpckhps	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   unpckhps	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        unpcklps	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   unpcklps	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        xorps	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   xorps	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   rsqrtss	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50    *      *      U     sfence
+# CHECK-NEXT:  1      2     0.50                        shufps	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   shufps	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      9     10.50                       sqrtps	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    10.50   *                   sqrtps	(%rax), %xmm2
+# CHECK-NEXT:  1      9     10.50                       sqrtss	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    10.50   *                   sqrtss	(%rax), %xmm2
+# CHECK-NEXT:  2      1     0.50    *      *      U     stmxcsr	(%rax)
+# CHECK-NEXT:  1      5     1.00                        subps	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   subps	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        subss	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   subss	(%rax), %xmm2
+# CHECK-NEXT:  2      1     1.00                        ucomiss	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   ucomiss	(%rax), %xmm1
+# CHECK-NEXT:  1      2     0.50                        unpckhps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   unpckhps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        unpcklps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   unpcklps	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        xorps	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   xorps	(%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -     112.00 41.00  55.50  10.00  34.50  33.50  33.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 33.50  33.50   -      -      -     15.00   -      -     115.50 115.50 9.50   9.50   2.00   25.00  50.50  66.50   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   addps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   addss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     andnps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   andnps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     andps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   andps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmpps	$0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cmpps	$0, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmpss	$0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cmpss	$0, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     comiss	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   comiss	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cvtpi2ps	%mm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvtpi2ps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cvtps2pi	%xmm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvtps2pi	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     cvtsi2ssl	%ecx, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     cvtsi2ssq	%rcx, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtsi2ssl	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtsi2ssl	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvtss2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvtss2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvtss2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvtss2si	(%rax), %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cvttps2pi	%xmm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvttps2pi	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvttss2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvttss2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvttss2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvttss2si	(%rax), %rcx
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     divps	%xmm0, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   divps	(%rax), %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     divss	%xmm0, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   divss	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   0.50   0.50   ldmxcsr	(%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     maskmovq	%mm0, %mm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     maxps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   maxps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     maxss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   maxss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     minps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   minps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     minss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   minss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movaps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movaps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movaps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movhlps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movlhps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movhps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   movhps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movlps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   movlps	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movmskps	%xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntps	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntq	%mm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movss	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movups	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movups	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movups	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     mulps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   mulps	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     mulss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   mulss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     orps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   orps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pavgb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pavgb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pavgw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pavgw	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     pextrw	$1, %mm0, %ecx
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     pinsrw	$1, %eax, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pinsrw	$1, (%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pmaxsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pmaxsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pmaxub	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pmaxub	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pminsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pminsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pminub	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   pminub	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmovmskb	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhuw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhuw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetcht0	(%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetcht1	(%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetcht2	(%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   prefetchnta	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psadbw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   psadbw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     pshufw	$1, %mm0, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   pshufw	$1, (%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     rcpps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   rcpps	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     rcpss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   rcpss	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     rsqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   rsqrtps	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     rsqrtss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   rsqrtss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   sfence
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     shufps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   shufps	$1, (%rax), %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     sqrtps	%xmm0, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   sqrtps	(%rax), %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     sqrtss	%xmm0, %xmm2
-# CHECK-NEXT:  -     14.00  1.00    -      -      -     0.50   0.50   sqrtss	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   0.50   0.50   stmxcsr	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     subps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   subps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     subss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   subss	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     ucomiss	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   ucomiss	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     unpckhps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   unpckhps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     unpcklps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   unpcklps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     xorps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   xorps	(%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andnps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andnps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmpps	$0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmpps	$0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmpss	$0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmpss	$0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     comiss	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     comiss	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpi2ps	%mm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpi2ps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtps2pi	%xmm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtps2pi	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2ssl	%ecx, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2ssq	%rcx, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2ssl	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2ssl	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtss2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtss2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtss2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtss2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttps2pi	%xmm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttps2pi	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttss2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttss2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttss2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttss2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divss	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     ldmxcsr	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     maskmovq	%mm0, %mm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movaps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movaps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movaps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movhlps	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movlhps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movhps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movhps	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movlps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movlps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     movmskps	%xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movntps	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movntq	%mm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movss	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movups	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movups	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movups	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     orps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     orps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pextrw	$1, %mm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrw	$1, %eax, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrw	$1, (%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxub	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxub	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminub	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminub	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pmovmskb	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhuw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhuw	(%rax), %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetcht0	(%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetcht1	(%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetcht2	(%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     prefetchnta	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psadbw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psadbw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshufw	$1, %mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshufw	$1, (%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rcpps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rcpps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rcpss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rcpss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rsqrtps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rsqrtps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rsqrtss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     rsqrtss	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     sfence
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     shufps	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     shufps	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     sqrtps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     sqrtps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     sqrtss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     10.50  10.50   -      -      -      -      -     1.00    -      -      -      -     sqrtss	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     stmxcsr	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ucomiss	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ucomiss	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpckhps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpckhps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpcklps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpcklps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     xorps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     xorps	(%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 addpd       %xmm0, %xmm2
 addpd       (%rax), %xmm2
@@ -402,548 +402,560 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        addpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   addpd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        addsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   addsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        andnpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   andnpd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        andpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   andpd	(%rax), %xmm2
-# CHECK-NEXT:  4      5     1.00    *      *      U     clflush	(%rax)
-# CHECK-NEXT:  1      3     1.00                        cmppd	$0, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cmppd	$0, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        cmpsd	$0, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cmpsd	$0, (%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        comisd	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   comisd	(%rax), %xmm1
-# CHECK-NEXT:  2      4     1.00                        cvtdq2pd	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtdq2pd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        cvtdq2ps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cvtdq2ps	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        cvtpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtpd2dq	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        cvtpd2pi	%xmm0, %mm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtpd2pi	(%rax), %mm2
-# CHECK-NEXT:  2      4     1.00                        cvtpd2ps	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtpd2ps	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        cvtpi2pd	%mm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtpi2pd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        cvtps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cvtps2dq	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        cvtps2pd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   cvtps2pd	(%rax), %xmm2
-# CHECK-NEXT:  2      5     1.00                        cvtsd2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        cvtsd2si	%xmm0, %rcx
-# CHECK-NEXT:  3      9     1.00    *                   cvtsd2si	(%rax), %ecx
-# CHECK-NEXT:  3      9     1.00    *                   cvtsd2si	(%rax), %rcx
-# CHECK-NEXT:  2      4     1.00                        cvtsd2ss	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvtsd2ss	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        addpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   addpd	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        addsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   addsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        andnpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   andnpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        andpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   andpd	(%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *      *      U     clflush	(%rax)
+# CHECK-NEXT:  1      2     1.00                        cmppd	$0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   cmppd	$0, (%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        cmpsd	$0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   cmpsd	$0, (%rax), %xmm2
+# CHECK-NEXT:  2      1     1.00                        comisd	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   comisd	(%rax), %xmm1
+# CHECK-NEXT:  2      8     1.00                        cvtdq2pd	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   cvtdq2pd	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        cvtdq2ps	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   cvtdq2ps	(%rax), %xmm2
+# CHECK-NEXT:  2      8     1.00                        cvtpd2dq	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   cvtpd2dq	(%rax), %xmm2
+# CHECK-NEXT:  2      6     1.00                        cvtpd2pi	%xmm0, %mm2
+# CHECK-NEXT:  2      13    1.00    *                   cvtpd2pi	(%rax), %mm2
+# CHECK-NEXT:  2      8     1.00                        cvtpd2ps	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   cvtpd2ps	(%rax), %xmm2
+# CHECK-NEXT:  2      6     1.00                        cvtpi2pd	%mm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   cvtpi2pd	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        cvtps2dq	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   cvtps2dq	(%rax), %xmm2
+# CHECK-NEXT:  2      8     1.00                        cvtps2pd	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   cvtps2pd	(%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        cvtsd2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        cvtsd2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   cvtsd2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   cvtsd2si	(%rax), %rcx
+# CHECK-NEXT:  1      4     1.00                        cvtsd2ss	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   cvtsd2ss	(%rax), %xmm2
 # CHECK-NEXT:  2      4     1.00                        cvtsi2sdl	%ecx, %xmm2
-# CHECK-NEXT:  2      4     1.00                        cvtsi2sdq	%rcx, %xmm2
+# CHECK-NEXT:  2      13    1.00                        cvtsi2sdq	%rcx, %xmm2
 # CHECK-NEXT:  2      9     1.00    *                   cvtsi2sdl	(%rax), %xmm2
 # CHECK-NEXT:  2      9     1.00    *                   cvtsi2sdl	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        cvtss2sd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   cvtss2sd	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        cvttpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  3      10    1.00    *                   cvttpd2dq	(%rax), %xmm2
-# CHECK-NEXT:  2      4     1.00                        cvttpd2pi	%xmm0, %mm2
-# CHECK-NEXT:  3      10    1.00    *                   cvttpd2pi	(%rax), %mm2
-# CHECK-NEXT:  1      3     1.00                        cvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   cvttps2dq	(%rax), %xmm2
-# CHECK-NEXT:  2      5     1.00                        cvttsd2si	%xmm0, %ecx
-# CHECK-NEXT:  2      5     1.00                        cvttsd2si	%xmm0, %rcx
-# CHECK-NEXT:  3      9     1.00    *                   cvttsd2si	(%rax), %ecx
-# CHECK-NEXT:  3      9     1.00    *                   cvttsd2si	(%rax), %rcx
-# CHECK-NEXT:  1      22    22.00                       divpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      28    22.00   *                   divpd	(%rax), %xmm2
-# CHECK-NEXT:  1      22    22.00                       divsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      28    22.00   *                   divsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00    *      *      U     lfence
+# CHECK-NEXT:  1      4     1.00                        cvtss2sd	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   cvtss2sd	(%rax), %xmm2
+# CHECK-NEXT:  2      8     1.00                        cvttpd2dq	%xmm0, %xmm2
+# CHECK-NEXT:  2      13    1.00    *                   cvttpd2dq	(%rax), %xmm2
+# CHECK-NEXT:  2      6     1.00                        cvttpd2pi	%xmm0, %mm2
+# CHECK-NEXT:  2      13    1.00    *                   cvttpd2pi	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        cvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   cvttps2dq	(%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        cvttsd2si	%xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00                        cvttsd2si	%xmm0, %rcx
+# CHECK-NEXT:  2      18    1.00    *                   cvttsd2si	(%rax), %ecx
+# CHECK-NEXT:  2      18    1.00    *                   cvttsd2si	(%rax), %rcx
+# CHECK-NEXT:  1      9     9.50                        divpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   divpd	(%rax), %xmm2
+# CHECK-NEXT:  1      9     9.50                        divsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    9.50    *                   divsd	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50    *      *      U     lfence
 # CHECK-NEXT:  1      1     1.00    *      *      U     maskmovdqu	%xmm0, %xmm1
-# CHECK-NEXT:  1      3     1.00                        maxpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   maxpd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        maxsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   maxsd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        minpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   minpd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        minsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   minsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movapd	%xmm0, %xmm2
+# CHECK-NEXT:  1      2     1.00                        maxpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   maxpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        maxsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   maxsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        minpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   minpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00                        minsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     1.00    *                   minsd	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50                        movapd	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            movapd	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movapd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movd	%eax, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   movd	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        movd	%xmm0, %ecx
-# CHECK-NEXT:  1      1     1.00           *            movd	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     0.33                        movdqa	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movapd	(%rax), %xmm2
+# CHECK-NEXT:  2      10    0.50                        movd	%eax, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movd	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00                        movd	%xmm0, %ecx
+# CHECK-NEXT:  1      2     1.00           *            movd	%xmm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        movdqa	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            movdqa	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movdqa	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.33                        movdqu	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movdqa	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        movdqu	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            movdqu	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movdqu	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        movdq2q	%xmm0, %mm2
-# CHECK-NEXT:  1      1     1.00           *            movhpd	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   movhpd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00           *            movlpd	%xmm0, (%rax)
-# CHECK-NEXT:  2      7     1.00    *                   movlpd	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        movmskpd	%xmm0, %ecx
-# CHECK-NEXT:  1      1     1.00           *            movntil	%eax, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            movntiq	%rax, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            movntdq	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            movntpd	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     0.33                        movq	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00                        movq	%rax, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   movq	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        movq	%xmm0, %rcx
-# CHECK-NEXT:  1      1     1.00           *            movq	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     0.33                        movq2dq	%mm0, %xmm2
-# CHECK-NEXT:  1      1     1.00                        movsd	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            movsd	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movupd	%xmm0, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movdqu	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        movdq2q	%xmm0, %mm2
+# CHECK-NEXT:  2      2     1.00           *            movhpd	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   movhpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     1.00           *            movlpd	%xmm0, (%rax)
+# CHECK-NEXT:  1      7     0.50    *                   movlpd	(%rax), %xmm2
+# CHECK-NEXT:  2      10    1.00                        movmskpd	%xmm0, %ecx
+# CHECK-NEXT:  1      1     0.50           *            movntil	%eax, (%rax)
+# CHECK-NEXT:  1      1     0.50           *            movntiq	%rax, (%rax)
+# CHECK-NEXT:  1      2     1.00           *            movntdq	%xmm0, (%rax)
+# CHECK-NEXT:  1      3     1.00           *            movntpd	%xmm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        movq	%xmm0, %xmm2
+# CHECK-NEXT:  2      10    0.50                        movq	%rax, %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movq	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00                        movq	%xmm0, %rcx
+# CHECK-NEXT:  1      2     1.00           *            movq	%xmm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        movq2dq	%mm0, %xmm2
+# CHECK-NEXT:  1      2     0.50                        movsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      2     1.00           *            movsd	%xmm0, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   movsd	(%rax), %xmm2
+# CHECK-NEXT:  1      1     0.50                        movupd	%xmm0, %xmm2
 # CHECK-NEXT:  1      1     1.00           *            movupd	%xmm0, (%rax)
-# CHECK-NEXT:  1      6     0.50    *                   movupd	(%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movupd	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        mulpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   mulpd	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   mulpd	(%rax), %xmm2
 # CHECK-NEXT:  1      5     1.00                        mulsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   mulsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        orpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   orpd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        packssdw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   packssdw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        packsswb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   packsswb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        packuswb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   packuswb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddq	%mm0, %mm2
-# CHECK-NEXT:  2      7     0.50    *                   paddq	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        paddq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddsb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddsb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddusb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddusb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddusw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddusw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        paddw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   paddw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.33                        pand	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pand	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.33                        pandn	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pandn	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pavgb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pavgb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pavgw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pavgw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpeqb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpeqd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpeqw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpgtb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpgtb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpgtd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpgtd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpgtw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpgtw	(%rax), %xmm2
-# CHECK-NEXT:  2      3     1.00                        pextrw	$1, %xmm0, %ecx
-# CHECK-NEXT:  1      5     1.00                        pmaddwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmaddwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmaxsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmaxsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmaxub	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmaxub	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pminsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pminsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pminub	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pminub	(%rax), %xmm2
-# CHECK-NEXT:  1      2     1.00                        pmovmskb	%xmm0, %ecx
-# CHECK-NEXT:  1      5     1.00                        pmulhuw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmulhuw	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmulhw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmulhw	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmullw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmullw	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmuludq	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmuludq	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmuludq	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmuludq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.33                        por	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   por	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        psadbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   psadbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pshufd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pshufd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pshufhw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pshufhw	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pshuflw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pshuflw	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        pslld	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        pslld	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   pslld	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pslldq	$1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        psllq	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psllq	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psllq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        psllw	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psllw	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psllw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        psrad	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psrad	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psrad	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        psraw	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psraw	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psraw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        psrld	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psrld	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psrld	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psrldq	$1, %xmm2
-# CHECK-NEXT:  1      1     1.00                        psrlq	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psrlq	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psrlq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        psrlw	$1, %xmm2
-# CHECK-NEXT:  2      2     1.00                        psrlw	%xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   psrlw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        psubq	%mm0, %mm2
-# CHECK-NEXT:  2      8     1.00    *                   psubq	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        psubq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubsb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubsb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubusb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubusb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubusw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubusw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psubw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpckhbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpckhbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpckhdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpckhdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpckhqdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpckhqdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpckhwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpckhwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpcklbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpcklbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpckldq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpckldq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpcklqdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpcklqdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        punpcklwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   punpcklwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.33                        pxor	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pxor	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        shufpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   shufpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      21    21.00                       sqrtpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      27    21.00   *                   sqrtpd	(%rax), %xmm2
-# CHECK-NEXT:  1      21    21.00                       sqrtsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      27    21.00   *                   sqrtsd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        subpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   subpd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        subsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   subsd	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        ucomisd	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   ucomisd	(%rax), %xmm1
-# CHECK-NEXT:  1      1     1.00                        unpckhpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   unpckhpd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        unpcklpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   unpcklpd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        xorpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   xorpd	(%rax), %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   mulsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        orpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   orpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        packssdw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   packssdw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        packsswb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   packsswb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        packuswb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   packuswb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddq	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   paddq	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        paddq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddsb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddsb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddsw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddusb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddusb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddusw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddusw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        paddw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   paddw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pand	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pand	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pandn	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pandn	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pavgb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pavgb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pavgw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pavgw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtw	(%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        pextrw	$1, %xmm0, %ecx
+# CHECK-NEXT:  1      4     1.00                        pmaddwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmaddwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmaxsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxsw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmaxub	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxub	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pminsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pminsw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pminub	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pminub	(%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        pmovmskb	%xmm0, %ecx
+# CHECK-NEXT:  1      4     1.00                        pmulhuw	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhuw	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        pmulhw	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhw	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        pmullw	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmullw	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        pmuludq	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmuludq	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmuludq	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmuludq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        por	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   por	(%rax), %xmm2
+# CHECK-NEXT:  2      4     0.50                        psadbw	%xmm0, %xmm2
+# CHECK-NEXT:  2      9     0.50    *                   psadbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pshufd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pshufd	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pshufhw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pshufhw	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pshuflw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pshuflw	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pslld	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        pslld	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   pslld	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pslldq	$1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        psllq	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psllq	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psllq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psllw	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psllw	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psllw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psrad	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psrad	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psrad	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psraw	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psraw	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psraw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psrld	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psrld	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psrld	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psrldq	$1, %xmm2
+# CHECK-NEXT:  1      2     0.50                        psrlq	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psrlq	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psrlq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psrlw	$1, %xmm2
+# CHECK-NEXT:  1      3     0.50                        psrlw	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     0.50    *                   psrlw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubq	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psubq	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psubq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubsb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubsb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubsw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubusb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubusb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubusw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubusw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psubw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpckhbw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpckhdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpckhqdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhqdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpckhwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckhwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpcklbw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpcklbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpckldq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpckldq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpcklqdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpcklqdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        punpcklwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   punpcklwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pxor	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pxor	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        shufpd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   shufpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      9     13.50                       sqrtpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    13.50   *                   sqrtpd	(%rax), %xmm2
+# CHECK-NEXT:  1      9     13.50                       sqrtsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      14    13.50   *                   sqrtsd	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        subpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   subpd	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        subsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   subsd	(%rax), %xmm2
+# CHECK-NEXT:  2      1     1.00                        ucomisd	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   ucomisd	(%rax), %xmm1
+# CHECK-NEXT:  1      2     0.50                        unpckhpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   unpckhpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        unpcklpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   unpcklpd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        xorpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   xorpd	(%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -     172.00 75.83  117.33 16.00  98.83  66.00  66.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 66.00  66.00   -      -      -     17.00   -      -     124.50 124.50 66.50  66.50  12.00  50.00  119.50 140.50  -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   addpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   addsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     andnpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   andnpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     andpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   andpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.50   0.50   1.00   1.00   0.50   0.50   clflush	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmppd	$0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cmppd	$0, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmpsd	$0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cmpsd	$0, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     comisd	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   comisd	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtdq2pd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtdq2pd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cvtdq2ps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvtdq2ps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtpd2dq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtpd2pi	%xmm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtpd2pi	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtpd2ps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtpd2ps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtpi2pd	%mm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtpi2pd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cvtps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvtps2dq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     cvtps2pd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   cvtps2pd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvtsd2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvtsd2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvtsd2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvtsd2si	(%rax), %rcx
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtsd2ss	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvtsd2ss	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtsi2sdl	%ecx, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvtsi2sdq	%rcx, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvtsi2sdl	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvtsi2sdl	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     cvtss2sd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   cvtss2sd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvttpd2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvttpd2dq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     cvttpd2pi	%xmm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   cvttpd2pi	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cvttps2dq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   cvttps2dq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvttsd2si	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     cvttsd2si	%xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvttsd2si	(%rax), %ecx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   cvttsd2si	(%rax), %rcx
-# CHECK-NEXT:  -     22.00  1.00    -      -      -      -      -     divpd	%xmm0, %xmm2
-# CHECK-NEXT:  -     22.00  1.00    -      -      -     0.50   0.50   divpd	(%rax), %xmm2
-# CHECK-NEXT:  -     22.00  1.00    -      -      -      -      -     divsd	%xmm0, %xmm2
-# CHECK-NEXT:  -     22.00  1.00    -      -      -     0.50   0.50   divsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   lfence
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   maskmovdqu	%xmm0, %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     maxpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   maxpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     maxsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   maxsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     minpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   minpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     minsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   minsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movapd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movapd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movapd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movd	%eax, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movd	%xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movdqa	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movdqa	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movdqa	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movdqu	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movdqu	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movdqu	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     1.33    -      -     movdq2q	%xmm0, %mm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movhpd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   movhpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movlpd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   movlpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movmskpd	%xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntil	%eax, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntiq	%rax, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntdq	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntpd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movq	%rax, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     movq	%xmm0, %rcx
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movq	%xmm0, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movq2dq	%mm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movsd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movupd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movupd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movupd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     mulpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   mulpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     mulsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   mulsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     orpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   orpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     packssdw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   packssdw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     packsswb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   packsswb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     packuswb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   packuswb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddsb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddsb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddusb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddusb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddusw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddusw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     paddw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   paddw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     pand	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   pand	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     pandn	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   pandn	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pavgb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pavgb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pavgw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pavgw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpeqb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpeqb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpeqd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpeqd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpeqw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpeqw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpgtb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpgtb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpgtd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpgtd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpgtw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpgtw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     pextrw	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmaddwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmaddwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmaxsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmaxsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmaxub	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmaxub	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pminsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pminsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pminub	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pminub	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmovmskb	%xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhuw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhuw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmullw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmullw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmuludq	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmuludq	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmuludq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmuludq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     por	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   por	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psadbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   psadbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pshufd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pshufd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pshufhw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pshufhw	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pshuflw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pshuflw	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pslld	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     pslld	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   pslld	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pslldq	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psllq	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psllq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psllq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psllw	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psllw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psllw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psrad	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psrad	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psrad	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psraw	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psraw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psraw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psrld	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psrld	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psrld	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psrldq	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psrlq	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psrlq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psrlq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     psrlw	$1, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     psrlw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50   0.50   0.50   psrlw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubq	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   psubq	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubsb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubsb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubusb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubusb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubusw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubusw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psubw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psubw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpckhbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpckhbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpckhdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpckhdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpckhqdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpckhqdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpckhwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpckhwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpcklbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpcklbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpckldq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpckldq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpcklqdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpcklqdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     punpcklwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   punpcklwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     pxor	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   pxor	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     shufpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   shufpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -      -      -     sqrtpd	%xmm0, %xmm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -     0.50   0.50   sqrtpd	(%rax), %xmm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -      -      -     sqrtsd	%xmm0, %xmm2
-# CHECK-NEXT:  -     21.00  1.00    -      -      -     0.50   0.50   sqrtsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     subpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   subpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     subsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   subsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     ucomisd	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   ucomisd	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     unpckhpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   unpckhpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     unpcklpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   unpcklpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     xorpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   xorpd	(%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andnpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andnpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     andpd	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     clflush	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmppd	$0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmppd	$0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmpsd	$0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     cmpsd	$0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     comisd	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     comisd	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtdq2pd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtdq2pd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtdq2ps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtdq2ps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpd2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpd2dq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpd2pi	%xmm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpd2pi	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpd2ps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpd2ps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpi2pd	%mm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtpi2pd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtps2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtps2dq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtps2pd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtps2pd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtsd2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtsd2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtsd2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvtsd2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsd2ss	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsd2ss	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2sdl	%ecx, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2sdq	%rcx, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2sdl	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtsi2sdl	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtss2sd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvtss2sd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttpd2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttpd2dq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttpd2pi	%xmm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttpd2pi	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttps2dq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     cvttps2dq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttsd2si	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttsd2si	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttsd2si	(%rax), %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -     1.00    -     1.00    -      -      -      -     cvttsd2si	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     divsd	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     lfence
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     maskmovdqu	%xmm0, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     maxsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     minsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movapd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movapd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movapd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movd	%eax, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     movd	%xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movd	%xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movdqa	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movdqa	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movdqa	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movdqu	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movdqu	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movdqu	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movdq2q	%xmm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movhpd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movhpd	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movlpd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movlpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     movmskpd	%xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     movntil	%eax, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     movntiq	%rax, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movntdq	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movntpd	%xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movq	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movq	%rax, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     movq	%xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movq	%xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movq2dq	%mm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movsd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movupd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movupd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movupd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     mulsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     orpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     orpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packssdw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packssdw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packsswb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packsswb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packuswb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packuswb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddusw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     paddw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pand	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pand	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pandn	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pandn	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pavgw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pextrw	$1, %xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxub	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxub	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminub	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminub	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pmovmskb	%xmm0, %ecx
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhuw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhuw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmullw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmullw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmuludq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmuludq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmuludq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmuludq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     por	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     por	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psadbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psadbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshufd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshufd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshufhw	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshufhw	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshuflw	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pshuflw	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslld	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslld	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslld	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pslldq	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllq	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllw	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psllw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrad	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrad	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrad	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psraw	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psraw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psraw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrld	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrld	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrld	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrldq	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlq	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlw	$1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psrlw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubq	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubq	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubusw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psubw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhqdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhqdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckhwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckldq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpckldq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklqdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklqdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     punpcklwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pxor	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pxor	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     shufpd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     shufpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     sqrtpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     sqrtpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     sqrtsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     13.50  13.50   -      -      -      -      -     1.00    -      -      -      -     sqrtsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     subsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ucomisd	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ucomisd	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpckhpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpckhpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpcklpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     unpcklpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     xorpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     xorpd	(%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse3.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse3.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse3.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 addsubpd  %xmm0, %xmm2
 addsubpd  (%rax),  %xmm2
@@ -39,58 +39,70 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        addsubpd	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   addsubpd	(%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        addsubps	%xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   addsubps	(%rax), %xmm2
-# CHECK-NEXT:  3      5     2.00                        haddpd	%xmm0, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   haddpd	(%rax), %xmm2
-# CHECK-NEXT:  3      5     2.00                        haddps	%xmm0, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   haddps	(%rax), %xmm2
-# CHECK-NEXT:  3      5     2.00                        hsubpd	%xmm0, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   hsubpd	(%rax), %xmm2
-# CHECK-NEXT:  3      5     2.00                        hsubps	%xmm0, %xmm2
-# CHECK-NEXT:  4      11    2.00    *                   hsubps	(%rax), %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   lddqu	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movddup	%xmm0, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   movddup	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movshdup	%xmm0, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   movshdup	(%rax), %xmm2
-# CHECK-NEXT:  1      1     1.00                        movsldup	%xmm0, %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   movsldup	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        addsubpd	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   addsubpd	(%rax), %xmm2
+# CHECK-NEXT:  1      5     1.00                        addsubps	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    1.00    *                   addsubps	(%rax), %xmm2
+# CHECK-NEXT:  3      11    1.00                        haddpd	%xmm0, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   haddpd	(%rax), %xmm2
+# CHECK-NEXT:  3      11    1.00                        haddps	%xmm0, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   haddps	(%rax), %xmm2
+# CHECK-NEXT:  3      11    1.00                        hsubpd	%xmm0, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   hsubpd	(%rax), %xmm2
+# CHECK-NEXT:  3      11    1.00                        hsubps	%xmm0, %xmm2
+# CHECK-NEXT:  4      16    1.00    *                   hsubps	(%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   lddqu	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        movddup	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   movddup	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        movshdup	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   movshdup	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        movsldup	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   movsldup	(%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     12.00   -     19.00  5.00   5.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 5.00   5.00    -      -      -      -      -      -     9.00   9.00   0.50   0.50    -      -     15.50  3.50    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addsubpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   addsubpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     addsubps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   addsubps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     haddpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   haddpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     haddps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   haddps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     hsubpd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   hsubpd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00    -      -     hsubps	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     2.00   0.50   0.50   hsubps	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   lddqu	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movddup	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movddup	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movshdup	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movshdup	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     movsldup	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movsldup	(%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addsubpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addsubpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addsubps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     addsubps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     haddpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     haddpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     haddps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     haddps	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     hsubpd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     hsubpd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     hsubps	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     hsubps	(%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     lddqu	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movddup	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movddup	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movshdup	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movshdup	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movsldup	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     movsldup	(%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse41.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse41.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse41.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 blendpd     $11, %xmm0, %xmm2
 blendpd     $11, (%rax), %xmm2
@@ -155,212 +155,224 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.50                        blendpd	$11, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   blendpd	$11, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        blendps	$11, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   blendps	$11, (%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        blendvpd	%xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   blendvpd	%xmm0, (%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        blendvps	%xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   blendvps	%xmm0, (%rax), %xmm2
-# CHECK-NEXT:  3      9     1.00                        dppd	$22, %xmm0, %xmm2
-# CHECK-NEXT:  4      15    1.00    *                   dppd	$22, (%rax), %xmm2
-# CHECK-NEXT:  4      12    2.00                        dpps	$22, %xmm0, %xmm2
-# CHECK-NEXT:  5      18    2.00    *                   dpps	$22, (%rax), %xmm2
-# CHECK-NEXT:  2      3     1.00                        extractps	$1, %xmm0, %ecx
-# CHECK-NEXT:  3      5     1.00           *            extractps	$1, %xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00                        insertps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     1.00    *                   insertps	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      6     0.50    *                   movntdqa	(%rax), %xmm2
-# CHECK-NEXT:  3      7     1.00                        mpsadbw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  4      13    1.00    *                   mpsadbw	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        packusdw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   packusdw	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        pblendvb	%xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  3      8     1.00    *                   pblendvb	%xmm0, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pblendw	$11, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pblendw	$11, (%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pcmpeqq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pcmpeqq	(%rax), %xmm2
-# CHECK-NEXT:  2      3     1.00                        pextrb	$1, %xmm0, %ecx
-# CHECK-NEXT:  3      5     1.00           *            pextrb	$1, %xmm0, (%rax)
-# CHECK-NEXT:  2      3     1.00                        pextrd	$1, %xmm0, %ecx
-# CHECK-NEXT:  4      5     1.00           *            pextrd	$1, %xmm0, (%rax)
-# CHECK-NEXT:  2      3     1.00                        pextrq	$1, %xmm0, %rcx
-# CHECK-NEXT:  4      5     1.00           *            pextrq	$1, %xmm0, (%rax)
-# CHECK-NEXT:  3      5     1.00           *            pextrw	$1, %xmm0, (%rax)
-# CHECK-NEXT:  1      5     1.00                        phminposuw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   phminposuw	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        pinsrb	$1, %eax, %xmm1
-# CHECK-NEXT:  2      7     0.50    *                   pinsrb	$1, (%rax), %xmm1
-# CHECK-NEXT:  2      2     1.00                        pinsrd	$1, %eax, %xmm1
-# CHECK-NEXT:  2      7     0.50    *                   pinsrd	$1, (%rax), %xmm1
-# CHECK-NEXT:  2      2     1.00                        pinsrq	$1, %rax, %xmm1
-# CHECK-NEXT:  2      7     0.50    *                   pinsrq	$1, (%rax), %xmm1
-# CHECK-NEXT:  1      1     0.50                        pmaxsb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmaxsb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmaxsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmaxsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmaxud	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmaxud	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmaxuw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmaxuw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pminsb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pminsb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pminsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pminsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pminud	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pminud	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pminuw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pminuw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovsxbd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovsxbd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovsxbq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovsxbq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovsxbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovsxbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovsxdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovsxdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovsxwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovsxwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovsxwq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovsxwq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovzxbd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovzxbd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovzxbq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovzxbq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovzxbw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovzxbw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovzxdq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovzxdq	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovzxwd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovzxwd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pmovzxwq	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pmovzxwq	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmuldq	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmuldq	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmulld	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmulld	(%rax), %xmm2
-# CHECK-NEXT:  2      2     1.00                        ptest	%xmm0, %xmm1
-# CHECK-NEXT:  3      8     1.00    *                   ptest	(%rax), %xmm1
-# CHECK-NEXT:  1      3     1.00                        roundpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   roundpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        roundps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   roundps	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        roundsd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   roundsd	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      3     1.00                        roundss	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      9     1.00    *                   roundss	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        blendpd	$11, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   blendpd	$11, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        blendps	$11, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   blendps	$11, (%rax), %xmm2
+# CHECK-NEXT:  1      2     2.00                        blendvpd	%xmm0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     2.00    *                   blendvpd	%xmm0, (%rax), %xmm2
+# CHECK-NEXT:  1      2     2.00                        blendvps	%xmm0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     2.00    *                   blendvps	%xmm0, (%rax), %xmm2
+# CHECK-NEXT:  15     15    1.50                        dppd	$22, %xmm0, %xmm2
+# CHECK-NEXT:  17     20    1.50    *                   dppd	$22, (%rax), %xmm2
+# CHECK-NEXT:  16     25    1.50                        dpps	$22, %xmm0, %xmm2
+# CHECK-NEXT:  18     30    1.50    *                   dpps	$22, (%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        extractps	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            extractps	$1, %xmm0, (%rax)
+# CHECK-NEXT:  1      2     0.50                        insertps	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   insertps	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      5     0.50    *                   movntdqa	(%rax), %xmm2
+# CHECK-NEXT:  9      9     2.00                        mpsadbw	$1, %xmm0, %xmm2
+# CHECK-NEXT:  9      14    2.00    *                   mpsadbw	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        packusdw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   packusdw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     2.00                        pblendvb	%xmm0, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     2.00    *                   pblendvb	%xmm0, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pblendw	$11, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pblendw	$11, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpeqq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpeqq	(%rax), %xmm2
+# CHECK-NEXT:  2      13    1.00                        pextrb	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            pextrb	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      13    1.00                        pextrd	$1, %xmm0, %ecx
+# CHECK-NEXT:  2      13    1.00           *            pextrd	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      13    1.00                        pextrq	$1, %xmm0, %rcx
+# CHECK-NEXT:  2      13    1.00           *            pextrq	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      13    1.00           *            pextrw	$1, %xmm0, (%rax)
+# CHECK-NEXT:  2      4     1.00                        phminposuw	%xmm0, %xmm2
+# CHECK-NEXT:  2      9     1.00    *                   phminposuw	(%rax), %xmm2
+# CHECK-NEXT:  2      2     0.50                        pinsrb	$1, %eax, %xmm1
+# CHECK-NEXT:  2      6     0.50    *                   pinsrb	$1, (%rax), %xmm1
+# CHECK-NEXT:  2      2     0.50                        pinsrd	$1, %eax, %xmm1
+# CHECK-NEXT:  2      6     0.50    *                   pinsrd	$1, (%rax), %xmm1
+# CHECK-NEXT:  2      2     0.50                        pinsrq	$1, %rax, %xmm1
+# CHECK-NEXT:  2      6     0.50    *                   pinsrq	$1, (%rax), %xmm1
+# CHECK-NEXT:  1      2     0.50                        pmaxsb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxsb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmaxsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmaxud	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxud	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmaxuw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmaxuw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pminsb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pminsb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pminsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pminsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pminud	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pminud	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pminuw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pminuw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovsxbd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovsxbd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovsxbq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovsxbq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovsxbw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovsxbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovsxdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovsxdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovsxwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovsxwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovsxwq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovsxwq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovzxbd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovzxbd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovzxbq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovzxbq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovzxbw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovzxbw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovzxdq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovzxdq	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovzxwd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovzxwd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pmovzxwq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pmovzxwq	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        pmuldq	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmuldq	(%rax), %xmm2
+# CHECK-NEXT:  1      5     2.00                        pmulld	%xmm0, %xmm2
+# CHECK-NEXT:  1      10    2.00    *                   pmulld	(%rax), %xmm2
+# CHECK-NEXT:  2      1     1.00                        ptest	%xmm0, %xmm1
+# CHECK-NEXT:  2      6     1.00    *                   ptest	(%rax), %xmm1
+# CHECK-NEXT:  1      4     1.00                        roundpd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   roundpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        roundps	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   roundps	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        roundsd	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   roundsd	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        roundss	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   roundss	$1, (%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     26.00  47.50  5.00   52.50  24.50  24.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 24.50  24.50   -      -      -     6.00    -      -     20.00  20.00  32.50  32.50  10.00  13.00  49.50  50.50   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     blendpd	$11, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   blendpd	$11, (%rax), %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     blendps	$11, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   blendps	$11, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     blendvpd	%xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   blendvpd	%xmm0, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     blendvps	%xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   blendvps	%xmm0, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     dppd	$22, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   0.50   0.50   dppd	$22, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   2.00    -     1.00    -      -     dpps	$22, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   2.00    -     1.00   0.50   0.50   dpps	$22, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     extractps	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -     1.00   1.00   0.50   0.50   extractps	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     insertps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   insertps	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movntdqa	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     mpsadbw	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   0.50   0.50   mpsadbw	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     packusdw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   packusdw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -     pblendvb	%xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00   0.50   0.50   pblendvb	%xmm0, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pblendw	$11, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pblendw	$11, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pcmpeqq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pcmpeqq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     pextrb	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -     0.50   1.00   0.50   0.50   0.50   pextrb	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     pextrd	$1, %xmm0, %ecx
-# CHECK-NEXT:  -      -     1.00   0.50   1.00   0.50   0.50   0.50   pextrd	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -     1.00   0.50    -     0.50    -      -     pextrq	$1, %xmm0, %rcx
-# CHECK-NEXT:  -      -     1.00   0.50   1.00   0.50   0.50   0.50   pextrq	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -      -     0.50   1.00   0.50   0.50   0.50   pextrw	$1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     phminposuw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   phminposuw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     pinsrb	$1, %eax, %xmm1
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pinsrb	$1, (%rax), %xmm1
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     pinsrd	$1, %eax, %xmm1
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pinsrd	$1, (%rax), %xmm1
-# CHECK-NEXT:  -      -      -     0.50    -     1.50    -      -     pinsrq	$1, %rax, %xmm1
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pinsrq	$1, (%rax), %xmm1
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmaxsb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmaxsb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmaxsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmaxsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmaxud	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmaxud	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmaxuw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmaxuw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pminsb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pminsb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pminsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pminsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pminud	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pminud	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pminuw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pminuw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovsxbd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovsxbd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovsxbq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovsxbq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovsxbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovsxbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovsxdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovsxdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovsxwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovsxwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovsxwq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovsxwq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovzxbd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovzxbd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovzxbq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovzxbq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovzxbw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovzxbw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovzxdq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovzxdq	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovzxwd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovzxwd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pmovzxwq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pmovzxwq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmuldq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmuldq	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulld	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulld	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     ptest	%xmm0, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -     1.00   0.50   0.50   ptest	(%rax), %xmm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     roundpd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   roundpd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     roundps	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   roundps	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     roundsd	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   roundsd	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     roundss	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   roundss	$1, (%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     blendpd	$11, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     blendpd	$11, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     blendps	$11, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     blendps	$11, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     blendvpd	%xmm0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     blendvpd	%xmm0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     blendvps	%xmm0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     blendvps	%xmm0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     dppd	$22, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     dppd	$22, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     dpps	$22, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     1.50   1.50    -      -      -      -      -     1.00    -      -      -      -     dpps	$22, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     extractps	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     extractps	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     insertps	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     0.50   0.50    -      -      -      -     insertps	$1, (%rax), %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     movntdqa	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     2.00    -     1.00    -      -      -      -      -     mpsadbw	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     2.00    -     1.00    -      -      -      -      -     mpsadbw	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packusdw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     packusdw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     pblendvb	%xmm0, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     pblendvb	%xmm0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pblendw	$11, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pblendw	$11, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpeqq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pextrb	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pextrb	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pextrd	$1, %xmm0, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pextrd	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     pextrq	$1, %xmm0, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pextrq	$1, %xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     pextrw	$1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     1.00    -      -      -      -      -     phminposuw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     1.00    -      -      -      -      -     phminposuw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrb	$1, %eax, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrb	$1, (%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrd	$1, %eax, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrd	$1, (%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrq	$1, %rax, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pinsrq	$1, (%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxud	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxud	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxuw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmaxuw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminud	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminud	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminuw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pminuw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxbd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxbd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxbq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxbq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxwq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovsxwq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxbd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxbd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxbq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxbq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxbw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxbw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxdq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxdq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxwd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxwd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxwq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pmovzxwq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmuldq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmuldq	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     pmulld	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     pmulld	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ptest	%xmm0, %xmm1
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ptest	(%rax), %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundpd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundpd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundps	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundps	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundsd	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundsd	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundss	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     roundss	$1, (%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse42.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse42.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse42.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 crc32b      %al, %ecx
 crc32b      (%rax), %ecx
@@ -40,60 +40,72 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        crc32b	%al, %ecx
-# CHECK-NEXT:  2      8     1.00    *                   crc32b	(%rax), %ecx
-# CHECK-NEXT:  1      3     1.00                        crc32l	%eax, %ecx
-# CHECK-NEXT:  2      8     1.00    *                   crc32l	(%rax), %ecx
-# CHECK-NEXT:  1      3     1.00                        crc32w	%ax, %ecx
-# CHECK-NEXT:  2      8     1.00    *                   crc32w	(%rax), %ecx
-# CHECK-NEXT:  1      3     1.00                        crc32b	%al, %rcx
-# CHECK-NEXT:  2      8     1.00    *                   crc32b	(%rax), %rcx
-# CHECK-NEXT:  1      3     1.00                        crc32q	%rax, %rcx
-# CHECK-NEXT:  2      8     1.00    *                   crc32q	(%rax), %rcx
-# CHECK-NEXT:  1      4     2.67                        pcmpestri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  1      4     2.33    *                   pcmpestri	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      11    2.67                        pcmpestrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  1      11    2.33    *                   pcmpestrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  3      11    3.00                        pcmpistri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  4      17    3.00    *                   pcmpistri	$1, (%rax), %xmm2
-# CHECK-NEXT:  3      11    3.00                        pcmpistrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  4      17    3.00    *                   pcmpistrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pcmpgtq	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pcmpgtq	(%rax), %xmm2
+# CHECK-NEXT:  3      3     2.00                        crc32b	%al, %ecx
+# CHECK-NEXT:  3      7     2.00    *                   crc32b	(%rax), %ecx
+# CHECK-NEXT:  7      6     2.00                        crc32l	%eax, %ecx
+# CHECK-NEXT:  3      7     2.00    *                   crc32l	(%rax), %ecx
+# CHECK-NEXT:  5      5     2.00                        crc32w	%ax, %ecx
+# CHECK-NEXT:  3      7     2.00    *                   crc32w	(%rax), %ecx
+# CHECK-NEXT:  3      3     2.00                        crc32b	%al, %rcx
+# CHECK-NEXT:  3      7     2.00    *                   crc32b	(%rax), %rcx
+# CHECK-NEXT:  11     10    2.00                        crc32q	%rax, %rcx
+# CHECK-NEXT:  3      7     2.00    *                   crc32q	(%rax), %rcx
+# CHECK-NEXT:  27     15    4.00                        pcmpestri	$1, %xmm0, %xmm2
+# CHECK-NEXT:  28     20    4.50    *                   pcmpestri	$1, (%rax), %xmm2
+# CHECK-NEXT:  27     10    4.00                        pcmpestrm	$1, %xmm0, %xmm2
+# CHECK-NEXT:  28     15    4.50    *                   pcmpestrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  7      14    1.00                        pcmpistri	$1, %xmm0, %xmm2
+# CHECK-NEXT:  8      19    1.00    *                   pcmpistri	$1, (%rax), %xmm2
+# CHECK-NEXT:  7      6     1.00                        pcmpistrm	$1, %xmm0, %xmm2
+# CHECK-NEXT:  9      11    1.00    *                   pcmpistrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtq	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pcmpgtq	(%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     24.00  20.00   -     10.00  5.00   5.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 21.00  21.00   -      -      -     28.00  20.00   -     6.00   6.00   9.00   9.00    -      -     1.00   9.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     crc32b	%al, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   crc32b	(%rax), %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     crc32l	%eax, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   crc32l	(%rax), %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     crc32w	%ax, %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   crc32w	(%rax), %ecx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     crc32b	%al, %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   crc32b	(%rax), %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     crc32q	%rax, %rcx
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   crc32q	(%rax), %rcx
-# CHECK-NEXT:  -      -     2.67   2.67    -     2.67    -      -     pcmpestri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     2.33   2.33    -     2.33   0.50   0.50   pcmpestri	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     2.67   2.67    -     2.67    -      -     pcmpestrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     2.33   2.33    -     2.33   0.50   0.50   pcmpestrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -      -      -     pcmpistri	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -     0.50   0.50   pcmpistri	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -      -      -     pcmpistrm	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -     3.00    -      -      -     0.50   0.50   pcmpistrm	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pcmpgtq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pcmpgtq	(%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32b	%al, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32b	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32l	%eax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32l	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32w	%ax, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32w	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32b	%al, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32b	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32q	%rax, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     crc32q	(%rax), %rcx
+# CHECK-NEXT: 4.00   4.00    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     pcmpestri	$1, %xmm0, %xmm2
+# CHECK-NEXT: 4.50   4.50    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     pcmpestri	$1, (%rax), %xmm2
+# CHECK-NEXT: 4.00   4.00    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     pcmpestrm	$1, %xmm0, %xmm2
+# CHECK-NEXT: 4.50   4.50    -      -      -     1.00    -      -     0.50   0.50   2.00   2.00    -      -      -     1.00    -      -      -      -     pcmpestrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     pcmpistri	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     pcmpistri	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     pcmpistrm	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -     1.00    -      -      -      -     pcmpistrm	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtq	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pcmpgtq	(%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse4a.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse4a.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-sse4a.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 extrq       %xmm0, %xmm2
 extrq       $22, $2, %xmm2
@@ -19,32 +19,44 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.50                        extrq	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     0.50                        extrq	$22, $2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        insertq	%xmm0, %xmm2
-# CHECK-NEXT:  1      1     0.50                        insertq	$22, $22, %xmm0, %xmm2
-# CHECK-NEXT:  1      1     1.00           *            movntsd	%xmm0, (%rax)
-# CHECK-NEXT:  1      1     1.00           *            movntss	%xmm0, (%rax)
+# CHECK-NEXT:  1      3     0.50                        extrq	%xmm0, %xmm2
+# CHECK-NEXT:  1      3     0.50                        extrq	$22, $2, %xmm2
+# CHECK-NEXT:  1      3     2.00                        insertq	%xmm0, %xmm2
+# CHECK-NEXT:  1      3     2.00                        insertq	$22, $22, %xmm0, %xmm2
+# CHECK-NEXT:  1      3     1.00           *            movntsd	%xmm0, (%rax)
+# CHECK-NEXT:  1      3     1.00           *            movntss	%xmm0, (%rax)
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     2.00   2.00   2.00   1.00   1.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -     5.00   5.00    -     2.00   2.00   4.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     extrq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     extrq	$22, $2, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     insertq	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     insertq	$22, $22, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntsd	%xmm0, (%rax)
-# CHECK-NEXT:  -      -      -      -     1.00    -     0.50   0.50   movntss	%xmm0, (%rax)
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     extrq	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     extrq	$22, $2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     insertq	%xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     insertq	$22, $22, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movntsd	%xmm0, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     movntss	%xmm0, (%rax)
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-ssse3.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-ssse3.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-ssse3.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 pabsb       %mm0, %mm2
 pabsb       (%rax), %mm2
@@ -106,148 +106,160 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.50                        pabsb	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pabsb	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        pabsb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pabsb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pabsd	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pabsd	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        pabsd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pabsd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pabsw	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pabsw	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        pabsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pabsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        palignr	$1, %mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   palignr	$1, (%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        palignr	$1, %xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   palignr	$1, (%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        phaddd	%mm0, %mm2
-# CHECK-NEXT:  4      8     1.50    *                   phaddd	(%rax), %mm2
-# CHECK-NEXT:  3      3     1.50                        phaddd	%xmm0, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   phaddd	(%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        phaddsw	%mm0, %mm2
-# CHECK-NEXT:  4      8     1.50    *                   phaddsw	(%rax), %mm2
-# CHECK-NEXT:  3      3     1.50                        phaddsw	%xmm0, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   phaddsw	(%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        phaddw	%mm0, %mm2
-# CHECK-NEXT:  4      8     1.50    *                   phaddw	(%rax), %mm2
-# CHECK-NEXT:  3      3     1.50                        phaddw	%xmm0, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   phaddw	(%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        phsubd	%mm0, %mm2
-# CHECK-NEXT:  4      8     1.50    *                   phsubd	(%rax), %mm2
-# CHECK-NEXT:  3      3     1.50                        phsubd	%xmm0, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   phsubd	(%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        phsubsw	%mm0, %mm2
-# CHECK-NEXT:  4      8     1.50    *                   phsubsw	(%rax), %mm2
-# CHECK-NEXT:  3      3     1.50                        phsubsw	%xmm0, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   phsubsw	(%rax), %xmm2
-# CHECK-NEXT:  3      3     1.50                        phsubw	%mm0, %mm2
-# CHECK-NEXT:  4      8     1.50    *                   phsubw	(%rax), %mm2
-# CHECK-NEXT:  3      3     1.50                        phsubw	%xmm0, %xmm2
-# CHECK-NEXT:  4      9     1.50    *                   phsubw	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmaddubsw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmaddubsw	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmaddubsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmaddubsw	(%rax), %xmm2
-# CHECK-NEXT:  1      5     1.00                        pmulhrsw	%mm0, %mm2
-# CHECK-NEXT:  2      10    1.00    *                   pmulhrsw	(%rax), %mm2
-# CHECK-NEXT:  1      5     1.00                        pmulhrsw	%xmm0, %xmm2
-# CHECK-NEXT:  2      11    1.00    *                   pmulhrsw	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        pshufb	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   pshufb	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        pshufb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   pshufb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psignb	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   psignb	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        psignb	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psignb	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psignd	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   psignd	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        psignd	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psignd	(%rax), %xmm2
-# CHECK-NEXT:  1      1     0.50                        psignw	%mm0, %mm2
-# CHECK-NEXT:  2      6     0.50    *                   psignw	(%rax), %mm2
-# CHECK-NEXT:  1      1     0.50                        psignw	%xmm0, %xmm2
-# CHECK-NEXT:  2      7     0.50    *                   psignw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pabsb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pabsb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pabsb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pabsb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pabsd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pabsd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pabsd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pabsd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        pabsw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   pabsw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        pabsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   pabsw	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        palignr	$1, %mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   palignr	$1, (%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        palignr	$1, %xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   palignr	$1, (%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        phaddd	%mm0, %mm2
+# CHECK-NEXT:  4      10    0.50    *                   phaddd	(%rax), %mm2
+# CHECK-NEXT:  3      5     0.50                        phaddd	%xmm0, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   phaddd	(%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        phaddsw	%mm0, %mm2
+# CHECK-NEXT:  4      10    0.50    *                   phaddsw	(%rax), %mm2
+# CHECK-NEXT:  3      5     0.50                        phaddsw	%xmm0, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   phaddsw	(%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        phaddw	%mm0, %mm2
+# CHECK-NEXT:  4      10    0.50    *                   phaddw	(%rax), %mm2
+# CHECK-NEXT:  3      5     0.50                        phaddw	%xmm0, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   phaddw	(%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        phsubd	%mm0, %mm2
+# CHECK-NEXT:  4      10    0.50    *                   phsubd	(%rax), %mm2
+# CHECK-NEXT:  3      5     0.50                        phsubd	%xmm0, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   phsubd	(%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        phsubsw	%mm0, %mm2
+# CHECK-NEXT:  4      10    0.50    *                   phsubsw	(%rax), %mm2
+# CHECK-NEXT:  3      5     0.50                        phsubsw	%xmm0, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   phsubsw	(%rax), %xmm2
+# CHECK-NEXT:  3      5     0.50                        phsubw	%mm0, %mm2
+# CHECK-NEXT:  4      10    0.50    *                   phsubw	(%rax), %mm2
+# CHECK-NEXT:  3      5     0.50                        phsubw	%xmm0, %xmm2
+# CHECK-NEXT:  4      10    0.50    *                   phsubw	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        pmaddubsw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmaddubsw	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmaddubsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmaddubsw	(%rax), %xmm2
+# CHECK-NEXT:  1      4     1.00                        pmulhrsw	%mm0, %mm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhrsw	(%rax), %mm2
+# CHECK-NEXT:  1      4     1.00                        pmulhrsw	%xmm0, %xmm2
+# CHECK-NEXT:  1      9     1.00    *                   pmulhrsw	(%rax), %xmm2
+# CHECK-NEXT:  1      3     2.00                        pshufb	%mm0, %mm2
+# CHECK-NEXT:  1      8     2.00    *                   pshufb	(%rax), %mm2
+# CHECK-NEXT:  1      3     2.00                        pshufb	%xmm0, %xmm2
+# CHECK-NEXT:  1      8     2.00    *                   pshufb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psignb	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psignb	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psignb	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psignb	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psignd	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psignd	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psignd	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psignd	(%rax), %xmm2
+# CHECK-NEXT:  1      2     0.50                        psignw	%mm0, %mm2
+# CHECK-NEXT:  1      7     0.50    *                   psignw	(%rax), %mm2
+# CHECK-NEXT:  1      2     0.50                        psignw	%xmm0, %xmm2
+# CHECK-NEXT:  1      7     0.50    *                   psignw	(%rax), %xmm2
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     8.00   52.00   -     52.00  16.00  16.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 16.00  16.00   -      -      -      -      -      -      -      -     34.00  34.00  8.00    -     36.00  28.00   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pabsb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pabsb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pabsb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pabsb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pabsd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pabsd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pabsd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pabsd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pabsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pabsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pabsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pabsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     palignr	$1, %mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   palignr	$1, (%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     palignr	$1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   palignr	$1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phaddd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phaddd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phaddd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phaddd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phaddsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phaddsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phaddsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phaddsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phaddw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phaddw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phaddw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phaddw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phsubd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phsubd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phsubd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phsubd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phsubsw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phsubsw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phsubsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phsubsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phsubw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phsubw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     phsubw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   phsubw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmaddubsw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmaddubsw	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmaddubsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmaddubsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhrsw	%mm0, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhrsw	(%rax), %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pmulhrsw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   pmulhrsw	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pshufb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pshufb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     pshufb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   pshufb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psignb	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psignb	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psignb	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psignb	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psignd	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psignd	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psignd	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psignd	(%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psignw	%mm0, %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psignw	(%rax), %mm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     psignw	%xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   psignw	(%rax), %xmm2
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     pabsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     palignr	$1, %mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     palignr	$1, (%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     palignr	$1, %xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     palignr	$1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phaddw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     phsubw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddubsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddubsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddubsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmaddubsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhrsw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhrsw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhrsw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     pmulhrsw	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     pshufb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     pshufb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     pshufb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     pshufb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignb	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignb	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignb	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignb	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignd	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignd	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignd	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignd	(%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignw	%mm0, %mm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignw	(%rax), %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignw	%xmm0, %xmm2
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     psignw	(%rax), %xmm2
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-tbm.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-tbm.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-tbm.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 bextr        $8192, %ebx, %ecx
 bextr        $8192, (%rbx), %ecx
@@ -70,100 +70,112 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      2     1.00                        bextrl	$8192, %ebx, %ecx
-# CHECK-NEXT:  3      7     1.00    *                   bextrl	$8192, (%rbx), %ecx
-# CHECK-NEXT:  2      2     1.00                        bextrq	$16384, %rbx, %rcx
-# CHECK-NEXT:  3      7     1.00    *                   bextrq	$16384, (%rbx), %rcx
-# CHECK-NEXT:  1      1     0.33                        blcfilll	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        bextrl	$8192, %ebx, %ecx
+# CHECK-NEXT:  2      6     0.50    *                   bextrl	$8192, (%rbx), %ecx
+# CHECK-NEXT:  2      2     0.50                        bextrq	$16384, %rbx, %rcx
+# CHECK-NEXT:  2      6     0.50    *                   bextrq	$16384, (%rbx), %rcx
+# CHECK-NEXT:  2      2     0.50                        blcfilll	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blcfilll	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blcfillq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blcfillq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blcfillq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blcil	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blcil	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blcil	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blciq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blciq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blciq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blcicl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blcicl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blcicl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blcicq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blcicq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blcicq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blcmskl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blcmskl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blcmskl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blcmskq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blcmskq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blcmskq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blcsl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blcsl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blcsl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blcsq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blcsq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blcsq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blsfilll	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blsfilll	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blsfilll	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blsfillq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blsfillq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blsfillq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        blsicl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        blsicl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   blsicl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        blsicq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        blsicq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   blsicq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        t1mskcl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        t1mskcl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   t1mskcl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        t1mskcq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        t1mskcq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   t1mskcq	(%rax), %rcx
-# CHECK-NEXT:  1      1     0.33                        tzmskl	%eax, %ecx
+# CHECK-NEXT:  2      2     0.50                        tzmskl	%eax, %ecx
 # CHECK-NEXT:  2      6     0.50    *                   tzmskl	(%rax), %ecx
-# CHECK-NEXT:  1      1     0.33                        tzmskq	%rax, %rcx
+# CHECK-NEXT:  2      2     0.50                        tzmskq	%rax, %rcx
 # CHECK-NEXT:  2      6     0.50    *                   tzmskq	(%rax), %rcx
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     14.00  16.00   -     14.00  10.00  10.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 1.00   1.00    -      -      -     20.00  20.00   -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50    -      -     bextrl	$8192, %ebx, %ecx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50   0.50   0.50   bextrl	$8192, (%rbx), %ecx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50    -      -     bextrq	$16384, %rbx, %rcx
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50   0.50   0.50   bextrq	$16384, (%rbx), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcfilll	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcfilll	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcfillq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcfillq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcil	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcil	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blciq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blciq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcicl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcicl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcicq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcicq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcmskl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcmskl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcmskq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcmskq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcsl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcsl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blcsq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blcsq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsfilll	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsfilll	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsfillq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsfillq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsicl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsicl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     blsicq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   blsicq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     t1mskcl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   t1mskcl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     t1mskcq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   t1mskcq	(%rax), %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     tzmskl	%eax, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   tzmskl	(%rax), %ecx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     tzmskq	%rax, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   tzmskq	(%rax), %rcx
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrl	$8192, %ebx, %ecx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrl	$8192, (%rbx), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrq	$16384, %rbx, %rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bextrq	$16384, (%rbx), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcfilll	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcfilll	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcfillq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcfillq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcil	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcil	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blciq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blciq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcicl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcicl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcicq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcicq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcmskl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcmskl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcmskq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcmskq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcsl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcsl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcsq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blcsq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsfilll	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsfilll	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsfillq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsfillq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsicl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsicl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsicq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     blsicq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     t1mskcl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     t1mskcl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     t1mskcq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     t1mskcq	(%rax), %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     tzmskl	%eax, %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     tzmskl	(%rax), %ecx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     tzmskq	%rax, %rcx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     tzmskq	(%rax), %rcx
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x86_32.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x86_32.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x86_32.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=i686-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=i686-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 aaa
 
@@ -33,46 +33,58 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      100   0.33                        aaa
-# CHECK-NEXT:  1      100   0.33                        aad
-# CHECK-NEXT:  1      100   0.33                        aad	$7
-# CHECK-NEXT:  1      100   0.33                        aam
-# CHECK-NEXT:  1      100   0.33                        aam	$7
-# CHECK-NEXT:  1      100   0.33                        aas
-# CHECK-NEXT:  1      100   0.33                  U     bound	%bx, (%eax)
-# CHECK-NEXT:  1      100   0.33                  U     bound	%ebx, (%eax)
-# CHECK-NEXT:  1      100   0.33                        daa
-# CHECK-NEXT:  1      100   0.33                        das
-# CHECK-NEXT:  1      100   0.33                  U     into
-# CHECK-NEXT:  3      7     0.67    *                   leave
-# CHECK-NEXT:  1      1     0.33                  U     salc
+# CHECK-NEXT:  1      100   0.50                        aaa
+# CHECK-NEXT:  1      100   0.50                        aad
+# CHECK-NEXT:  1      100   0.50                        aad	$7
+# CHECK-NEXT:  1      100   0.50                        aam
+# CHECK-NEXT:  1      100   0.50                        aam	$7
+# CHECK-NEXT:  1      100   0.50                        aas
+# CHECK-NEXT:  1      100   0.50                  U     bound	%bx, (%eax)
+# CHECK-NEXT:  1      100   0.50                  U     bound	%ebx, (%eax)
+# CHECK-NEXT:  1      100   0.50                        daa
+# CHECK-NEXT:  1      100   0.50                        das
+# CHECK-NEXT:  1      100   0.50                  U     into
+# CHECK-NEXT:  1      1     0.50    *                   leave
+# CHECK-NEXT:  1      1     0.50                  U     salc
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     4.67   4.67    -     4.67   0.50   0.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     6.50   6.50    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     aaa
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     aad
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     aad	$7
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     aam
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     aam	$7
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     aas
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     bound	%bx, (%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     bound	%ebx, (%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     daa
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     das
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     into
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67   0.50   0.50   leave
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     salc
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     aaa
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     aad
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     aad	$7
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     aam
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     aam	$7
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     aas
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bound	%bx, (%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     bound	%ebx, (%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     daa
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     das
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     into
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leave
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     salc
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x86_64.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x86_64.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x86_64.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 adcb $7, %al
 adcb $7, %dil
@@ -861,1512 +861,1524 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      2     0.67                        adcb	$7, %al
-# CHECK-NEXT:  2      2     0.67                        adcb	$7, %dil
-# CHECK-NEXT:  6      9     1.00    *      *            adcb	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcb	%sil, %dil
-# CHECK-NEXT:  6      9     1.00    *      *            adcb	%sil, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   adcb	(%rax), %dil
-# CHECK-NEXT:  2      2     0.67                        adcw	$511, %ax
-# CHECK-NEXT:  2      2     0.67                        adcw	$511, %di
-# CHECK-NEXT:  6      9     1.00    *      *            adcw	$511, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcw	$7, %di
-# CHECK-NEXT:  6      9     1.00    *      *            adcw	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcw	%si, %di
-# CHECK-NEXT:  6      9     1.00    *      *            adcw	%si, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   adcw	(%rax), %di
-# CHECK-NEXT:  2      2     0.67                        adcl	$665536, %eax
-# CHECK-NEXT:  2      2     0.67                        adcl	$665536, %edi
-# CHECK-NEXT:  6      9     1.00    *      *            adcl	$665536, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcl	$7, %edi
-# CHECK-NEXT:  6      9     1.00    *      *            adcl	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcl	%esi, %edi
-# CHECK-NEXT:  6      9     1.00    *      *            adcl	%esi, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   adcl	(%rax), %edi
-# CHECK-NEXT:  2      2     0.67                        adcq	$665536, %rax
-# CHECK-NEXT:  2      2     0.67                        adcq	$665536, %rdi
-# CHECK-NEXT:  6      9     1.00    *      *            adcq	$665536, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcq	$7, %rdi
-# CHECK-NEXT:  6      9     1.00    *      *            adcq	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        adcq	%rsi, %rdi
-# CHECK-NEXT:  6      9     1.00    *      *            adcq	%rsi, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   adcq	(%rax), %rdi
-# CHECK-NEXT:  1      1     0.33                        addb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        addb	$7, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            addb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addb	%sil, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            addb	%sil, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   addb	(%rax), %dil
-# CHECK-NEXT:  1      1     0.33                        addw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        addw	$511, %di
-# CHECK-NEXT:  3      7     1.00    *      *            addw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addw	$7, %di
-# CHECK-NEXT:  3      7     1.00    *      *            addw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addw	%si, %di
-# CHECK-NEXT:  3      7     1.00    *      *            addw	%si, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   addw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        addl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        addl	$665536, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            addl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addl	$7, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            addl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addl	%esi, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            addl	%esi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   addl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        addq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        addq	$665536, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            addq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addq	$7, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            addq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        addq	%rsi, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            addq	%rsi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   addq	(%rax), %rdi
-# CHECK-NEXT:  1      1     0.33                        andb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        andb	$7, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            andb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andb	%sil, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            andb	%sil, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   andb	(%rax), %dil
-# CHECK-NEXT:  1      1     0.33                        andw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        andw	$511, %di
-# CHECK-NEXT:  3      7     1.00    *      *            andw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andw	$7, %di
-# CHECK-NEXT:  3      7     1.00    *      *            andw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andw	%si, %di
-# CHECK-NEXT:  3      7     1.00    *      *            andw	%si, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   andw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        andl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        andl	$665536, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            andl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andl	$7, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            andl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andl	%esi, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            andl	%esi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   andl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        andq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        andq	$665536, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            andq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andq	$7, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            andq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        andq	%rsi, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            andq	%rsi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   andq	(%rax), %rdi
-# CHECK-NEXT:  1      3     1.00                        bsfw	%si, %di
-# CHECK-NEXT:  1      3     1.00                        bsrw	%si, %di
-# CHECK-NEXT:  2      8     1.00    *                   bsfw	(%rax), %di
-# CHECK-NEXT:  2      8     1.00    *                   bsrw	(%rax), %di
-# CHECK-NEXT:  1      3     1.00                        bsfl	%esi, %edi
-# CHECK-NEXT:  1      3     1.00                        bsrl	%esi, %edi
-# CHECK-NEXT:  2      8     1.00    *                   bsfl	(%rax), %edi
-# CHECK-NEXT:  2      8     1.00    *                   bsrl	(%rax), %edi
-# CHECK-NEXT:  1      3     1.00                        bsfq	%rsi, %rdi
-# CHECK-NEXT:  1      3     1.00                        bsrq	%rsi, %rdi
-# CHECK-NEXT:  2      8     1.00    *                   bsfq	(%rax), %rdi
-# CHECK-NEXT:  2      8     1.00    *                   bsrq	(%rax), %rdi
+# CHECK-NEXT:  1      1     1.00                        adcb	$7, %al
+# CHECK-NEXT:  1      1     1.00                        adcb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            adcb	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            adcb	%sil, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   adcb	(%rax), %dil
+# CHECK-NEXT:  1      1     1.00                        adcw	$511, %ax
+# CHECK-NEXT:  1      1     1.00                        adcw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            adcw	$511, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            adcw	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            adcw	%si, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   adcw	(%rax), %di
+# CHECK-NEXT:  1      1     1.00                        adcl	$665536, %eax
+# CHECK-NEXT:  1      1     1.00                        adcl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            adcl	$665536, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            adcl	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            adcl	%esi, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   adcl	(%rax), %edi
+# CHECK-NEXT:  1      1     1.00                        adcq	$665536, %rax
+# CHECK-NEXT:  1      1     1.00                        adcq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            adcq	$665536, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            adcq	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        adcq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            adcq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   adcq	(%rax), %rdi
+# CHECK-NEXT:  1      1     0.50                        addb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        addb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            addb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            addb	%sil, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   addb	(%rax), %dil
+# CHECK-NEXT:  1      1     0.50                        addw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        addw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            addw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            addw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            addw	%si, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   addw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        addl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            addl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            addl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            addl	%esi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   addl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        addq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        addq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            addq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            addq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        addq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            addq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   addq	(%rax), %rdi
+# CHECK-NEXT:  1      1     0.50                        andb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        andb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            andb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            andb	%sil, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   andb	(%rax), %dil
+# CHECK-NEXT:  1      1     0.50                        andw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        andw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            andw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            andw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            andw	%si, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   andw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        andl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        andl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            andl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            andl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            andl	%esi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   andl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        andq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        andq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            andq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            andq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        andq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            andq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   andq	(%rax), %rdi
+# CHECK-NEXT:  6      3     2.00                        bsfw	%si, %di
+# CHECK-NEXT:  7      4     2.00                        bsrw	%si, %di
+# CHECK-NEXT:  8      7     2.00    *                   bsfw	(%rax), %di
+# CHECK-NEXT:  9      8     2.00    *                   bsrw	(%rax), %di
+# CHECK-NEXT:  6      3     2.00                        bsfl	%esi, %edi
+# CHECK-NEXT:  7      4     2.00                        bsrl	%esi, %edi
+# CHECK-NEXT:  8      7     2.00    *                   bsfl	(%rax), %edi
+# CHECK-NEXT:  9      8     2.00    *                   bsrl	(%rax), %edi
+# CHECK-NEXT:  6      3     2.00                        bsfq	%rsi, %rdi
+# CHECK-NEXT:  7      4     2.00                        bsrq	%rsi, %rdi
+# CHECK-NEXT:  8      7     2.00    *                   bsfq	(%rax), %rdi
+# CHECK-NEXT:  9      8     2.00    *                   bsrq	(%rax), %rdi
 # CHECK-NEXT:  1      1     1.00                        bswapl	%eax
-# CHECK-NEXT:  2      2     1.00                        bswapq	%rax
+# CHECK-NEXT:  1      1     1.00                        bswapq	%rax
 # CHECK-NEXT:  1      1     0.50                        btw	%si, %di
-# CHECK-NEXT:  1      1     0.50                        btcw	%si, %di
-# CHECK-NEXT:  1      1     0.50                        btrw	%si, %di
-# CHECK-NEXT:  1      1     0.50                        btsw	%si, %di
-# CHECK-NEXT:  6      9     1.00    *                   btw	%si, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btcw	%si, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btrw	%si, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btsw	%si, (%rax)
+# CHECK-NEXT:  2      2     0.50                        btcw	%si, %di
+# CHECK-NEXT:  2      2     0.50                        btrw	%si, %di
+# CHECK-NEXT:  2      2     0.50                        btsw	%si, %di
+# CHECK-NEXT:  7      5     0.50    *                   btw	%si, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btcw	%si, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btrw	%si, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btsw	%si, (%rax)
 # CHECK-NEXT:  1      1     0.50                        btw	$7, %di
-# CHECK-NEXT:  1      1     0.50                        btcw	$7, %di
-# CHECK-NEXT:  1      1     0.50                        btrw	$7, %di
-# CHECK-NEXT:  1      1     0.50                        btsw	$7, %di
-# CHECK-NEXT:  2      6     0.50    *                   btw	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btcw	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btrw	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btsw	$7, (%rax)
+# CHECK-NEXT:  2      2     0.50                        btcw	$7, %di
+# CHECK-NEXT:  2      2     0.50                        btrw	$7, %di
+# CHECK-NEXT:  2      2     0.50                        btsw	$7, %di
+# CHECK-NEXT:  1      5     0.50    *                   btw	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btcw	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btrw	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btsw	$7, (%rax)
 # CHECK-NEXT:  1      1     0.50                        btl	%esi, %edi
-# CHECK-NEXT:  1      1     0.50                        btcl	%esi, %edi
-# CHECK-NEXT:  1      1     0.50                        btrl	%esi, %edi
-# CHECK-NEXT:  1      1     0.50                        btsl	%esi, %edi
-# CHECK-NEXT:  6      9     1.00    *                   btl	%esi, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btcl	%esi, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btrl	%esi, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btsl	%esi, (%rax)
+# CHECK-NEXT:  2      2     0.50                        btcl	%esi, %edi
+# CHECK-NEXT:  2      2     0.50                        btrl	%esi, %edi
+# CHECK-NEXT:  2      2     0.50                        btsl	%esi, %edi
+# CHECK-NEXT:  7      5     0.50    *                   btl	%esi, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btcl	%esi, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btrl	%esi, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btsl	%esi, (%rax)
 # CHECK-NEXT:  1      1     0.50                        btl	$7, %edi
-# CHECK-NEXT:  1      1     0.50                        btcl	$7, %edi
-# CHECK-NEXT:  1      1     0.50                        btrl	$7, %edi
-# CHECK-NEXT:  1      1     0.50                        btsl	$7, %edi
-# CHECK-NEXT:  2      6     0.50    *                   btl	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btcl	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btrl	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btsl	$7, (%rax)
+# CHECK-NEXT:  2      2     0.50                        btcl	$7, %edi
+# CHECK-NEXT:  2      2     0.50                        btrl	$7, %edi
+# CHECK-NEXT:  2      2     0.50                        btsl	$7, %edi
+# CHECK-NEXT:  1      5     0.50    *                   btl	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btcl	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btrl	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btsl	$7, (%rax)
 # CHECK-NEXT:  1      1     0.50                        btq	%rsi, %rdi
-# CHECK-NEXT:  1      1     0.50                        btcq	%rsi, %rdi
-# CHECK-NEXT:  1      1     0.50                        btrq	%rsi, %rdi
-# CHECK-NEXT:  1      1     0.50                        btsq	%rsi, %rdi
-# CHECK-NEXT:  6      9     1.00    *                   btq	%rsi, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btcq	%rsi, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btrq	%rsi, (%rax)
-# CHECK-NEXT:  6      9     1.00    *      *            btsq	%rsi, (%rax)
+# CHECK-NEXT:  2      2     0.50                        btcq	%rsi, %rdi
+# CHECK-NEXT:  2      2     0.50                        btrq	%rsi, %rdi
+# CHECK-NEXT:  2      2     0.50                        btsq	%rsi, %rdi
+# CHECK-NEXT:  7      5     0.50    *                   btq	%rsi, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btcq	%rsi, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btrq	%rsi, (%rax)
+# CHECK-NEXT:  11     7     1.00    *      *            btsq	%rsi, (%rax)
 # CHECK-NEXT:  1      1     0.50                        btq	$7, %rdi
-# CHECK-NEXT:  1      1     0.50                        btcq	$7, %rdi
-# CHECK-NEXT:  1      1     0.50                        btrq	$7, %rdi
-# CHECK-NEXT:  1      1     0.50                        btsq	$7, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   btq	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btcq	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btrq	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            btsq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cbtw
-# CHECK-NEXT:  1      1     0.33                        cwtl
-# CHECK-NEXT:  1      1     0.33                        cltq
-# CHECK-NEXT:  2      2     1.00                        cwtd
+# CHECK-NEXT:  2      2     0.50                        btcq	$7, %rdi
+# CHECK-NEXT:  2      2     0.50                        btrq	$7, %rdi
+# CHECK-NEXT:  2      2     0.50                        btsq	$7, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   btq	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btcq	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btrq	$7, (%rax)
+# CHECK-NEXT:  5      7     1.00    *      *            btsq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cbtw
+# CHECK-NEXT:  1      1     0.50                        cwtl
+# CHECK-NEXT:  1      1     0.50                        cltq
+# CHECK-NEXT:  1      1     0.50                        cwtd
 # CHECK-NEXT:  1      1     0.50                        cltd
 # CHECK-NEXT:  1      1     0.50                        cqto
-# CHECK-NEXT:  1      1     0.25                  U     clc
-# CHECK-NEXT:  1      1     0.33                  U     cld
-# CHECK-NEXT:  1      1     0.33                  U     cmc
-# CHECK-NEXT:  1      1     0.33                        cmpb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        cmpb	$7, %dil
-# CHECK-NEXT:  2      6     0.50    *                   cmpb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpb	%sil, %dil
-# CHECK-NEXT:  2      6     0.50    *                   cmpb	%sil, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   cmpb	(%rax), %dil
-# CHECK-NEXT:  1      1     0.33                        cmpw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        cmpw	$511, %di
-# CHECK-NEXT:  2      6     0.50    *                   cmpw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpw	$7, %di
-# CHECK-NEXT:  2      6     0.50    *                   cmpw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpw	%si, %di
-# CHECK-NEXT:  2      6     0.50    *                   cmpw	%si, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   cmpw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        cmpl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        cmpl	$665536, %edi
-# CHECK-NEXT:  2      6     0.50    *                   cmpl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpl	$7, %edi
-# CHECK-NEXT:  2      6     0.50    *                   cmpl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpl	%esi, %edi
-# CHECK-NEXT:  2      6     0.50    *                   cmpl	%esi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   cmpl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        cmpq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        cmpq	$665536, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   cmpq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpq	$7, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   cmpq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        cmpq	%rsi, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   cmpq	%rsi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   cmpq	(%rax), %rdi
-# CHECK-NEXT:  5      8     1.00                  U     cmpsb	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  5      8     1.00                  U     cmpsw	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  5      8     1.00                  U     cmpsl	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  5      8     1.00                  U     cmpsq	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  4      5     1.33                        cmpxchgb	%cl, %bl
-# CHECK-NEXT:  6      8     2.00    *      *            cmpxchgb	%cl, (%rbx)
-# CHECK-NEXT:  4      5     1.33                        cmpxchgw	%cx, %bx
-# CHECK-NEXT:  6      8     2.00    *      *            cmpxchgw	%cx, (%rbx)
-# CHECK-NEXT:  4      5     1.33                        cmpxchgl	%ecx, %ebx
-# CHECK-NEXT:  6      8     2.00    *      *            cmpxchgl	%ecx, (%rbx)
-# CHECK-NEXT:  4      5     1.33                        cmpxchgq	%rcx, %rbx
-# CHECK-NEXT:  6      8     2.00    *      *            cmpxchgq	%rcx, (%rbx)
-# CHECK-NEXT:  1      100   0.33                  U     cpuid
-# CHECK-NEXT:  1      1     0.33                        decb	%dil
-# CHECK-NEXT:  3      7     1.00    *      *            decb	(%rax)
-# CHECK-NEXT:  1      1     0.33                        decw	%di
-# CHECK-NEXT:  3      7     1.00    *      *            decw	(%rax)
-# CHECK-NEXT:  1      1     0.33                        decl	%edi
-# CHECK-NEXT:  3      7     1.00    *      *            decl	(%rax)
-# CHECK-NEXT:  1      1     0.33                        decq	%rdi
-# CHECK-NEXT:  3      7     1.00    *      *            decq	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     divb	%dil
-# CHECK-NEXT:  2      30    10.00   *             U     divb	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     divw	%si
-# CHECK-NEXT:  2      30    10.00   *             U     divw	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     divl	%edx
-# CHECK-NEXT:  2      30    10.00   *             U     divl	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     divq	%rcx
-# CHECK-NEXT:  2      30    10.00   *             U     divq	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     idivb	%dil
-# CHECK-NEXT:  2      30    10.00   *             U     idivb	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     idivw	%si
-# CHECK-NEXT:  2      30    10.00   *             U     idivw	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     idivl	%edx
-# CHECK-NEXT:  2      30    10.00   *             U     idivl	(%rax)
-# CHECK-NEXT:  1      25    10.00                 U     idivq	%rcx
-# CHECK-NEXT:  2      30    10.00   *             U     idivq	(%rax)
-# CHECK-NEXT:  1      3     1.00                        imulb	%dil
-# CHECK-NEXT:  2      8     1.00    *                   imulb	(%rax)
-# CHECK-NEXT:  4      4     1.33                        imulw	%di
-# CHECK-NEXT:  5      9     1.33    *                   imulw	(%rax)
-# CHECK-NEXT:  1      3     1.00                        imulw	%si, %di
-# CHECK-NEXT:  2      8     1.00    *                   imulw	(%rax), %di
-# CHECK-NEXT:  2      4     1.00                        imulw	$511, %si, %di
-# CHECK-NEXT:  3      8     1.00    *                   imulw	$511, (%rax), %di
-# CHECK-NEXT:  2      4     1.00                        imulw	$7, %si, %di
-# CHECK-NEXT:  3      8     1.00    *                   imulw	$7, (%rax), %di
-# CHECK-NEXT:  3      4     1.00                        imull	%edi
-# CHECK-NEXT:  4      9     1.00    *                   imull	(%rax)
-# CHECK-NEXT:  1      3     1.00                        imull	%esi, %edi
-# CHECK-NEXT:  2      8     1.00    *                   imull	(%rax), %edi
-# CHECK-NEXT:  1      3     1.00                        imull	$665536, %esi, %edi
+# CHECK-NEXT:  1      1     0.50                  U     clc
+# CHECK-NEXT:  1      1     0.50                  U     cld
+# CHECK-NEXT:  1      1     0.50                  U     cmc
+# CHECK-NEXT:  1      1     0.50                        cmpb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        cmpb	$7, %dil
+# CHECK-NEXT:  1      5     0.50    *                   cmpb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpb	%sil, %dil
+# CHECK-NEXT:  1      5     0.50    *                   cmpb	%sil, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   cmpb	(%rax), %dil
+# CHECK-NEXT:  1      1     0.50                        cmpw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        cmpw	$511, %di
+# CHECK-NEXT:  1      5     0.50    *                   cmpw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpw	$7, %di
+# CHECK-NEXT:  1      5     0.50    *                   cmpw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpw	%si, %di
+# CHECK-NEXT:  1      5     0.50    *                   cmpw	%si, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   cmpw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        cmpl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        cmpl	$665536, %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmpl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpl	$7, %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmpl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpl	%esi, %edi
+# CHECK-NEXT:  1      5     0.50    *                   cmpl	%esi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   cmpl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        cmpq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        cmpq	$665536, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmpq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpq	$7, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmpq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        cmpq	%rsi, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   cmpq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   cmpq	(%rax), %rdi
+# CHECK-NEXT:  1      100   0.50                  U     cmpsb	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  1      100   0.50                  U     cmpsw	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  1      100   0.50                  U     cmpsl	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  1      100   0.50                  U     cmpsq	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  3      3     1.00                        cmpxchgb	%cl, %bl
+# CHECK-NEXT:  5      3     1.00    *      *            cmpxchgb	%cl, (%rbx)
+# CHECK-NEXT:  5      3     1.00                        cmpxchgw	%cx, %bx
+# CHECK-NEXT:  6      3     1.00    *      *            cmpxchgw	%cx, (%rbx)
+# CHECK-NEXT:  5      3     1.00                        cmpxchgl	%ecx, %ebx
+# CHECK-NEXT:  6      3     1.00    *      *            cmpxchgl	%ecx, (%rbx)
+# CHECK-NEXT:  5      3     1.00                        cmpxchgq	%rcx, %rbx
+# CHECK-NEXT:  6      3     1.00    *      *            cmpxchgq	%rcx, (%rbx)
+# CHECK-NEXT:  1      100   0.50                  U     cpuid
+# CHECK-NEXT:  1      1     0.50                        decb	%dil
+# CHECK-NEXT:  2      6     1.00    *      *            decb	(%rax)
+# CHECK-NEXT:  1      1     0.50                        decw	%di
+# CHECK-NEXT:  2      6     1.00    *      *            decw	(%rax)
+# CHECK-NEXT:  1      1     0.50                        decl	%edi
+# CHECK-NEXT:  2      6     1.00    *      *            decl	(%rax)
+# CHECK-NEXT:  1      1     0.50                        decq	%rdi
+# CHECK-NEXT:  2      6     1.00    *      *            decq	(%rax)
+# CHECK-NEXT:  1      12    12.00                 U     divb	%dil
+# CHECK-NEXT:  1      16    12.00   *             U     divb	(%rax)
+# CHECK-NEXT:  2      15    15.00                 U     divw	%si
+# CHECK-NEXT:  2      19    15.00   *             U     divw	(%rax)
+# CHECK-NEXT:  2      14    14.00                 U     divl	%edx
+# CHECK-NEXT:  2      18    14.00   *             U     divl	(%rax)
+# CHECK-NEXT:  2      14    14.00                 U     divq	%rcx
+# CHECK-NEXT:  2      18    14.00   *             U     divq	(%rax)
+# CHECK-NEXT:  1      12    12.00                 U     idivb	%dil
+# CHECK-NEXT:  1      16    12.00   *             U     idivb	(%rax)
+# CHECK-NEXT:  2      15    17.00                 U     idivw	%si
+# CHECK-NEXT:  2      19    17.00   *             U     idivw	(%rax)
+# CHECK-NEXT:  2      14    25.00                 U     idivl	%edx
+# CHECK-NEXT:  2      18    25.00   *             U     idivl	(%rax)
+# CHECK-NEXT:  2      14    14.00                 U     idivq	%rcx
+# CHECK-NEXT:  2      18    14.00   *             U     idivq	(%rax)
+# CHECK-NEXT:  1      4     1.00                        imulb	%dil
+# CHECK-NEXT:  1      8     1.00    *                   imulb	(%rax)
+# CHECK-NEXT:  2      4     1.00                        imulw	%di
+# CHECK-NEXT:  2      8     1.00    *                   imulw	(%rax)
+# CHECK-NEXT:  1      4     1.00                        imulw	%si, %di
+# CHECK-NEXT:  1      8     1.00    *                   imulw	(%rax), %di
+# CHECK-NEXT:  2      5     1.00                        imulw	$511, %si, %di
+# CHECK-NEXT:  2      9     1.00    *                   imulw	$511, (%rax), %di
+# CHECK-NEXT:  2      5     1.00                        imulw	$7, %si, %di
+# CHECK-NEXT:  2      9     1.00    *                   imulw	$7, (%rax), %di
+# CHECK-NEXT:  1      4     1.00                        imull	%edi
+# CHECK-NEXT:  1      8     1.00    *                   imull	(%rax)
+# CHECK-NEXT:  1      4     1.00                        imull	%esi, %edi
+# CHECK-NEXT:  1      8     1.00    *                   imull	(%rax), %edi
+# CHECK-NEXT:  1      4     1.00                        imull	$665536, %esi, %edi
 # CHECK-NEXT:  2      8     1.00    *                   imull	$665536, (%rax), %edi
-# CHECK-NEXT:  1      3     1.00                        imull	$7, %esi, %edi
+# CHECK-NEXT:  1      4     1.00                        imull	$7, %esi, %edi
 # CHECK-NEXT:  2      8     1.00    *                   imull	$7, (%rax), %edi
-# CHECK-NEXT:  2      4     1.00                        imulq	%rdi
-# CHECK-NEXT:  3      9     1.00    *                   imulq	(%rax)
-# CHECK-NEXT:  1      3     1.00                        imulq	%rsi, %rdi
-# CHECK-NEXT:  2      8     1.00    *                   imulq	(%rax), %rdi
-# CHECK-NEXT:  1      3     1.00                        imulq	$665536, %rsi, %rdi
-# CHECK-NEXT:  2      8     1.00    *                   imulq	$665536, (%rax), %rdi
-# CHECK-NEXT:  1      3     1.00                        imulq	$7, %rsi, %rdi
-# CHECK-NEXT:  2      8     1.00    *                   imulq	$7, (%rax), %rdi
-# CHECK-NEXT:  1      100   0.33                  U     inb	$7, %al
-# CHECK-NEXT:  1      100   0.33                  U     inb	%dx, %al
-# CHECK-NEXT:  1      100   0.33                  U     inw	$7, %ax
-# CHECK-NEXT:  1      100   0.33                  U     inw	%dx, %ax
-# CHECK-NEXT:  1      100   0.33                  U     inl	$7, %eax
-# CHECK-NEXT:  1      100   0.33                  U     inl	%dx, %eax
-# CHECK-NEXT:  1      1     0.33                        incb	%dil
-# CHECK-NEXT:  3      7     1.00    *      *            incb	(%rax)
-# CHECK-NEXT:  1      1     0.33                        incw	%di
-# CHECK-NEXT:  3      7     1.00    *      *            incw	(%rax)
-# CHECK-NEXT:  1      1     0.33                        incl	%edi
-# CHECK-NEXT:  3      7     1.00    *      *            incl	(%rax)
-# CHECK-NEXT:  1      1     0.33                        incq	%rdi
-# CHECK-NEXT:  3      7     1.00    *      *            incq	(%rax)
-# CHECK-NEXT:  1      100   0.33                  U     insb	%dx, %es:(%rdi)
-# CHECK-NEXT:  1      100   0.33                  U     insw	%dx, %es:(%rdi)
-# CHECK-NEXT:  1      100   0.33                  U     insl	%dx, %es:(%rdi)
-# CHECK-NEXT:  1      100   0.33    *      *      U     int	$7
-# CHECK-NEXT:  1      1     0.50                        lahf
-# CHECK-NEXT:  3      7     0.67                  U     lodsb	(%rsi), %al
-# CHECK-NEXT:  3      7     0.67                  U     lodsw	(%rsi), %ax
-# CHECK-NEXT:  2      6     0.50                  U     lodsl	(%rsi), %eax
-# CHECK-NEXT:  2      6     0.50                  U     lodsq	(%rsi), %rax
-# CHECK-NEXT:  5      8     1.00                  U     movsb	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  5      8     1.00                  U     movsw	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  5      8     1.00                  U     movsl	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  5      8     1.00                  U     movsq	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  1      1     0.33                        movsbw	%al, %di
-# CHECK-NEXT:  1      1     0.33                        movzbw	%al, %di
+# CHECK-NEXT:  1      6     4.00                        imulq	%rdi
+# CHECK-NEXT:  1      10    4.00    *                   imulq	(%rax)
+# CHECK-NEXT:  1      6     4.00                        imulq	%rsi, %rdi
+# CHECK-NEXT:  1      10    4.00    *                   imulq	(%rax), %rdi
+# CHECK-NEXT:  1      6     4.00                        imulq	$665536, %rsi, %rdi
+# CHECK-NEXT:  2      10    4.00    *                   imulq	$665536, (%rax), %rdi
+# CHECK-NEXT:  1      6     4.00                        imulq	$7, %rsi, %rdi
+# CHECK-NEXT:  2      10    4.00    *                   imulq	$7, (%rax), %rdi
+# CHECK-NEXT:  1      100   0.50                  U     inb	$7, %al
+# CHECK-NEXT:  1      100   0.50                  U     inb	%dx, %al
+# CHECK-NEXT:  1      100   0.50                  U     inw	$7, %ax
+# CHECK-NEXT:  1      100   0.50                  U     inw	%dx, %ax
+# CHECK-NEXT:  1      100   0.50                  U     inl	$7, %eax
+# CHECK-NEXT:  1      100   0.50                  U     inl	%dx, %eax
+# CHECK-NEXT:  1      1     0.50                        incb	%dil
+# CHECK-NEXT:  2      6     1.00    *      *            incb	(%rax)
+# CHECK-NEXT:  1      1     0.50                        incw	%di
+# CHECK-NEXT:  2      6     1.00    *      *            incw	(%rax)
+# CHECK-NEXT:  1      1     0.50                        incl	%edi
+# CHECK-NEXT:  2      6     1.00    *      *            incl	(%rax)
+# CHECK-NEXT:  1      1     0.50                        incq	%rdi
+# CHECK-NEXT:  2      6     1.00    *      *            incq	(%rax)
+# CHECK-NEXT:  1      100   0.50                  U     insb	%dx, %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     insw	%dx, %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     insl	%dx, %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50    *      *      U     int	$7
+# CHECK-NEXT:  4      2     0.50                        lahf
+# CHECK-NEXT:  1      100   0.50                  U     lodsb	(%rsi), %al
+# CHECK-NEXT:  1      100   0.50                  U     lodsw	(%rsi), %ax
+# CHECK-NEXT:  1      100   0.50                  U     lodsl	(%rsi), %eax
+# CHECK-NEXT:  1      100   0.50                  U     lodsq	(%rsi), %rax
+# CHECK-NEXT:  1      100   0.50                  U     movsb	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     movsw	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     movsl	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     movsq	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  1      1     0.50                        movsbw	%al, %di
+# CHECK-NEXT:  1      1     0.50                        movzbw	%al, %di
 # CHECK-NEXT:  1      5     0.50    *                   movsbw	(%rax), %di
 # CHECK-NEXT:  1      5     0.50    *                   movzbw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        movsbl	%al, %edi
-# CHECK-NEXT:  1      1     0.33                        movzbl	%al, %edi
+# CHECK-NEXT:  1      1     0.50                        movsbl	%al, %edi
+# CHECK-NEXT:  1      1     0.50                        movzbl	%al, %edi
 # CHECK-NEXT:  1      5     0.50    *                   movsbl	(%rax), %edi
 # CHECK-NEXT:  1      5     0.50    *                   movzbl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        movsbq	%al, %rdi
-# CHECK-NEXT:  1      1     0.33                        movzbq	%al, %rdi
+# CHECK-NEXT:  1      1     0.50                        movsbq	%al, %rdi
+# CHECK-NEXT:  1      1     0.50                        movzbq	%al, %rdi
 # CHECK-NEXT:  1      5     0.50    *                   movsbq	(%rax), %rdi
 # CHECK-NEXT:  1      5     0.50    *                   movzbq	(%rax), %rdi
-# CHECK-NEXT:  1      1     0.33                        movswl	%ax, %edi
-# CHECK-NEXT:  1      1     0.33                        movzwl	%ax, %edi
+# CHECK-NEXT:  1      1     0.50                        movswl	%ax, %edi
+# CHECK-NEXT:  1      1     0.50                        movzwl	%ax, %edi
 # CHECK-NEXT:  1      5     0.50    *                   movswl	(%rax), %edi
 # CHECK-NEXT:  1      5     0.50    *                   movzwl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        movswq	%ax, %rdi
-# CHECK-NEXT:  1      1     0.33                        movzwq	%ax, %rdi
+# CHECK-NEXT:  1      1     0.50                        movswq	%ax, %rdi
+# CHECK-NEXT:  1      1     0.50                        movzwq	%ax, %rdi
 # CHECK-NEXT:  1      5     0.50    *                   movswq	(%rax), %rdi
 # CHECK-NEXT:  1      5     0.50    *                   movzwq	(%rax), %rdi
-# CHECK-NEXT:  1      1     0.33                        movslq	%eax, %rdi
+# CHECK-NEXT:  1      1     0.50                        movslq	%eax, %rdi
 # CHECK-NEXT:  1      5     0.50    *                   movslq	(%rax), %rdi
-# CHECK-NEXT:  1      3     1.00                        mulb	%dil
-# CHECK-NEXT:  2      8     1.00    *                   mulb	(%rax)
-# CHECK-NEXT:  4      4     1.33                        mulw	%si
-# CHECK-NEXT:  5      9     1.33    *                   mulw	(%rax)
-# CHECK-NEXT:  3      4     1.00                        mull	%edx
-# CHECK-NEXT:  4      9     1.00    *                   mull	(%rax)
-# CHECK-NEXT:  2      4     1.00                        mulq	%rcx
-# CHECK-NEXT:  3      9     1.00    *                   mulq	(%rax)
-# CHECK-NEXT:  1      1     0.33                        negb	%dil
-# CHECK-NEXT:  3      7     1.00    *      *            negb	(%r8)
-# CHECK-NEXT:  1      1     0.33                        negw	%si
-# CHECK-NEXT:  3      7     1.00    *      *            negw	(%r9)
-# CHECK-NEXT:  1      1     0.33                        negl	%edx
-# CHECK-NEXT:  3      7     1.00    *      *            negl	(%rax)
-# CHECK-NEXT:  1      1     0.33                        negq	%rcx
-# CHECK-NEXT:  3      7     1.00    *      *            negq	(%r10)
-# CHECK-NEXT:  1      1     0.25                        nop
-# CHECK-NEXT:  1      1     0.25                        nopw	%di
-# CHECK-NEXT:  1      1     0.25                        nopw	(%rcx)
-# CHECK-NEXT:  1      1     0.25                        nopl	%esi
-# CHECK-NEXT:  1      1     0.25                        nopl	(%r8)
-# CHECK-NEXT:  1      1     0.25                        nopq	%rdx
-# CHECK-NEXT:  1      1     0.25                        nopq	(%r9)
-# CHECK-NEXT:  1      1     0.33                        notb	%dil
-# CHECK-NEXT:  3      7     1.00    *      *            notb	(%r8)
-# CHECK-NEXT:  1      1     0.33                        notw	%si
-# CHECK-NEXT:  3      7     1.00    *      *            notw	(%r9)
-# CHECK-NEXT:  1      1     0.33                        notl	%edx
-# CHECK-NEXT:  3      7     1.00    *      *            notl	(%rax)
-# CHECK-NEXT:  1      1     0.33                        notq	%rcx
-# CHECK-NEXT:  3      7     1.00    *      *            notq	(%r10)
-# CHECK-NEXT:  1      1     0.33                        orb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        orb	$7, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            orb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orb	%sil, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            orb	%sil, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   orb	(%rax), %dil
-# CHECK-NEXT:  1      1     0.33                        orw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        orw	$511, %di
-# CHECK-NEXT:  3      7     1.00    *      *            orw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orw	$7, %di
-# CHECK-NEXT:  3      7     1.00    *      *            orw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orw	%si, %di
-# CHECK-NEXT:  3      7     1.00    *      *            orw	%si, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   orw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        orl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        orl	$665536, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            orl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orl	$7, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            orl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orl	%esi, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            orl	%esi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   orl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        orq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        orq	$665536, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            orq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orq	$7, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            orq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        orq	%rsi, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            orq	%rsi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   orq	(%rax), %rdi
-# CHECK-NEXT:  1      100   0.33                  U     outb	%al, $7
-# CHECK-NEXT:  1      100   0.33                  U     outb	%al, %dx
-# CHECK-NEXT:  1      100   0.33                  U     outw	%ax, $7
-# CHECK-NEXT:  1      100   0.33                  U     outw	%ax, %dx
-# CHECK-NEXT:  1      100   0.33                  U     outl	%eax, $7
-# CHECK-NEXT:  1      100   0.33                  U     outl	%eax, %dx
-# CHECK-NEXT:  1      100   0.33                  U     outsb	(%rsi), %dx
-# CHECK-NEXT:  1      100   0.33                  U     outsw	(%rsi), %dx
-# CHECK-NEXT:  1      100   0.33                  U     outsl	(%rsi), %dx
-# CHECK-NEXT:  4      4     1.33    *      *      U     pause
-# CHECK-NEXT:  3      2     1.50                        rclb	%dil
-# CHECK-NEXT:  3      2     1.50                        rcrb	%dil
-# CHECK-NEXT:  11     11    3.50           *            rclb	(%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrb	(%rax)
-# CHECK-NEXT:  8      5     4.00                        rclb	$7, %dil
-# CHECK-NEXT:  8      5     4.00                        rcrb	$7, %dil
-# CHECK-NEXT:  11     11    3.50           *            rclb	$7, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrb	$7, (%rax)
-# CHECK-NEXT:  8      5     4.00                        rclb	%cl, %dil
-# CHECK-NEXT:  8      5     4.00                        rcrb	%cl, %dil
-# CHECK-NEXT:  11     11    3.50           *            rclb	%cl, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrb	%cl, (%rax)
-# CHECK-NEXT:  3      2     1.50                        rclw	%di
-# CHECK-NEXT:  3      2     1.50                        rcrw	%di
-# CHECK-NEXT:  11     11    3.50           *            rclw	(%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrw	(%rax)
-# CHECK-NEXT:  8      5     4.00                        rclw	$7, %di
-# CHECK-NEXT:  8      5     4.00                        rcrw	$7, %di
-# CHECK-NEXT:  11     11    3.50           *            rclw	$7, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrw	$7, (%rax)
-# CHECK-NEXT:  8      5     4.00                        rclw	%cl, %di
-# CHECK-NEXT:  8      5     4.00                        rcrw	%cl, %di
-# CHECK-NEXT:  11     11    3.50           *            rclw	%cl, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrw	%cl, (%rax)
-# CHECK-NEXT:  3      2     1.50                        rcll	%edi
-# CHECK-NEXT:  3      2     1.50                        rcrl	%edi
-# CHECK-NEXT:  11     11    3.50           *            rcll	(%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrl	(%rax)
-# CHECK-NEXT:  8      5     4.00                        rcll	$7, %edi
-# CHECK-NEXT:  8      5     4.00                        rcrl	$7, %edi
-# CHECK-NEXT:  11     11    3.50           *            rcll	$7, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrl	$7, (%rax)
-# CHECK-NEXT:  8      5     4.00                        rcll	%cl, %edi
-# CHECK-NEXT:  8      5     4.00                        rcrl	%cl, %edi
-# CHECK-NEXT:  11     11    3.50           *            rcll	%cl, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrl	%cl, (%rax)
-# CHECK-NEXT:  3      2     1.50                        rclq	%rdi
-# CHECK-NEXT:  3      2     1.50                        rcrq	%rdi
-# CHECK-NEXT:  11     11    3.50           *            rclq	(%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrq	(%rax)
-# CHECK-NEXT:  8      5     4.00                        rclq	$7, %rdi
-# CHECK-NEXT:  8      5     4.00                        rcrq	$7, %rdi
-# CHECK-NEXT:  11     11    3.50           *            rclq	$7, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrq	$7, (%rax)
-# CHECK-NEXT:  8      5     4.00                        rclq	%cl, %rdi
-# CHECK-NEXT:  8      5     4.00                        rcrq	%cl, %rdi
-# CHECK-NEXT:  11     11    3.50           *            rclq	%cl, (%rax)
-# CHECK-NEXT:  11     11    3.50           *            rcrq	%cl, (%rax)
-# CHECK-NEXT:  2      2     1.00                        rolb	%dil
-# CHECK-NEXT:  2      2     1.00                        rorb	%dil
-# CHECK-NEXT:  5      8     1.00    *      *            rolb	(%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorb	(%rax)
-# CHECK-NEXT:  2      2     1.00                        rolb	$7, %dil
-# CHECK-NEXT:  2      2     1.00                        rorb	$7, %dil
-# CHECK-NEXT:  5      8     1.00    *      *            rolb	$7, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorb	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        rolb	%cl, %dil
-# CHECK-NEXT:  3      3     1.50                        rorb	%cl, %dil
-# CHECK-NEXT:  6      9     1.50    *      *            rolb	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            rorb	%cl, (%rax)
-# CHECK-NEXT:  2      2     1.00                        rolw	%di
-# CHECK-NEXT:  2      2     1.00                        rorw	%di
-# CHECK-NEXT:  5      8     1.00    *      *            rolw	(%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorw	(%rax)
-# CHECK-NEXT:  2      2     1.00                        rolw	$7, %di
-# CHECK-NEXT:  2      2     1.00                        rorw	$7, %di
-# CHECK-NEXT:  5      8     1.00    *      *            rolw	$7, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorw	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        rolw	%cl, %di
-# CHECK-NEXT:  3      3     1.50                        rorw	%cl, %di
-# CHECK-NEXT:  6      9     1.50    *      *            rolw	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            rorw	%cl, (%rax)
-# CHECK-NEXT:  2      2     1.00                        roll	%edi
-# CHECK-NEXT:  2      2     1.00                        rorl	%edi
-# CHECK-NEXT:  5      8     1.00    *      *            roll	(%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorl	(%rax)
-# CHECK-NEXT:  2      2     1.00                        roll	$7, %edi
-# CHECK-NEXT:  2      2     1.00                        rorl	$7, %edi
-# CHECK-NEXT:  5      8     1.00    *      *            roll	$7, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorl	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        roll	%cl, %edi
-# CHECK-NEXT:  3      3     1.50                        rorl	%cl, %edi
-# CHECK-NEXT:  6      9     1.50    *      *            roll	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            rorl	%cl, (%rax)
-# CHECK-NEXT:  2      2     1.00                        rolq	%rdi
-# CHECK-NEXT:  2      2     1.00                        rorq	%rdi
-# CHECK-NEXT:  5      8     1.00    *      *            rolq	(%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorq	(%rax)
-# CHECK-NEXT:  2      2     1.00                        rolq	$7, %rdi
-# CHECK-NEXT:  2      2     1.00                        rorq	$7, %rdi
-# CHECK-NEXT:  5      8     1.00    *      *            rolq	$7, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            rorq	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        rolq	%cl, %rdi
-# CHECK-NEXT:  3      3     1.50                        rorq	%cl, %rdi
-# CHECK-NEXT:  6      9     1.50    *      *            rolq	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            rorq	%cl, (%rax)
-# CHECK-NEXT:  1      1     0.50                        sahf
+# CHECK-NEXT:  1      4     1.00                        mulb	%dil
+# CHECK-NEXT:  1      8     1.00    *                   mulb	(%rax)
+# CHECK-NEXT:  2      4     1.00                        mulw	%si
+# CHECK-NEXT:  2      8     1.00    *                   mulw	(%rax)
+# CHECK-NEXT:  1      4     1.00                        mull	%edx
+# CHECK-NEXT:  1      8     1.00    *                   mull	(%rax)
+# CHECK-NEXT:  1      6     4.00                        mulq	%rcx
+# CHECK-NEXT:  1      10    4.00    *                   mulq	(%rax)
+# CHECK-NEXT:  1      1     0.50                        negb	%dil
+# CHECK-NEXT:  2      6     1.00    *      *            negb	(%r8)
+# CHECK-NEXT:  1      1     0.50                        negw	%si
+# CHECK-NEXT:  2      6     1.00    *      *            negw	(%r9)
+# CHECK-NEXT:  1      1     0.50                        negl	%edx
+# CHECK-NEXT:  2      6     1.00    *      *            negl	(%rax)
+# CHECK-NEXT:  1      1     0.50                        negq	%rcx
+# CHECK-NEXT:  2      6     1.00    *      *            negq	(%r10)
+# CHECK-NEXT:  1      1     0.50                        nop
+# CHECK-NEXT:  1      1     0.50                        nopw	%di
+# CHECK-NEXT:  1      1     0.50                        nopw	(%rcx)
+# CHECK-NEXT:  1      1     0.50                        nopl	%esi
+# CHECK-NEXT:  1      1     0.50                        nopl	(%r8)
+# CHECK-NEXT:  1      1     0.50                        nopq	%rdx
+# CHECK-NEXT:  1      1     0.50                        nopq	(%r9)
+# CHECK-NEXT:  1      1     0.50                        notb	%dil
+# CHECK-NEXT:  2      6     1.00    *      *            notb	(%r8)
+# CHECK-NEXT:  1      1     0.50                        notw	%si
+# CHECK-NEXT:  2      6     1.00    *      *            notw	(%r9)
+# CHECK-NEXT:  1      1     0.50                        notl	%edx
+# CHECK-NEXT:  2      6     1.00    *      *            notl	(%rax)
+# CHECK-NEXT:  1      1     0.50                        notq	%rcx
+# CHECK-NEXT:  2      6     1.00    *      *            notq	(%r10)
+# CHECK-NEXT:  1      1     0.50                        orb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        orb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            orb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            orb	%sil, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   orb	(%rax), %dil
+# CHECK-NEXT:  1      1     0.50                        orw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        orw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            orw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            orw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            orw	%si, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   orw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        orl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        orl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            orl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            orl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            orl	%esi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   orl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        orq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        orq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            orq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            orq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        orq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            orq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   orq	(%rax), %rdi
+# CHECK-NEXT:  1      100   0.50                  U     outb	%al, $7
+# CHECK-NEXT:  1      100   0.50                  U     outb	%al, %dx
+# CHECK-NEXT:  1      100   0.50                  U     outw	%ax, $7
+# CHECK-NEXT:  1      100   0.50                  U     outw	%ax, %dx
+# CHECK-NEXT:  1      100   0.50                  U     outl	%eax, $7
+# CHECK-NEXT:  1      100   0.50                  U     outl	%eax, %dx
+# CHECK-NEXT:  1      100   0.50                  U     outsb	(%rsi), %dx
+# CHECK-NEXT:  1      100   0.50                  U     outsw	(%rsi), %dx
+# CHECK-NEXT:  1      100   0.50                  U     outsl	(%rsi), %dx
+# CHECK-NEXT:  1      1     0.50    *      *      U     pause
+# CHECK-NEXT:  1      1     0.50                        rclb	%dil
+# CHECK-NEXT:  1      1     0.50                        rcrb	%dil
+# CHECK-NEXT:  2      5     1.00           *            rclb	(%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrb	(%rax)
+# CHECK-NEXT:  25     13    0.50                        rclb	$7, %dil
+# CHECK-NEXT:  23     12    0.50                        rcrb	$7, %dil
+# CHECK-NEXT:  2      5     1.00           *            rclb	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrb	$7, (%rax)
+# CHECK-NEXT:  26     12    0.50                        rclb	%cl, %dil
+# CHECK-NEXT:  24     11    0.50                        rcrb	%cl, %dil
+# CHECK-NEXT:  2      5     1.00           *            rclb	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrb	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rclw	%di
+# CHECK-NEXT:  1      1     0.50                        rcrw	%di
+# CHECK-NEXT:  2      5     1.00           *            rclw	(%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrw	(%rax)
+# CHECK-NEXT:  21     11    0.50                        rclw	$7, %di
+# CHECK-NEXT:  19     10    0.50                        rcrw	$7, %di
+# CHECK-NEXT:  2      5     1.00           *            rclw	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrw	$7, (%rax)
+# CHECK-NEXT:  22     10    0.50                        rclw	%cl, %di
+# CHECK-NEXT:  20     9     0.50                        rcrw	%cl, %di
+# CHECK-NEXT:  2      5     1.00           *            rclw	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrw	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rcll	%edi
+# CHECK-NEXT:  1      1     0.50                        rcrl	%edi
+# CHECK-NEXT:  2      5     1.00           *            rcll	(%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrl	(%rax)
+# CHECK-NEXT:  16     8     0.50                        rcll	$7, %edi
+# CHECK-NEXT:  15     7     0.50                        rcrl	$7, %edi
+# CHECK-NEXT:  2      5     1.00           *            rcll	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrl	$7, (%rax)
+# CHECK-NEXT:  17     7     0.50                        rcll	%cl, %edi
+# CHECK-NEXT:  16     7     0.50                        rcrl	%cl, %edi
+# CHECK-NEXT:  2      5     1.00           *            rcll	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrl	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rclq	%rdi
+# CHECK-NEXT:  1      1     0.50                        rcrq	%rdi
+# CHECK-NEXT:  2      5     1.00           *            rclq	(%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrq	(%rax)
+# CHECK-NEXT:  16     8     0.50                        rclq	$7, %rdi
+# CHECK-NEXT:  15     7     0.50                        rcrq	$7, %rdi
+# CHECK-NEXT:  2      5     1.00           *            rclq	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrq	$7, (%rax)
+# CHECK-NEXT:  17     7     0.50                        rclq	%cl, %rdi
+# CHECK-NEXT:  16     7     0.50                        rcrq	%cl, %rdi
+# CHECK-NEXT:  2      5     1.00           *            rclq	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00           *            rcrq	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rolb	%dil
+# CHECK-NEXT:  1      1     0.50                        rorb	%dil
+# CHECK-NEXT:  2      5     1.00    *      *            rolb	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorb	(%rax)
+# CHECK-NEXT:  1      1     0.50                        rolb	$7, %dil
+# CHECK-NEXT:  1      1     0.50                        rorb	$7, %dil
+# CHECK-NEXT:  2      5     1.00    *      *            rolb	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rolb	%cl, %dil
+# CHECK-NEXT:  1      1     0.50                        rorb	%cl, %dil
+# CHECK-NEXT:  2      5     1.00    *      *            rolb	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorb	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rolw	%di
+# CHECK-NEXT:  1      1     0.50                        rorw	%di
+# CHECK-NEXT:  2      5     1.00    *      *            rolw	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorw	(%rax)
+# CHECK-NEXT:  1      1     0.50                        rolw	$7, %di
+# CHECK-NEXT:  1      1     0.50                        rorw	$7, %di
+# CHECK-NEXT:  2      5     1.00    *      *            rolw	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rolw	%cl, %di
+# CHECK-NEXT:  1      1     0.50                        rorw	%cl, %di
+# CHECK-NEXT:  2      5     1.00    *      *            rolw	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorw	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        roll	%edi
+# CHECK-NEXT:  1      1     0.50                        rorl	%edi
+# CHECK-NEXT:  2      5     1.00    *      *            roll	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorl	(%rax)
+# CHECK-NEXT:  1      1     0.50                        roll	$7, %edi
+# CHECK-NEXT:  1      1     0.50                        rorl	$7, %edi
+# CHECK-NEXT:  2      5     1.00    *      *            roll	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        roll	%cl, %edi
+# CHECK-NEXT:  1      1     0.50                        rorl	%cl, %edi
+# CHECK-NEXT:  2      5     1.00    *      *            roll	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorl	%cl, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rolq	%rdi
+# CHECK-NEXT:  1      1     0.50                        rorq	%rdi
+# CHECK-NEXT:  2      5     1.00    *      *            rolq	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorq	(%rax)
+# CHECK-NEXT:  1      1     0.50                        rolq	$7, %rdi
+# CHECK-NEXT:  1      1     0.50                        rorq	$7, %rdi
+# CHECK-NEXT:  2      5     1.00    *      *            rolq	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        rolq	%cl, %rdi
+# CHECK-NEXT:  1      1     0.50                        rorq	%cl, %rdi
+# CHECK-NEXT:  2      5     1.00    *      *            rolq	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            rorq	%cl, (%rax)
+# CHECK-NEXT:  2      2     0.50                        sahf
 # CHECK-NEXT:  1      1     0.50                        sarb	%dil
 # CHECK-NEXT:  1      1     0.50                        shlb	%dil
 # CHECK-NEXT:  1      1     0.50                        shrb	%dil
-# CHECK-NEXT:  4      7     1.00    *      *            sarb	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shlb	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrb	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarb	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlb	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrb	(%rax)
 # CHECK-NEXT:  1      1     0.50                        sarb	$7, %dil
 # CHECK-NEXT:  1      1     0.50                        shlb	$7, %dil
 # CHECK-NEXT:  1      1     0.50                        shrb	$7, %dil
-# CHECK-NEXT:  4      7     1.00    *      *            sarb	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shlb	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrb	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        sarb	%cl, %dil
-# CHECK-NEXT:  3      3     1.50                        shlb	%cl, %dil
-# CHECK-NEXT:  3      3     1.50                        shrb	%cl, %dil
-# CHECK-NEXT:  6      9     1.50    *      *            sarb	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shlb	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shrb	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarb	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlb	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        sarb	%cl, %dil
+# CHECK-NEXT:  1      1     0.50                        shlb	%cl, %dil
+# CHECK-NEXT:  1      1     0.50                        shrb	%cl, %dil
+# CHECK-NEXT:  2      5     1.00    *      *            sarb	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlb	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrb	%cl, (%rax)
 # CHECK-NEXT:  1      1     0.50                        sarw	%di
 # CHECK-NEXT:  1      1     0.50                        shlw	%di
 # CHECK-NEXT:  1      1     0.50                        shrw	%di
-# CHECK-NEXT:  4      7     1.00    *      *            sarw	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shlw	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrw	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarw	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlw	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrw	(%rax)
 # CHECK-NEXT:  1      1     0.50                        sarw	$7, %di
 # CHECK-NEXT:  1      1     0.50                        shlw	$7, %di
 # CHECK-NEXT:  1      1     0.50                        shrw	$7, %di
-# CHECK-NEXT:  4      7     1.00    *      *            sarw	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shlw	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrw	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        sarw	%cl, %di
-# CHECK-NEXT:  3      3     1.50                        shlw	%cl, %di
-# CHECK-NEXT:  3      3     1.50                        shrw	%cl, %di
-# CHECK-NEXT:  6      9     1.50    *      *            sarw	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shlw	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shrw	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarw	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlw	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        sarw	%cl, %di
+# CHECK-NEXT:  1      1     0.50                        shlw	%cl, %di
+# CHECK-NEXT:  1      1     0.50                        shrw	%cl, %di
+# CHECK-NEXT:  2      5     1.00    *      *            sarw	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlw	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrw	%cl, (%rax)
 # CHECK-NEXT:  1      1     0.50                        sarl	%edi
 # CHECK-NEXT:  1      1     0.50                        shll	%edi
 # CHECK-NEXT:  1      1     0.50                        shrl	%edi
-# CHECK-NEXT:  4      7     1.00    *      *            sarl	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shll	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrl	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarl	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shll	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrl	(%rax)
 # CHECK-NEXT:  1      1     0.50                        sarl	$7, %edi
 # CHECK-NEXT:  1      1     0.50                        shll	$7, %edi
 # CHECK-NEXT:  1      1     0.50                        shrl	$7, %edi
-# CHECK-NEXT:  4      7     1.00    *      *            sarl	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shll	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrl	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        sarl	%cl, %edi
-# CHECK-NEXT:  3      3     1.50                        shll	%cl, %edi
-# CHECK-NEXT:  3      3     1.50                        shrl	%cl, %edi
-# CHECK-NEXT:  6      9     1.50    *      *            sarl	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shll	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shrl	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarl	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shll	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        sarl	%cl, %edi
+# CHECK-NEXT:  1      1     0.50                        shll	%cl, %edi
+# CHECK-NEXT:  1      1     0.50                        shrl	%cl, %edi
+# CHECK-NEXT:  2      5     1.00    *      *            sarl	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shll	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrl	%cl, (%rax)
 # CHECK-NEXT:  1      1     0.50                        sarq	%rdi
 # CHECK-NEXT:  1      1     0.50                        shlq	%rdi
 # CHECK-NEXT:  1      1     0.50                        shrq	%rdi
-# CHECK-NEXT:  4      7     1.00    *      *            sarq	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shlq	(%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrq	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            sarq	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlq	(%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrq	(%rax)
 # CHECK-NEXT:  1      1     0.50                        sarq	$7, %rdi
 # CHECK-NEXT:  1      1     0.50                        shlq	$7, %rdi
 # CHECK-NEXT:  1      1     0.50                        shrq	$7, %rdi
-# CHECK-NEXT:  4      7     1.00    *      *            sarq	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shlq	$7, (%rax)
-# CHECK-NEXT:  4      7     1.00    *      *            shrq	$7, (%rax)
-# CHECK-NEXT:  3      3     1.50                        sarq	%cl, %rdi
-# CHECK-NEXT:  3      3     1.50                        shlq	%cl, %rdi
-# CHECK-NEXT:  3      3     1.50                        shrq	%cl, %rdi
-# CHECK-NEXT:  6      9     1.50    *      *            sarq	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shlq	%cl, (%rax)
-# CHECK-NEXT:  6      9     1.50    *      *            shrq	%cl, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbb	$7, %al
-# CHECK-NEXT:  2      2     0.67                        sbbb	$7, %dil
-# CHECK-NEXT:  6      9     1.00    *      *            sbbb	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbb	%sil, %dil
-# CHECK-NEXT:  6      9     1.00    *      *            sbbb	%sil, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   sbbb	(%rax), %dil
-# CHECK-NEXT:  2      2     0.67                        sbbw	$511, %ax
-# CHECK-NEXT:  2      2     0.67                        sbbw	$511, %di
-# CHECK-NEXT:  6      9     1.00    *      *            sbbw	$511, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbw	$7, %di
-# CHECK-NEXT:  6      9     1.00    *      *            sbbw	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbw	%si, %di
-# CHECK-NEXT:  6      9     1.00    *      *            sbbw	%si, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   sbbw	(%rax), %di
-# CHECK-NEXT:  2      2     0.67                        sbbl	$665536, %eax
-# CHECK-NEXT:  2      2     0.67                        sbbl	$665536, %edi
-# CHECK-NEXT:  6      9     1.00    *      *            sbbl	$665536, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbl	$7, %edi
-# CHECK-NEXT:  6      9     1.00    *      *            sbbl	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbl	%esi, %edi
-# CHECK-NEXT:  6      9     1.00    *      *            sbbl	%esi, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   sbbl	(%rax), %edi
-# CHECK-NEXT:  2      2     0.67                        sbbq	$665536, %rax
-# CHECK-NEXT:  2      2     0.67                        sbbq	$665536, %rdi
-# CHECK-NEXT:  6      9     1.00    *      *            sbbq	$665536, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbq	$7, %rdi
-# CHECK-NEXT:  6      9     1.00    *      *            sbbq	$7, (%rax)
-# CHECK-NEXT:  2      2     0.67                        sbbq	%rsi, %rdi
-# CHECK-NEXT:  6      9     1.00    *      *            sbbq	%rsi, (%rax)
-# CHECK-NEXT:  3      7     0.67    *                   sbbq	(%rax), %rdi
-# CHECK-NEXT:  2      2     0.67                  U     scasb	%es:(%rdi), %al
-# CHECK-NEXT:  2      2     0.67                  U     scasw	%es:(%rdi), %ax
-# CHECK-NEXT:  2      2     0.67                  U     scasl	%es:(%rdi), %eax
-# CHECK-NEXT:  2      2     0.67                  U     scasq	%es:(%rdi), %rax
+# CHECK-NEXT:  2      5     1.00    *      *            sarq	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlq	$7, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        sarq	%cl, %rdi
+# CHECK-NEXT:  1      1     0.50                        shlq	%cl, %rdi
+# CHECK-NEXT:  1      1     0.50                        shrq	%cl, %rdi
+# CHECK-NEXT:  2      5     1.00    *      *            sarq	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shlq	%cl, (%rax)
+# CHECK-NEXT:  2      5     1.00    *      *            shrq	%cl, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbb	$7, %al
+# CHECK-NEXT:  1      1     1.00                        sbbb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            sbbb	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            sbbb	%sil, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   sbbb	(%rax), %dil
+# CHECK-NEXT:  1      1     1.00                        sbbw	$511, %ax
+# CHECK-NEXT:  1      1     1.00                        sbbw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            sbbw	$511, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            sbbw	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            sbbw	%si, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   sbbw	(%rax), %di
+# CHECK-NEXT:  1      1     1.00                        sbbl	$665536, %eax
+# CHECK-NEXT:  1      1     1.00                        sbbl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            sbbl	$665536, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            sbbl	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            sbbl	%esi, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   sbbl	(%rax), %edi
+# CHECK-NEXT:  1      1     1.00                        sbbq	$665536, %rax
+# CHECK-NEXT:  1      1     1.00                        sbbq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            sbbq	$665536, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            sbbq	$7, (%rax)
+# CHECK-NEXT:  1      1     1.00                        sbbq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            sbbq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     1.00    *                   sbbq	(%rax), %rdi
+# CHECK-NEXT:  1      100   0.50                  U     scasb	%es:(%rdi), %al
+# CHECK-NEXT:  1      100   0.50                  U     scasw	%es:(%rdi), %ax
+# CHECK-NEXT:  1      100   0.50                  U     scasl	%es:(%rdi), %eax
+# CHECK-NEXT:  1      100   0.50                  U     scasq	%es:(%rdi), %rax
 # CHECK-NEXT:  1      1     0.50                        seto	%al
-# CHECK-NEXT:  3      2     1.00           *            seto	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            seto	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setno	%al
-# CHECK-NEXT:  3      2     1.00           *            setno	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setno	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setb	%al
-# CHECK-NEXT:  3      2     1.00           *            setb	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setb	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setae	%al
-# CHECK-NEXT:  3      2     1.00           *            setae	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setae	(%rax)
 # CHECK-NEXT:  1      1     0.50                        sete	%al
-# CHECK-NEXT:  3      2     1.00           *            sete	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            sete	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setne	%al
-# CHECK-NEXT:  3      2     1.00           *            setne	(%rax)
-# CHECK-NEXT:  2      2     1.00                        seta	%al
-# CHECK-NEXT:  4      3     1.00           *            seta	(%rax)
-# CHECK-NEXT:  2      2     1.00                        setbe	%al
-# CHECK-NEXT:  4      3     1.00           *            setbe	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setne	(%rax)
+# CHECK-NEXT:  1      1     0.50                        seta	%al
+# CHECK-NEXT:  1      1     0.50           *            seta	(%rax)
+# CHECK-NEXT:  1      1     0.50                        setbe	%al
+# CHECK-NEXT:  1      1     0.50           *            setbe	(%rax)
 # CHECK-NEXT:  1      1     0.50                        sets	%al
-# CHECK-NEXT:  3      2     1.00           *            sets	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            sets	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setns	%al
-# CHECK-NEXT:  3      2     1.00           *            setns	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setns	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setp	%al
-# CHECK-NEXT:  3      2     1.00           *            setp	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setp	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setnp	%al
-# CHECK-NEXT:  3      2     1.00           *            setnp	(%rax)
+# CHECK-NEXT:  1      1     0.50           *            setnp	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setl	%al
-# CHECK-NEXT:  3      2     1.00           *            setl	(%rax)
+# CHECK-NEXT:  2      1     1.00           *            setl	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setge	%al
-# CHECK-NEXT:  3      2     1.00           *            setge	(%rax)
+# CHECK-NEXT:  2      1     1.00           *            setge	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setg	%al
-# CHECK-NEXT:  3      2     1.00           *            setg	(%rax)
+# CHECK-NEXT:  2      1     1.00           *            setg	(%rax)
 # CHECK-NEXT:  1      1     0.50                        setle	%al
-# CHECK-NEXT:  3      2     1.00           *            setle	(%rax)
-# CHECK-NEXT:  4      4     1.50                        shldw	%cl, %si, %di
-# CHECK-NEXT:  4      4     1.50                        shrdw	%cl, %si, %di
-# CHECK-NEXT:  7      10    1.50    *      *            shldw	%cl, %si, (%rax)
-# CHECK-NEXT:  7      10    1.50    *      *            shrdw	%cl, %si, (%rax)
-# CHECK-NEXT:  2      2     0.67                        shldw	$7, %si, %di
-# CHECK-NEXT:  2      2     0.67                        shrdw	$7, %si, %di
-# CHECK-NEXT:  5      8     1.00    *      *            shldw	$7, %si, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            shrdw	$7, %si, (%rax)
-# CHECK-NEXT:  4      4     1.50                        shldl	%cl, %esi, %edi
-# CHECK-NEXT:  4      4     1.50                        shrdl	%cl, %esi, %edi
-# CHECK-NEXT:  7      10    1.50    *      *            shldl	%cl, %esi, (%rax)
-# CHECK-NEXT:  7      10    1.50    *      *            shrdl	%cl, %esi, (%rax)
-# CHECK-NEXT:  2      2     0.67                        shldl	$7, %esi, %edi
-# CHECK-NEXT:  2      2     0.67                        shrdl	$7, %esi, %edi
-# CHECK-NEXT:  5      8     1.00    *      *            shldl	$7, %esi, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            shrdl	$7, %esi, (%rax)
-# CHECK-NEXT:  4      4     1.50                        shldq	%cl, %rsi, %rdi
-# CHECK-NEXT:  4      4     1.50                        shrdq	%cl, %rsi, %rdi
-# CHECK-NEXT:  7      10    1.50    *      *            shldq	%cl, %rsi, (%rax)
-# CHECK-NEXT:  7      10    1.50    *      *            shrdq	%cl, %rsi, (%rax)
-# CHECK-NEXT:  2      2     0.67                        shldq	$7, %rsi, %rdi
-# CHECK-NEXT:  2      2     0.67                        shrdq	$7, %rsi, %rdi
-# CHECK-NEXT:  5      8     1.00    *      *            shldq	$7, %rsi, (%rax)
-# CHECK-NEXT:  5      8     1.00    *      *            shrdq	$7, %rsi, (%rax)
-# CHECK-NEXT:  1      1     0.33                  U     stc
-# CHECK-NEXT:  1      1     0.33                  U     std
-# CHECK-NEXT:  3      5     1.00                  U     stosb	%al, %es:(%rdi)
-# CHECK-NEXT:  3      5     1.00                  U     stosw	%ax, %es:(%rdi)
-# CHECK-NEXT:  3      5     1.00                  U     stosl	%eax, %es:(%rdi)
-# CHECK-NEXT:  3      5     1.00                  U     stosq	%rax, %es:(%rdi)
-# CHECK-NEXT:  1      1     0.33                        subb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        subb	$7, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            subb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subb	%sil, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            subb	%sil, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   subb	(%rax), %dil
-# CHECK-NEXT:  1      1     0.33                        subw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        subw	$511, %di
-# CHECK-NEXT:  3      7     1.00    *      *            subw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subw	$7, %di
-# CHECK-NEXT:  3      7     1.00    *      *            subw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subw	%si, %di
-# CHECK-NEXT:  3      7     1.00    *      *            subw	%si, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   subw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        subl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        subl	$665536, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            subl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subl	$7, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            subl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subl	%esi, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            subl	%esi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   subl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        subq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        subq	$665536, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            subq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subq	$7, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            subq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        subq	%rsi, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            subq	%rsi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   subq	(%rax), %rdi
-# CHECK-NEXT:  1      1     0.33                        testb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        testb	$7, %dil
-# CHECK-NEXT:  2      6     0.50    *                   testb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testb	%sil, %dil
-# CHECK-NEXT:  2      6     0.50    *                   testb	%sil, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        testw	$511, %di
-# CHECK-NEXT:  2      6     0.50    *                   testw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testw	$7, %di
-# CHECK-NEXT:  2      6     0.50    *                   testw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testw	%si, %di
-# CHECK-NEXT:  2      6     0.50    *                   testw	%si, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        testl	$665536, %edi
-# CHECK-NEXT:  2      6     0.50    *                   testl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testl	$7, %edi
-# CHECK-NEXT:  2      6     0.50    *                   testl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testl	%esi, %edi
-# CHECK-NEXT:  2      6     0.50    *                   testl	%esi, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        testq	$665536, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   testq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testq	$7, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   testq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        testq	%rsi, %rdi
-# CHECK-NEXT:  2      6     0.50    *                   testq	%rsi, (%rax)
-# CHECK-NEXT:  1      100   0.33    *             U     ud2
-# CHECK-NEXT:  3      2     1.00                        xaddb	%bl, %cl
-# CHECK-NEXT:  5      8     1.00    *      *            xaddb	%bl, (%rcx)
-# CHECK-NEXT:  3      2     1.00                        xaddw	%bx, %cx
-# CHECK-NEXT:  5      8     1.00    *      *            xaddw	%ax, (%rbx)
-# CHECK-NEXT:  3      2     1.00                        xaddl	%ebx, %ecx
-# CHECK-NEXT:  5      8     1.00    *      *            xaddl	%eax, (%rbx)
-# CHECK-NEXT:  3      2     1.00                        xaddq	%rbx, %rcx
-# CHECK-NEXT:  5      8     1.00    *      *            xaddq	%rax, (%rbx)
-# CHECK-NEXT:  3      2     1.00                        xchgb	%bl, %cl
-# CHECK-NEXT:  3      6     1.00    *      *            xchgb	%bl, (%rbx)
-# CHECK-NEXT:  3      2     1.00                        xchgw	%bx, %ax
-# CHECK-NEXT:  3      2     1.00                        xchgw	%bx, %cx
-# CHECK-NEXT:  3      6     1.00    *      *            xchgw	%ax, (%rbx)
-# CHECK-NEXT:  3      2     1.00                        xchgl	%ebx, %eax
-# CHECK-NEXT:  3      2     1.00                        xchgl	%ebx, %ecx
-# CHECK-NEXT:  3      6     1.00    *      *            xchgl	%eax, (%rbx)
-# CHECK-NEXT:  3      2     1.00                        xchgq	%rbx, %rax
-# CHECK-NEXT:  3      2     1.00                        xchgq	%rbx, %rcx
-# CHECK-NEXT:  3      6     1.00    *      *            xchgq	%rax, (%rbx)
-# CHECK-NEXT:  1      5     0.50    *                   xlatb
-# CHECK-NEXT:  1      1     0.33                        xorb	$7, %al
-# CHECK-NEXT:  1      1     0.33                        xorb	$7, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            xorb	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorb	%sil, %dil
-# CHECK-NEXT:  3      7     1.00    *      *            xorb	%sil, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   xorb	(%rax), %dil
-# CHECK-NEXT:  1      1     0.33                        xorw	$511, %ax
-# CHECK-NEXT:  1      1     0.33                        xorw	$511, %di
-# CHECK-NEXT:  3      7     1.00    *      *            xorw	$511, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorw	$7, %di
-# CHECK-NEXT:  3      7     1.00    *      *            xorw	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorw	%si, %di
-# CHECK-NEXT:  3      7     1.00    *      *            xorw	%si, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   xorw	(%rax), %di
-# CHECK-NEXT:  1      1     0.33                        xorl	$665536, %eax
-# CHECK-NEXT:  1      1     0.33                        xorl	$665536, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            xorl	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorl	$7, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            xorl	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorl	%esi, %edi
-# CHECK-NEXT:  3      7     1.00    *      *            xorl	%esi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   xorl	(%rax), %edi
-# CHECK-NEXT:  1      1     0.33                        xorq	$665536, %rax
-# CHECK-NEXT:  1      1     0.33                        xorq	$665536, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            xorq	$665536, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorq	$7, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            xorq	$7, (%rax)
-# CHECK-NEXT:  1      1     0.33                        xorq	%rsi, %rdi
-# CHECK-NEXT:  3      7     1.00    *      *            xorq	%rsi, (%rax)
-# CHECK-NEXT:  2      6     0.50    *                   xorq	(%rax), %rdi
+# CHECK-NEXT:  2      1     1.00           *            setle	(%rax)
+# CHECK-NEXT:  7      4     4.00                        shldw	%cl, %si, %di
+# CHECK-NEXT:  7      4     4.00                        shrdw	%cl, %si, %di
+# CHECK-NEXT:  8      4     11.00   *      *            shldw	%cl, %si, (%rax)
+# CHECK-NEXT:  8      4     11.00   *      *            shrdw	%cl, %si, (%rax)
+# CHECK-NEXT:  6      4     3.00                        shldw	$7, %si, %di
+# CHECK-NEXT:  6      3     3.00                        shrdw	$7, %si, %di
+# CHECK-NEXT:  8      4     11.00   *      *            shldw	$7, %si, (%rax)
+# CHECK-NEXT:  8      4     11.00   *      *            shrdw	$7, %si, (%rax)
+# CHECK-NEXT:  7      4     4.00                        shldl	%cl, %esi, %edi
+# CHECK-NEXT:  7      4     4.00                        shrdl	%cl, %esi, %edi
+# CHECK-NEXT:  8      4     11.00   *      *            shldl	%cl, %esi, (%rax)
+# CHECK-NEXT:  8      4     11.00   *      *            shrdl	%cl, %esi, (%rax)
+# CHECK-NEXT:  6      3     3.00                        shldl	$7, %esi, %edi
+# CHECK-NEXT:  6      4     3.00                        shrdl	$7, %esi, %edi
+# CHECK-NEXT:  8      4     11.00   *      *            shldl	$7, %esi, (%rax)
+# CHECK-NEXT:  8      4     11.00   *      *            shrdl	$7, %esi, (%rax)
+# CHECK-NEXT:  7      4     4.00                        shldq	%cl, %rsi, %rdi
+# CHECK-NEXT:  7      4     4.00                        shrdq	%cl, %rsi, %rdi
+# CHECK-NEXT:  8      4     11.00   *      *            shldq	%cl, %rsi, (%rax)
+# CHECK-NEXT:  8      4     11.00   *      *            shrdq	%cl, %rsi, (%rax)
+# CHECK-NEXT:  6      4     3.00                        shldq	$7, %rsi, %rdi
+# CHECK-NEXT:  6      4     3.00                        shrdq	$7, %rsi, %rdi
+# CHECK-NEXT:  8      4     11.00   *      *            shldq	$7, %rsi, (%rax)
+# CHECK-NEXT:  8      4     11.00   *      *            shrdq	$7, %rsi, (%rax)
+# CHECK-NEXT:  1      1     0.50                  U     stc
+# CHECK-NEXT:  1      1     0.50                  U     std
+# CHECK-NEXT:  1      100   0.50                  U     stosb	%al, %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     stosw	%ax, %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     stosl	%eax, %es:(%rdi)
+# CHECK-NEXT:  1      100   0.50                  U     stosq	%rax, %es:(%rdi)
+# CHECK-NEXT:  1      1     0.50                        subb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        subb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            subb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            subb	%sil, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   subb	(%rax), %dil
+# CHECK-NEXT:  1      1     0.50                        subw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        subw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            subw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            subw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            subw	%si, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   subw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        subl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        subl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            subl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            subl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            subl	%esi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   subl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        subq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        subq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            subq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            subq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        subq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            subq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   subq	(%rax), %rdi
+# CHECK-NEXT:  1      1     0.50                        testb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        testb	$7, %dil
+# CHECK-NEXT:  1      5     0.50    *                   testb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testb	%sil, %dil
+# CHECK-NEXT:  1      5     0.50    *                   testb	%sil, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        testw	$511, %di
+# CHECK-NEXT:  1      5     0.50    *                   testw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testw	$7, %di
+# CHECK-NEXT:  1      5     0.50    *                   testw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testw	%si, %di
+# CHECK-NEXT:  1      5     0.50    *                   testw	%si, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        testl	$665536, %edi
+# CHECK-NEXT:  1      5     0.50    *                   testl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testl	$7, %edi
+# CHECK-NEXT:  1      5     0.50    *                   testl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testl	%esi, %edi
+# CHECK-NEXT:  1      5     0.50    *                   testl	%esi, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        testq	$665536, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   testq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testq	$7, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   testq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        testq	%rsi, %rdi
+# CHECK-NEXT:  1      5     0.50    *                   testq	%rsi, (%rax)
+# CHECK-NEXT:  1      100   0.50    *             U     ud2
+# CHECK-NEXT:  4      2     1.00                        xaddb	%bl, %cl
+# CHECK-NEXT:  4      6     1.00    *      *            xaddb	%bl, (%rcx)
+# CHECK-NEXT:  4      2     1.00                        xaddw	%bx, %cx
+# CHECK-NEXT:  4      6     1.00    *      *            xaddw	%ax, (%rbx)
+# CHECK-NEXT:  4      2     1.00                        xaddl	%ebx, %ecx
+# CHECK-NEXT:  4      6     1.00    *      *            xaddl	%eax, (%rbx)
+# CHECK-NEXT:  4      2     1.00                        xaddq	%rbx, %rcx
+# CHECK-NEXT:  4      6     1.00    *      *            xaddq	%rax, (%rbx)
+# CHECK-NEXT:  2      1     1.00                        xchgb	%bl, %cl
+# CHECK-NEXT:  2      5     1.00    *      *            xchgb	%bl, (%rbx)
+# CHECK-NEXT:  2      1     1.00                        xchgw	%bx, %ax
+# CHECK-NEXT:  2      2     1.00                        xchgw	%bx, %cx
+# CHECK-NEXT:  2      5     1.00    *      *            xchgw	%ax, (%rbx)
+# CHECK-NEXT:  2      1     1.00                        xchgl	%ebx, %eax
+# CHECK-NEXT:  2      1     1.00                        xchgl	%ebx, %ecx
+# CHECK-NEXT:  2      5     1.00    *      *            xchgl	%eax, (%rbx)
+# CHECK-NEXT:  2      1     1.00                        xchgq	%rbx, %rax
+# CHECK-NEXT:  2      1     1.00                        xchgq	%rbx, %rcx
+# CHECK-NEXT:  2      5     1.00    *      *            xchgq	%rax, (%rbx)
+# CHECK-NEXT:  1      6     0.50    *                   xlatb
+# CHECK-NEXT:  1      1     0.50                        xorb	$7, %al
+# CHECK-NEXT:  1      1     0.50                        xorb	$7, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            xorb	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorb	%sil, %dil
+# CHECK-NEXT:  2      6     1.00    *      *            xorb	%sil, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   xorb	(%rax), %dil
+# CHECK-NEXT:  1      1     0.50                        xorw	$511, %ax
+# CHECK-NEXT:  1      1     0.50                        xorw	$511, %di
+# CHECK-NEXT:  2      6     1.00    *      *            xorw	$511, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorw	$7, %di
+# CHECK-NEXT:  2      6     1.00    *      *            xorw	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorw	%si, %di
+# CHECK-NEXT:  2      6     1.00    *      *            xorw	%si, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   xorw	(%rax), %di
+# CHECK-NEXT:  1      1     0.50                        xorl	$665536, %eax
+# CHECK-NEXT:  1      1     0.50                        xorl	$665536, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            xorl	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorl	$7, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            xorl	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorl	%esi, %edi
+# CHECK-NEXT:  2      6     1.00    *      *            xorl	%esi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   xorl	(%rax), %edi
+# CHECK-NEXT:  1      1     0.50                        xorq	$665536, %rax
+# CHECK-NEXT:  1      1     0.50                        xorq	$665536, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            xorq	$665536, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorq	$7, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            xorq	$7, (%rax)
+# CHECK-NEXT:  1      1     0.50                        xorq	%rsi, %rdi
+# CHECK-NEXT:  2      6     1.00    *      *            xorq	%rsi, (%rax)
+# CHECK-NEXT:  1      5     0.50    *                   xorq	(%rax), %rdi
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT: 160.00  -     571.83 221.33 222.00 571.83 316.00 316.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 262.00 262.00  -      -     246.00 547.50 622.50  -      -      -      -      -      -      -      -      -      -      -      -     64.00
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcb	$7, %al
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcb	$7, %dil
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcb	%sil, %dil
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   adcb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adcb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcw	$511, %ax
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcw	$511, %di
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcw	$7, %di
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcw	%si, %di
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   adcw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adcw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcl	$665536, %eax
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcl	$665536, %edi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcl	$7, %edi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcl	%esi, %edi
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   adcl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adcl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcq	$665536, %rax
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcq	$665536, %rdi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcq	$7, %rdi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   adcq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     adcq	%rsi, %rdi
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   adcq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   adcq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   addb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   addw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   addl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     addq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   addq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   addq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   andb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   andw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   andl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     andq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   andq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   andq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bsfw	%si, %di
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bsrw	%si, %di
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   bsfw	(%rax), %di
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   bsrw	(%rax), %di
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bsfl	%esi, %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bsrl	%esi, %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   bsfl	(%rax), %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   bsrl	(%rax), %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bsfq	%rsi, %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bsrq	%rsi, %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   bsfq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   bsrq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     bswapl	%eax
-# CHECK-NEXT:  -      -     0.50   1.00    -     0.50    -      -     bswapq	%rax
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btw	%si, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btcw	%si, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btrw	%si, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btsw	%si, %di
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btcw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btrw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btsw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btcw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btrw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btsw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   btw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btcw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btrw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btsw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btl	%esi, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btcl	%esi, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btrl	%esi, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btsl	%esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btcl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btrl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btsl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btl	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btcl	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btrl	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btsl	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   btl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btcl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btrl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btsl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btcq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btrq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btsq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btcq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btrq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   1.83   1.00   1.00   btsq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btcq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btrq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     btsq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50   0.50   0.50   btq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btcq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btrq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   btsq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cbtw
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cwtl
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cltq
-# CHECK-NEXT:  -      -     1.33   0.33    -     0.33    -      -     cwtd
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     cltd
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     cqto
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     clc
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cld
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmc
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cmpq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   cmpq	(%rax), %rdi
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   1.00   1.00   cmpsb	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   1.00   1.00   cmpsw	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   1.00   1.00   cmpsl	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00   1.00   1.00   cmpsq	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  -      -     1.50   1.00    -     1.50    -      -     cmpxchgb	%cl, %bl
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   2.33   1.00   1.00   cmpxchgb	%cl, (%rbx)
-# CHECK-NEXT:  -      -     1.50   1.00    -     1.50    -      -     cmpxchgw	%cx, %bx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   2.33   1.00   1.00   cmpxchgw	%cx, (%rbx)
-# CHECK-NEXT:  -      -     1.50   1.00    -     1.50    -      -     cmpxchgl	%ecx, %ebx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   2.33   1.00   1.00   cmpxchgl	%ecx, (%rbx)
-# CHECK-NEXT:  -      -     1.50   1.00    -     1.50    -      -     cmpxchgq	%rcx, %rbx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   2.33   1.00   1.00   cmpxchgq	%rcx, (%rbx)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     cpuid
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     decb	%dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   decb	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     decw	%di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   decw	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     decl	%edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   decl	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     decq	%rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   decq	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     divb	%dil
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   divb	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     divw	%si
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   divw	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     divl	%edx
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   divl	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     divq	%rcx
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   divq	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     idivb	%dil
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   idivb	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     idivw	%si
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   idivw	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     idivl	%edx
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   idivl	(%rax)
-# CHECK-NEXT: 10.00   -     1.00    -      -      -      -      -     idivq	%rcx
-# CHECK-NEXT: 10.00   -     1.00    -      -      -     0.50   0.50   idivq	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imulb	%dil
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imulb	(%rax)
-# CHECK-NEXT:  -      -     1.17   1.67    -     1.17    -      -     imulw	%di
-# CHECK-NEXT:  -      -     1.17   1.67    -     1.17   0.50   0.50   imulw	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imulw	%si, %di
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imulw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   1.33    -     0.33    -      -     imulw	$511, %si, %di
-# CHECK-NEXT:  -      -     0.33   1.33    -     0.33   0.50   0.50   imulw	$511, (%rax), %di
-# CHECK-NEXT:  -      -     0.33   1.33    -     0.33    -      -     imulw	$7, %si, %di
-# CHECK-NEXT:  -      -     0.33   1.33    -     0.33   0.50   0.50   imulw	$7, (%rax), %di
-# CHECK-NEXT:  -      -     0.83   1.33    -     0.83    -      -     imull	%edi
-# CHECK-NEXT:  -      -     0.83   1.33    -     0.83   0.50   0.50   imull	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imull	%esi, %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imull	(%rax), %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imull	$665536, %esi, %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imull	$665536, (%rax), %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imull	$7, %esi, %edi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imull	$7, (%rax), %edi
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     imulq	%rdi
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   imulq	(%rax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imulq	%rsi, %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imulq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imulq	$665536, %rsi, %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imulq	$665536, (%rax), %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     imulq	$7, %rsi, %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   imulq	$7, (%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     inb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     inb	%dx, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     inw	$7, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     inw	%dx, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     inl	$7, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     inl	%dx, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     incb	%dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   incb	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     incw	%di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   incw	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     incl	%edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   incl	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     incq	%rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   incq	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     insb	%dx, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     insw	%dx, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     insl	%dx, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     int	$7
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     lahf
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67   0.50   0.50   lodsb	(%rsi), %al
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67   0.50   0.50   lodsw	(%rsi), %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsl	(%rsi), %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   lodsq	(%rsi), %rax
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsb	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsw	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsl	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   movsq	(%rsi), %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movsbw	%al, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movzbw	%al, %di
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movsbw	(%rax), %di
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movzbw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movsbl	%al, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movzbl	%al, %edi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movsbl	(%rax), %edi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movzbl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movsbq	%al, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movzbq	%al, %rdi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movsbq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movzbq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movswl	%ax, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movzwl	%ax, %edi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movswl	(%rax), %edi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movzwl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movswq	%ax, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movzwq	%ax, %rdi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movswq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movzwq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     movslq	%eax, %rdi
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   movslq	(%rax), %rdi
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     mulb	%dil
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   mulb	(%rax)
-# CHECK-NEXT:  -      -     1.17   1.67    -     1.17    -      -     mulw	%si
-# CHECK-NEXT:  -      -     1.17   1.67    -     1.17   0.50   0.50   mulw	(%rax)
-# CHECK-NEXT:  -      -     0.83   1.33    -     0.83    -      -     mull	%edx
-# CHECK-NEXT:  -      -     0.83   1.33    -     0.83   0.50   0.50   mull	(%rax)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     mulq	%rcx
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   mulq	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     negb	%dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   negb	(%r8)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     negw	%si
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   negw	(%r9)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     negl	%edx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   negl	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     negq	%rcx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   negq	(%r10)
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nop
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nopw	%di
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nopw	(%rcx)
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nopl	%esi
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nopl	(%r8)
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nopq	%rdx
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     nopq	(%r9)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     notb	%dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   notb	(%r8)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     notw	%si
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   notw	(%r9)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     notl	%edx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   notl	(%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     notq	%rcx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   notq	(%r10)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   orb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   orw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   orl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     orq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   orq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   orq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outb	%al, $7
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outb	%al, %dx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outw	%ax, $7
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outw	%ax, %dx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outl	%eax, $7
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outl	%eax, %dx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outsb	(%rsi), %dx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outsw	(%rsi), %dx
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     outsl	(%rsi), %dx
-# CHECK-NEXT:  -      -     1.00   1.00    -     2.00    -      -     pause
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rclb	%dil
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rcrb	%dil
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclb	(%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrb	(%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rclb	$7, %dil
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrb	$7, %dil
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclb	$7, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrb	$7, (%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rclb	%cl, %dil
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrb	%cl, %dil
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclb	%cl, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrb	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rclw	%di
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rcrw	%di
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclw	(%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrw	(%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rclw	$7, %di
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrw	$7, %di
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclw	$7, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrw	$7, (%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rclw	%cl, %di
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrw	%cl, %di
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclw	%cl, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrw	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rcll	%edi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rcrl	%edi
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcll	(%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrl	(%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcll	$7, %edi
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrl	$7, %edi
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcll	$7, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrl	$7, (%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcll	%cl, %edi
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrl	%cl, %edi
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcll	%cl, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrl	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rclq	%rdi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rcrq	%rdi
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclq	(%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrq	(%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rclq	$7, %rdi
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrq	$7, %rdi
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclq	$7, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrq	$7, (%rax)
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rclq	%cl, %rdi
-# CHECK-NEXT:  -      -     4.00    -      -     4.00    -      -     rcrq	%cl, %rdi
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rclq	%cl, (%rax)
-# CHECK-NEXT:  -      -     3.50    -      -     3.50   2.00   2.00   rcrq	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rolb	%dil
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorb	%dil
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rolb	(%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorb	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rolb	$7, %dil
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorb	$7, %dil
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rolb	$7, (%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorb	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rolb	%cl, %dil
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rorb	%cl, %dil
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rolb	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rorb	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rolw	%di
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorw	%di
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rolw	(%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorw	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rolw	$7, %di
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorw	$7, %di
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rolw	$7, (%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorw	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rolw	%cl, %di
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rorw	%cl, %di
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rolw	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rorw	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     roll	%edi
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorl	%edi
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   roll	(%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorl	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     roll	$7, %edi
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorl	$7, %edi
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   roll	$7, (%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorl	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     roll	%cl, %edi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rorl	%cl, %edi
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   roll	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rorl	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rolq	%rdi
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorq	%rdi
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rolq	(%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorq	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rolq	$7, %rdi
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     rorq	$7, %rdi
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rolq	$7, (%rax)
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   1.00   1.00   rorq	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rolq	%cl, %rdi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     rorq	%cl, %rdi
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rolq	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   rorq	%cl, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sahf
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarb	%dil
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shlb	%dil
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrb	%dil
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarb	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shlb	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrb	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarb	$7, %dil
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shlb	$7, %dil
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrb	$7, %dil
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shlb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrb	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     sarb	%cl, %dil
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shlb	%cl, %dil
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shrb	%cl, %dil
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   sarb	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shlb	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shrb	%cl, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarw	%di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shlw	%di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrw	%di
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarw	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shlw	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrw	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shlw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrw	$7, %di
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shlw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrw	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     sarw	%cl, %di
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shlw	%cl, %di
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shrw	%cl, %di
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   sarw	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shlw	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shrw	%cl, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarl	%edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shll	%edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrl	%edi
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarl	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shll	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrl	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarl	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shll	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrl	$7, %edi
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shll	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrl	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     sarl	%cl, %edi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shll	%cl, %edi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shrl	%cl, %edi
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   sarl	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shll	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shrl	%cl, (%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarq	%rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shlq	%rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrq	%rdi
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarq	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shlq	(%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrq	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sarq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shlq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     shrq	$7, %rdi
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   sarq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shlq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   1.00   1.00   shrq	$7, (%rax)
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     sarq	%cl, %rdi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shlq	%cl, %rdi
-# CHECK-NEXT:  -      -     1.50    -      -     1.50    -      -     shrq	%cl, %rdi
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   sarq	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shlq	%cl, (%rax)
-# CHECK-NEXT:  -      -     1.50    -     1.00   1.50   1.00   1.00   shrq	%cl, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbb	$7, %al
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbb	$7, %dil
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbb	%sil, %dil
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   sbbb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   sbbb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbw	$511, %ax
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbw	$511, %di
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbw	$7, %di
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbw	%si, %di
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   sbbw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   sbbw	(%rax), %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbl	$665536, %eax
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbl	$665536, %edi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbl	$7, %edi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbl	%esi, %edi
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   sbbl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   sbbl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbq	$665536, %rax
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbq	$665536, %rdi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbq	$7, %rdi
-# CHECK-NEXT:  -      -     1.00   1.00   1.00   1.00   1.00   1.00   sbbq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     sbbq	%rsi, %rdi
-# CHECK-NEXT:  -      -     1.33   0.33   1.00   1.33   1.00   1.00   sbbq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83   0.50   0.50   sbbq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67    -      -     scasb	%es:(%rdi), %al
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67    -      -     scasw	%es:(%rdi), %ax
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67    -      -     scasl	%es:(%rdi), %eax
-# CHECK-NEXT:  -      -     0.67   0.67    -     0.67    -      -     scasq	%es:(%rdi), %rax
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     seto	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   seto	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setno	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setno	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setb	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setb	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setae	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setae	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sete	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   sete	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setne	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setne	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     seta	%al
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   0.50   0.50   seta	(%rax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     setbe	%al
-# CHECK-NEXT:  -      -     1.00    -     1.00   1.00   0.50   0.50   setbe	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     sets	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   sets	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setns	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setns	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setp	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setp	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setnp	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setnp	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setl	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setl	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setge	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setge	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setg	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setg	(%rax)
-# CHECK-NEXT:  -      -     0.50    -      -     0.50    -      -     setle	%al
-# CHECK-NEXT:  -      -     0.50    -     1.00   0.50   0.50   0.50   setle	(%rax)
-# CHECK-NEXT:  -      -     1.83   0.33    -     1.83    -      -     shldw	%cl, %si, %di
-# CHECK-NEXT:  -      -     1.83   0.33    -     1.83    -      -     shrdw	%cl, %si, %di
-# CHECK-NEXT:  -      -     1.83   0.33   1.00   1.83   1.00   1.00   shldw	%cl, %si, (%rax)
-# CHECK-NEXT:  -      -     1.83   0.33   1.00   1.83   1.00   1.00   shrdw	%cl, %si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     shldw	$7, %si, %di
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     shrdw	$7, %si, %di
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   0.83   1.00   1.00   shldw	$7, %si, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   0.83   1.00   1.00   shrdw	$7, %si, (%rax)
-# CHECK-NEXT:  -      -     1.83   0.33    -     1.83    -      -     shldl	%cl, %esi, %edi
-# CHECK-NEXT:  -      -     1.83   0.33    -     1.83    -      -     shrdl	%cl, %esi, %edi
-# CHECK-NEXT:  -      -     1.83   0.33   1.00   1.83   1.00   1.00   shldl	%cl, %esi, (%rax)
-# CHECK-NEXT:  -      -     1.83   0.33   1.00   1.83   1.00   1.00   shrdl	%cl, %esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     shldl	$7, %esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     shrdl	$7, %esi, %edi
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   0.83   1.00   1.00   shldl	$7, %esi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   0.83   1.00   1.00   shrdl	$7, %esi, (%rax)
-# CHECK-NEXT:  -      -     1.83   0.33    -     1.83    -      -     shldq	%cl, %rsi, %rdi
-# CHECK-NEXT:  -      -     1.83   0.33    -     1.83    -      -     shrdq	%cl, %rsi, %rdi
-# CHECK-NEXT:  -      -     1.83   0.33   1.00   1.83   1.00   1.00   shldq	%cl, %rsi, (%rax)
-# CHECK-NEXT:  -      -     1.83   0.33   1.00   1.83   1.00   1.00   shrdq	%cl, %rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     shldq	$7, %rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33    -     0.83    -      -     shrdq	$7, %rsi, %rdi
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   0.83   1.00   1.00   shldq	$7, %rsi, (%rax)
-# CHECK-NEXT:  -      -     0.83   0.33   1.00   0.83   1.00   1.00   shrdq	$7, %rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     stc
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     std
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   0.50   0.50   stosb	%al, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   0.50   0.50   stosw	%ax, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   0.50   0.50   stosl	%eax, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   0.50   0.50   stosq	%rax, %es:(%rdi)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   subb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   subw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   subl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     subq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   subq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   subq	(%rax), %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     testq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   testq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     ud2
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xaddb	%bl, %cl
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   xaddb	%bl, (%rcx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xaddw	%bx, %cx
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   xaddw	%ax, (%rbx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xaddl	%ebx, %ecx
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   xaddl	%eax, (%rbx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xaddq	%rbx, %rcx
-# CHECK-NEXT:  -      -     0.67   0.67   1.00   0.67   1.00   1.00   xaddq	%rax, (%rbx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgb	%bl, %cl
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xchgb	%bl, (%rbx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgw	%bx, %ax
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgw	%bx, %cx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xchgw	%ax, (%rbx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgl	%ebx, %eax
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgl	%ebx, %ecx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xchgl	%eax, (%rbx)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgq	%rbx, %rax
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     xchgq	%rbx, %rcx
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xchgq	%rax, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   xlatb
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorb	$7, %al
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorb	$7, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorb	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorb	%sil, %dil
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorb	%sil, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   xorb	(%rax), %dil
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorw	$511, %ax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorw	$511, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorw	$511, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorw	$7, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorw	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorw	%si, %di
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorw	%si, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   xorw	(%rax), %di
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorl	$665536, %eax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorl	$665536, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorl	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorl	$7, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorl	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorl	%esi, %edi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorl	%esi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   xorl	(%rax), %edi
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorq	$665536, %rax
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorq	$665536, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorq	$665536, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorq	$7, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorq	$7, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     xorq	%rsi, %rdi
-# CHECK-NEXT:  -      -     0.33   0.33   1.00   0.33   1.00   1.00   xorq	%rsi, (%rax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33   0.50   0.50   xorq	(%rax), %rdi
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     adcq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     andq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsfw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsrw	%si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsfw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsrw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsfl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsrl	%esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsfl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsrl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsfq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsrq	%rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsfq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     2.00   2.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bsrq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bswapl	%eax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     bswapq	%rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrw	%si, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsw	%si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btw	%si, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcw	%si, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrw	%si, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsw	%si, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsw	$7, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrl	%esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsl	%esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btl	%esi, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcl	%esi, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrl	%esi, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsl	%esi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btl	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcl	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrl	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsl	$7, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btl	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcl	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrl	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrq	%rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsq	%rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btq	%rsi, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcq	%rsi, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrq	%rsi, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsq	%rsi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsq	$7, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btcq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btrq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     btsq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cbtw
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cwtl
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cltq
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cwtd
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cltd
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cqto
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     clc
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cld
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmc
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpb	$7, %dil
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpb	%sil, %dil
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	$511, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	$7, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	%si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	$665536, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	$7, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	%esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	$665536, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	$7, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	%rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpsb	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpsw	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpsl	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpsq	%es:(%rdi), (%rsi)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgb	%cl, %bl
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgb	%cl, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgw	%cx, %bx
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgw	%cx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgl	%ecx, %ebx
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgl	%ecx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgq	%rcx, %rbx
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     cmpxchgq	%rcx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     cpuid
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decw	%di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decl	%edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decq	%rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     decq	(%rax)
+# CHECK-NEXT:  -      -      -      -     12.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divb	%dil
+# CHECK-NEXT: 0.50   0.50    -      -     12.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divb	(%rax)
+# CHECK-NEXT:  -      -      -      -     15.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divw	%si
+# CHECK-NEXT: 0.50   0.50    -      -     15.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divw	(%rax)
+# CHECK-NEXT:  -      -      -      -     14.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divl	%edx
+# CHECK-NEXT: 0.50   0.50    -      -     14.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divl	(%rax)
+# CHECK-NEXT:  -      -      -      -     14.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divq	%rcx
+# CHECK-NEXT: 0.50   0.50    -      -     14.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     divq	(%rax)
+# CHECK-NEXT:  -      -      -      -     12.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivb	%dil
+# CHECK-NEXT: 0.50   0.50    -      -     12.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivb	(%rax)
+# CHECK-NEXT:  -      -      -      -     17.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivw	%si
+# CHECK-NEXT: 0.50   0.50    -      -     17.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivw	(%rax)
+# CHECK-NEXT:  -      -      -      -     25.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivl	%edx
+# CHECK-NEXT: 0.50   0.50    -      -     25.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivl	(%rax)
+# CHECK-NEXT:  -      -      -      -     14.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivq	%rcx
+# CHECK-NEXT: 0.50   0.50    -      -     14.00   -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     idivq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulb	%dil
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	%di
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	%si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	$511, %si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	$511, (%rax), %di
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	$7, %si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imulw	$7, (%rax), %di
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	%edi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	%esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	$665536, %esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	$665536, (%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	$7, %esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   imull	$7, (%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	%rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	%rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	$665536, %rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	$665536, (%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	$7, %rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   imulq	$7, (%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     inb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     inb	%dx, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     inw	$7, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     inw	%dx, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     inl	$7, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     inl	%dx, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incw	%di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incl	%edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incq	%rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     incq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     insb	%dx, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     insw	%dx, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     insl	%dx, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     int	$7
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lahf
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lodsb	(%rsi), %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lodsw	(%rsi), %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lodsl	(%rsi), %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     lodsq	(%rsi), %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsb	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsw	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsl	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsq	(%rsi), %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsbw	%al, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzbw	%al, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsbw	(%rax), %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzbw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsbl	%al, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzbl	%al, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsbl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzbl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsbq	%al, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzbq	%al, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movsbq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzbq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movswl	%ax, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzwl	%ax, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movswl	(%rax), %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzwl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movswq	%ax, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzwq	%ax, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movswq	(%rax), %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movzwq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movslq	%eax, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     movslq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   mulb	%dil
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   mulb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   mulw	%si
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   mulw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   mull	%edx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00   mull	(%rax)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   mulq	%rcx
+# CHECK-NEXT: 0.50   0.50    -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     4.00   mulq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negb	(%r8)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negw	%si
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negw	(%r9)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negl	%edx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negq	%rcx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     negq	(%r10)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nop
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nopw	%di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nopw	(%rcx)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nopl	%esi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nopl	(%r8)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nopq	%rdx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     nopq	(%r9)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notb	(%r8)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notw	%si
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notw	(%r9)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notl	%edx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notq	%rcx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     notq	(%r10)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     orq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outb	%al, $7
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outb	%al, %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outw	%ax, $7
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outw	%ax, %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outl	%eax, $7
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outl	%eax, %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outsb	(%rsi), %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outsw	(%rsi), %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     outsl	(%rsi), %dx
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     pause
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclb	%dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclb	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclb	$7, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclb	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclb	%cl, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrb	%cl, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclb	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrb	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclw	%di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrw	%di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclw	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclw	%cl, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrw	%cl, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclw	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrw	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcll	%edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrl	%edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcll	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcll	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcll	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcll	%cl, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrl	%cl, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcll	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrl	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclq	%rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrq	%rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclq	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclq	%cl, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrq	%cl, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rclq	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rcrq	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolb	%dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolb	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolb	$7, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolb	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolb	%cl, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorb	%cl, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolb	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorb	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolw	%di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorw	%di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolw	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolw	%cl, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorw	%cl, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolw	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorw	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     roll	%edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorl	%edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     roll	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     roll	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     roll	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     roll	%cl, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorl	%cl, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     roll	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorl	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolq	%rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorq	%rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolq	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolq	%cl, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorq	%cl, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rolq	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     rorq	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sahf
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarb	%dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlb	%dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrb	%dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarb	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlb	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarb	$7, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlb	$7, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarb	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlb	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarb	%cl, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlb	%cl, %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrb	%cl, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarb	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlb	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrb	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarw	%di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlw	%di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrw	%di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarw	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlw	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrw	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlw	$7, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlw	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarw	%cl, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlw	%cl, %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrw	%cl, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarw	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlw	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrw	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarl	%edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shll	%edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrl	%edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarl	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shll	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarl	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shll	$7, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarl	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shll	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarl	%cl, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shll	%cl, %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrl	%cl, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarl	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shll	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrl	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarq	%rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlq	%rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrq	%rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarq	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlq	(%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrq	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlq	$7, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlq	$7, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarq	%cl, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlq	%cl, %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrq	%cl, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sarq	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shlq	%cl, (%rax)
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     shrq	%cl, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     sbbq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     scasb	%es:(%rdi), %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     scasw	%es:(%rdi), %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     scasl	%es:(%rdi), %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     scasq	%es:(%rdi), %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     seto	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     seto	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setno	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setno	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setb	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setb	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setae	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setae	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sete	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sete	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setne	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setne	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     seta	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     seta	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setbe	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setbe	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sets	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     sets	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setns	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setns	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setp	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setp	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setnp	%al
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setnp	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setl	%al
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     setl	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setge	%al
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     setge	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setg	%al
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     setg	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     setle	%al
+# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     setle	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shldw	%cl, %si, %di
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shrdw	%cl, %si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shldw	%cl, %si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shrdw	%cl, %si, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shldw	$7, %si, %di
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shrdw	$7, %si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shldw	$7, %si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shrdw	$7, %si, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shldl	%cl, %esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shrdl	%cl, %esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shldl	%cl, %esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shrdl	%cl, %esi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shldl	$7, %esi, %edi
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shrdl	$7, %esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shldl	$7, %esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shrdl	$7, %esi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shldq	%cl, %rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     4.00   4.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shrdq	%cl, %rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shldq	%cl, %rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shrdq	%cl, %rsi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shldq	$7, %rsi, %rdi
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -      -      -      -      -      -      -     shrdq	$7, %rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shldq	$7, %rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     11.00  11.00   -      -      -      -      -      -      -      -      -      -      -      -      -     shrdq	$7, %rsi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     stc
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     std
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     stosb	%al, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     stosw	%ax, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     stosl	%eax, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     stosq	%rax, %es:(%rdi)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     subq	(%rax), %rdi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testb	$7, %dil
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testb	%sil, %dil
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testb	%sil, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	$511, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	$7, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	%si, %di
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testw	%si, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	$665536, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	$7, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	%esi, %edi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testl	%esi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	$665536, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	$7, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	%rsi, %rdi
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     testq	%rsi, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     ud2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddb	%bl, %cl
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddb	%bl, (%rcx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddw	%bx, %cx
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddw	%ax, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddl	%ebx, %ecx
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddl	%eax, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddq	%rbx, %rcx
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xaddq	%rax, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgb	%bl, %cl
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgb	%bl, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgw	%bx, %ax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgw	%bx, %cx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgw	%ax, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgl	%ebx, %eax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgl	%ebx, %ecx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgl	%eax, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgq	%rbx, %rax
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgq	%rbx, %rcx
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xchgq	%rax, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xlatb
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorb	$7, %al
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorb	$7, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorb	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorb	%sil, %dil
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorb	%sil, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorb	(%rax), %dil
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	$511, %ax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	$511, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	$511, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	$7, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	%si, %di
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	%si, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorw	(%rax), %di
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	$665536, %eax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	$665536, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	$7, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	%esi, %edi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	%esi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	(%rax), %edi
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	$665536, %rax
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	$665536, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	$665536, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	$7, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	$7, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	%rsi, %rdi
+# CHECK-NEXT: 1.00   1.00    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	%rsi, (%rax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	(%rax), %rdi
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x87.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x87.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-x87.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 f2xm1
 
@@ -206,316 +206,328 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      100   0.33                  U     f2xm1
+# CHECK-NEXT:  1      100   0.50                  U     f2xm1
 # CHECK-NEXT:  1      1     1.00                  U     fabs
-# CHECK-NEXT:  1      3     1.00                  U     fadd	%st(0), %st(1)
-# CHECK-NEXT:  1      3     1.00                  U     fadd	%st(2)
-# CHECK-NEXT:  2      10    1.00    *             U     fadds	(%ecx)
-# CHECK-NEXT:  2      10    1.00    *             U     faddl	(%ecx)
-# CHECK-NEXT:  1      3     1.00                  U     faddp	%st(1)
-# CHECK-NEXT:  1      3     1.00                  U     faddp	%st(2)
-# CHECK-NEXT:  3      13    2.00    *             U     fiadds	(%ecx)
-# CHECK-NEXT:  3      13    2.00    *             U     fiaddl	(%ecx)
-# CHECK-NEXT:  1      100   0.33                  U     fbld	(%ecx)
-# CHECK-NEXT:  1      100   0.33                  U     fbstp	(%eax)
+# CHECK-NEXT:  1      5     1.00                  U     fadd	%st(0), %st(1)
+# CHECK-NEXT:  1      5     1.00                  U     fadd	%st(2)
+# CHECK-NEXT:  1      10    1.00    *             U     fadds	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     faddl	(%ecx)
+# CHECK-NEXT:  1      5     1.00                  U     faddp	%st(1)
+# CHECK-NEXT:  1      5     1.00                  U     faddp	%st(2)
+# CHECK-NEXT:  1      10    1.00    *             U     fiadds	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fiaddl	(%ecx)
+# CHECK-NEXT:  1      100   0.50                  U     fbld	(%ecx)
+# CHECK-NEXT:  1      100   0.50                  U     fbstp	(%eax)
 # CHECK-NEXT:  1      1     1.00                  U     fchs
-# CHECK-NEXT:  1      100   0.33                  U     fnclex
-# CHECK-NEXT:  3      3     2.00                  U     fcmovb	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmovbe	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmove	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmovnb	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmovnbe	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmovne	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmovnu	%st(1), %st(0)
-# CHECK-NEXT:  3      3     2.00                  U     fcmovu	%st(1), %st(0)
-# CHECK-NEXT:  1      1     1.00                  U     fcom	%st(1)
-# CHECK-NEXT:  1      1     1.00                  U     fcom	%st(3)
-# CHECK-NEXT:  2      8     1.00                  U     fcoms	(%ecx)
-# CHECK-NEXT:  2      8     1.00                  U     fcoml	(%eax)
-# CHECK-NEXT:  1      1     1.00                  U     fcomp	%st(1)
-# CHECK-NEXT:  1      1     1.00                  U     fcomp	%st(3)
-# CHECK-NEXT:  2      8     1.00                  U     fcomps	(%ecx)
-# CHECK-NEXT:  2      8     1.00                  U     fcompl	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     fcompp
-# CHECK-NEXT:  3      3     1.00                  U     fcomi	%st(3)
-# CHECK-NEXT:  3      3     1.00                  U     fcompi	%st(3)
-# CHECK-NEXT:  1      100   0.33                  U     fcos
-# CHECK-NEXT:  1      1     1.00                  U     fdecstp
-# CHECK-NEXT:  1      14    14.00                 U     fdiv	%st(0), %st(1)
-# CHECK-NEXT:  1      14    14.00                 U     fdiv	%st(2)
-# CHECK-NEXT:  2      31    1.00    *             U     fdivs	(%ecx)
-# CHECK-NEXT:  2      31    1.00    *             U     fdivl	(%eax)
-# CHECK-NEXT:  1      14    14.00                 U     fdivp	%st(1)
-# CHECK-NEXT:  1      14    14.00                 U     fdivp	%st(2)
-# CHECK-NEXT:  3      34    1.00    *             U     fidivs	(%ecx)
-# CHECK-NEXT:  3      34    1.00    *             U     fidivl	(%eax)
-# CHECK-NEXT:  1      14    14.00                 U     fdivr	%st(0), %st(1)
-# CHECK-NEXT:  1      14    14.00                 U     fdivr	%st(2)
-# CHECK-NEXT:  2      31    1.00    *             U     fdivrs	(%ecx)
-# CHECK-NEXT:  2      31    1.00    *             U     fdivrl	(%eax)
-# CHECK-NEXT:  1      14    14.00                 U     fdivrp	%st(1)
-# CHECK-NEXT:  1      14    14.00                 U     fdivrp	%st(2)
-# CHECK-NEXT:  3      34    1.00    *             U     fidivrs	(%ecx)
-# CHECK-NEXT:  3      34    1.00    *             U     fidivrl	(%eax)
-# CHECK-NEXT:  1      1     1.00                  U     ffree	%st(0)
-# CHECK-NEXT:  3      11    2.00                  U     ficoms	(%ecx)
-# CHECK-NEXT:  3      11    2.00                  U     ficoml	(%eax)
-# CHECK-NEXT:  3      11    2.00                  U     ficomps	(%ecx)
-# CHECK-NEXT:  3      11    2.00                  U     ficompl	(%eax)
-# CHECK-NEXT:  2      10    1.00    *             U     filds	(%edx)
-# CHECK-NEXT:  2      10    1.00    *             U     fildl	(%ecx)
-# CHECK-NEXT:  2      10    1.00    *             U     fildll	(%eax)
-# CHECK-NEXT:  1      1     1.00                  U     fincstp
-# CHECK-NEXT:  4      5     1.33                  U     fninit
-# CHECK-NEXT:  4      9     1.00           *      U     fists	(%edx)
-# CHECK-NEXT:  4      9     1.00           *      U     fistl	(%ecx)
-# CHECK-NEXT:  4      9     1.00           *      U     fistps	(%edx)
-# CHECK-NEXT:  4      9     1.00           *      U     fistpl	(%ecx)
-# CHECK-NEXT:  4      9     1.00           *      U     fistpll	(%eax)
-# CHECK-NEXT:  3      5     1.00           *      U     fisttps	(%edx)
-# CHECK-NEXT:  3      5     1.00           *      U     fisttpl	(%ecx)
-# CHECK-NEXT:  3      5     1.00           *      U     fisttpll	(%eax)
-# CHECK-NEXT:  1      1     1.00                  U     fld	%st(0)
-# CHECK-NEXT:  3      9     1.00    *             U     flds	(%edx)
-# CHECK-NEXT:  3      9     1.00    *             U     fldl	(%ecx)
-# CHECK-NEXT:  3      9     1.00    *             U     fldt	(%eax)
-# CHECK-NEXT:  5      8     2.00    *             U     fldcw	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     fldenv	(%eax)
-# CHECK-NEXT:  2      1     1.00                  U     fld1
-# CHECK-NEXT:  2      1     1.00                  U     fldl2e
-# CHECK-NEXT:  2      1     1.00                  U     fldl2t
-# CHECK-NEXT:  2      1     1.00                  U     fldlg2
-# CHECK-NEXT:  2      1     1.00                  U     fldln2
-# CHECK-NEXT:  2      1     1.00                  U     fldpi
-# CHECK-NEXT:  1      1     1.00                  U     fldz
+# CHECK-NEXT:  1      100   0.50                  U     fnclex
+# CHECK-NEXT:  1      1     1.00                  U     fcmovb	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmovbe	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmove	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmovnb	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmovnbe	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmovne	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmovnu	%st(1), %st(0)
+# CHECK-NEXT:  1      1     1.00                  U     fcmovu	%st(1), %st(0)
+# CHECK-NEXT:  2      1     1.00                  U     fcom	%st(1)
+# CHECK-NEXT:  2      1     1.00                  U     fcom	%st(3)
+# CHECK-NEXT:  1      6     1.00                  U     fcoms	(%ecx)
+# CHECK-NEXT:  1      6     1.00                  U     fcoml	(%eax)
+# CHECK-NEXT:  2      1     1.00                  U     fcomp	%st(1)
+# CHECK-NEXT:  2      1     1.00                  U     fcomp	%st(3)
+# CHECK-NEXT:  1      6     1.00                  U     fcomps	(%ecx)
+# CHECK-NEXT:  1      6     1.00                  U     fcompl	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fcompp
+# CHECK-NEXT:  2      1     1.00                  U     fcomi	%st(3)
+# CHECK-NEXT:  2      1     1.00                  U     fcompi	%st(3)
+# CHECK-NEXT:  1      100   0.50                  U     fcos
+# CHECK-NEXT:  1      100   0.50                  U     fdecstp
+# CHECK-NEXT:  1      9     9.50                  U     fdiv	%st(0), %st(1)
+# CHECK-NEXT:  1      9     9.50                  U     fdiv	%st(2)
+# CHECK-NEXT:  1      14    9.50    *             U     fdivs	(%ecx)
+# CHECK-NEXT:  1      14    9.50    *             U     fdivl	(%eax)
+# CHECK-NEXT:  1      9     9.50                  U     fdivp	%st(1)
+# CHECK-NEXT:  1      9     9.50                  U     fdivp	%st(2)
+# CHECK-NEXT:  1      14    9.50    *             U     fidivs	(%ecx)
+# CHECK-NEXT:  1      14    9.50    *             U     fidivl	(%eax)
+# CHECK-NEXT:  1      9     9.50                  U     fdivr	%st(0), %st(1)
+# CHECK-NEXT:  1      9     9.50                  U     fdivr	%st(2)
+# CHECK-NEXT:  1      14    9.50    *             U     fdivrs	(%ecx)
+# CHECK-NEXT:  1      14    9.50    *             U     fdivrl	(%eax)
+# CHECK-NEXT:  1      9     9.50                  U     fdivrp	%st(1)
+# CHECK-NEXT:  1      9     9.50                  U     fdivrp	%st(2)
+# CHECK-NEXT:  1      14    9.50    *             U     fidivrs	(%ecx)
+# CHECK-NEXT:  1      14    9.50    *             U     fidivrl	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     ffree	%st(0)
+# CHECK-NEXT:  2      6     1.00                  U     ficoms	(%ecx)
+# CHECK-NEXT:  2      6     1.00                  U     ficoml	(%eax)
+# CHECK-NEXT:  2      6     1.00                  U     ficomps	(%ecx)
+# CHECK-NEXT:  2      6     1.00                  U     ficompl	(%eax)
+# CHECK-NEXT:  1      5     0.50    *             U     filds	(%edx)
+# CHECK-NEXT:  1      5     0.50    *             U     fildl	(%ecx)
+# CHECK-NEXT:  1      5     0.50    *             U     fildll	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fincstp
+# CHECK-NEXT:  1      100   0.50                  U     fninit
+# CHECK-NEXT:  1      1     0.50           *      U     fists	(%edx)
+# CHECK-NEXT:  1      1     0.50           *      U     fistl	(%ecx)
+# CHECK-NEXT:  1      1     0.50           *      U     fistps	(%edx)
+# CHECK-NEXT:  1      1     0.50           *      U     fistpl	(%ecx)
+# CHECK-NEXT:  1      1     0.50           *      U     fistpll	(%eax)
+# CHECK-NEXT:  1      1     0.50           *      U     fisttps	(%edx)
+# CHECK-NEXT:  1      1     0.50           *      U     fisttpl	(%ecx)
+# CHECK-NEXT:  1      1     0.50           *      U     fisttpll	(%eax)
+# CHECK-NEXT:  1      1     0.50                  U     fld	%st(0)
+# CHECK-NEXT:  1      5     0.50    *             U     flds	(%edx)
+# CHECK-NEXT:  1      5     0.50    *             U     fldl	(%ecx)
+# CHECK-NEXT:  1      5     0.50    *             U     fldt	(%eax)
+# CHECK-NEXT:  1      5     0.50    *             U     fldcw	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fldenv	(%eax)
+# CHECK-NEXT:  1      3     1.00                  U     fld1
+# CHECK-NEXT:  1      3     1.00                  U     fldl2e
+# CHECK-NEXT:  1      3     1.00                  U     fldl2t
+# CHECK-NEXT:  1      3     1.00                  U     fldlg2
+# CHECK-NEXT:  1      3     1.00                  U     fldln2
+# CHECK-NEXT:  1      3     1.00                  U     fldpi
+# CHECK-NEXT:  1      3     1.00                  U     fldz
 # CHECK-NEXT:  1      5     1.00                  U     fmul	%st(0), %st(1)
 # CHECK-NEXT:  1      5     1.00                  U     fmul	%st(2)
-# CHECK-NEXT:  2      12    1.00    *             U     fmuls	(%ecx)
-# CHECK-NEXT:  2      12    1.00    *             U     fmull	(%eax)
+# CHECK-NEXT:  1      10    1.00    *             U     fmuls	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fmull	(%eax)
 # CHECK-NEXT:  1      5     1.00                  U     fmulp	%st(1)
 # CHECK-NEXT:  1      5     1.00                  U     fmulp	%st(2)
-# CHECK-NEXT:  3      15    1.00    *             U     fimuls	(%ecx)
-# CHECK-NEXT:  3      15    1.00    *             U     fimull	(%eax)
-# CHECK-NEXT:  1      1     1.00                  U     fnop
-# CHECK-NEXT:  1      100   0.33                  U     fpatan
-# CHECK-NEXT:  1      100   0.33                  U     fprem
-# CHECK-NEXT:  1      100   0.33                  U     fprem1
-# CHECK-NEXT:  1      100   0.33                  U     fptan
-# CHECK-NEXT:  1      100   0.33                  U     frndint
-# CHECK-NEXT:  1      100   0.33                  U     frstor	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     fnsave	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     fscale
-# CHECK-NEXT:  1      100   0.33                  U     fsin
-# CHECK-NEXT:  1      100   0.33                  U     fsincos
-# CHECK-NEXT:  1      24    24.00                 U     fsqrt
-# CHECK-NEXT:  1      1     1.00                  U     fst	%st(0)
-# CHECK-NEXT:  3      6     1.00           *      U     fsts	(%edx)
-# CHECK-NEXT:  3      6     1.00           *      U     fstl	(%ecx)
-# CHECK-NEXT:  1      1     1.00                  U     fstp	%st(0)
-# CHECK-NEXT:  3      6     1.00           *      U     fstpl	(%edx)
-# CHECK-NEXT:  3      6     1.00           *      U     fstpl	(%ecx)
-# CHECK-NEXT:  3      6     1.00           *      U     fstpt	(%eax)
-# CHECK-NEXT:  4      7     1.00           *      U     fnstcw	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     fnstenv	(%eax)
-# CHECK-NEXT:  4      7     1.00                  U     fnstsw	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     frstor	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     wait
-# CHECK-NEXT:  1      100   0.33                  U     fnsave	(%eax)
-# CHECK-NEXT:  1      3     1.00                  U     fsub	%st(0), %st(1)
-# CHECK-NEXT:  1      3     1.00                  U     fsub	%st(2)
-# CHECK-NEXT:  2      10    1.00    *             U     fsubs	(%ecx)
-# CHECK-NEXT:  2      10    1.00    *             U     fsubl	(%eax)
-# CHECK-NEXT:  1      3     1.00                  U     fsubp	%st(1)
-# CHECK-NEXT:  1      3     1.00                  U     fsubp	%st(2)
-# CHECK-NEXT:  3      13    2.00    *             U     fisubs	(%ecx)
-# CHECK-NEXT:  3      13    2.00    *             U     fisubl	(%eax)
-# CHECK-NEXT:  1      3     1.00                  U     fsubr	%st(0), %st(1)
-# CHECK-NEXT:  1      3     1.00                  U     fsubr	%st(2)
-# CHECK-NEXT:  2      10    1.00    *             U     fsubrs	(%ecx)
-# CHECK-NEXT:  2      10    1.00    *             U     fsubrl	(%eax)
-# CHECK-NEXT:  1      3     1.00                  U     fsubrp	%st(1)
-# CHECK-NEXT:  1      3     1.00                  U     fsubrp	%st(2)
-# CHECK-NEXT:  3      13    2.00    *             U     fisubrs	(%ecx)
-# CHECK-NEXT:  3      13    2.00    *             U     fisubrl	(%eax)
-# CHECK-NEXT:  1      3     1.00                  U     ftst
-# CHECK-NEXT:  1      1     1.00                  U     fucom	%st(1)
-# CHECK-NEXT:  1      1     1.00                  U     fucom	%st(3)
-# CHECK-NEXT:  1      1     1.00                  U     fucomp	%st(1)
-# CHECK-NEXT:  1      1     1.00                  U     fucomp	%st(3)
-# CHECK-NEXT:  1      3     1.00                  U     fucompp
-# CHECK-NEXT:  3      3     1.00                  U     fucomi	%st(3)
-# CHECK-NEXT:  3      3     1.00                  U     fucompi	%st(3)
-# CHECK-NEXT:  1      100   0.33                  U     wait
-# CHECK-NEXT:  1      100   0.33                  U     fxam
-# CHECK-NEXT:  1      1     0.33                  U     fxch	%st(1)
-# CHECK-NEXT:  1      1     0.33                  U     fxch	%st(3)
-# CHECK-NEXT:  5      5     2.00    *      *      U     fxrstor	(%eax)
-# CHECK-NEXT:  1      100   0.33    *      *      U     fxsave	(%eax)
-# CHECK-NEXT:  1      100   0.33                  U     fxtract
-# CHECK-NEXT:  1      100   0.33                  U     fyl2x
-# CHECK-NEXT:  1      100   0.33                  U     fyl2xp1
+# CHECK-NEXT:  1      10    1.00    *             U     fimuls	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fimull	(%eax)
+# CHECK-NEXT:  1      1     0.50                  U     fnop
+# CHECK-NEXT:  1      100   0.50                  U     fpatan
+# CHECK-NEXT:  1      100   0.50                  U     fprem
+# CHECK-NEXT:  1      100   0.50                  U     fprem1
+# CHECK-NEXT:  1      100   0.50                  U     fptan
+# CHECK-NEXT:  1      100   0.50                  U     frndint
+# CHECK-NEXT:  1      100   0.50                  U     frstor	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fnsave	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fscale
+# CHECK-NEXT:  1      100   0.50                  U     fsin
+# CHECK-NEXT:  1      100   0.50                  U     fsincos
+# CHECK-NEXT:  1      1     17.50                 U     fsqrt
+# CHECK-NEXT:  1      1     0.50                  U     fst	%st(0)
+# CHECK-NEXT:  1      1     0.50           *      U     fsts	(%edx)
+# CHECK-NEXT:  1      1     0.50           *      U     fstl	(%ecx)
+# CHECK-NEXT:  1      1     0.50                  U     fstp	%st(0)
+# CHECK-NEXT:  1      1     0.50           *      U     fstpl	(%edx)
+# CHECK-NEXT:  1      1     0.50           *      U     fstpl	(%ecx)
+# CHECK-NEXT:  1      1     0.50           *      U     fstpt	(%eax)
+# CHECK-NEXT:  1      1     0.50           *      U     fnstcw	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fnstenv	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fnstsw	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     frstor	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     wait
+# CHECK-NEXT:  1      100   0.50                  U     fnsave	(%eax)
+# CHECK-NEXT:  1      5     1.00                  U     fsub	%st(0), %st(1)
+# CHECK-NEXT:  1      5     1.00                  U     fsub	%st(2)
+# CHECK-NEXT:  1      10    1.00    *             U     fsubs	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fsubl	(%eax)
+# CHECK-NEXT:  1      5     1.00                  U     fsubp	%st(1)
+# CHECK-NEXT:  1      5     1.00                  U     fsubp	%st(2)
+# CHECK-NEXT:  1      10    1.00    *             U     fisubs	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fisubl	(%eax)
+# CHECK-NEXT:  1      5     1.00                  U     fsubr	%st(0), %st(1)
+# CHECK-NEXT:  1      5     1.00                  U     fsubr	%st(2)
+# CHECK-NEXT:  1      10    1.00    *             U     fsubrs	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fsubrl	(%eax)
+# CHECK-NEXT:  1      5     1.00                  U     fsubrp	%st(1)
+# CHECK-NEXT:  1      5     1.00                  U     fsubrp	%st(2)
+# CHECK-NEXT:  1      10    1.00    *             U     fisubrs	(%ecx)
+# CHECK-NEXT:  1      10    1.00    *             U     fisubrl	(%eax)
+# CHECK-NEXT:  1      1     1.00                  U     ftst
+# CHECK-NEXT:  2      1     1.00                  U     fucom	%st(1)
+# CHECK-NEXT:  2      1     1.00                  U     fucom	%st(3)
+# CHECK-NEXT:  2      1     1.00                  U     fucomp	%st(1)
+# CHECK-NEXT:  2      1     1.00                  U     fucomp	%st(3)
+# CHECK-NEXT:  1      1     1.00                  U     fucompp
+# CHECK-NEXT:  2      1     1.00                  U     fucomi	%st(3)
+# CHECK-NEXT:  2      1     1.00                  U     fucompi	%st(3)
+# CHECK-NEXT:  1      100   0.50                  U     wait
+# CHECK-NEXT:  1      100   0.50                  U     fxam
+# CHECK-NEXT:  1      1     0.50                  U     fxch	%st(1)
+# CHECK-NEXT:  1      1     0.50                  U     fxch	%st(3)
+# CHECK-NEXT:  1      100   0.50    *      *      U     fxrstor	(%eax)
+# CHECK-NEXT:  1      100   0.50    *      *      U     fxsave	(%eax)
+# CHECK-NEXT:  1      100   0.50                  U     fxtract
+# CHECK-NEXT:  1      100   0.50                  U     fyl2x
+# CHECK-NEXT:  1      100   0.50                  U     fyl2xp1
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -     136.00 52.67  90.67  17.00  54.67  34.00  34.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 24.00  24.00   -      -      -     36.00  20.00   -     201.50 201.50  -      -      -     7.00   48.00  40.00   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     f2xm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fabs
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fadd	%st(0), %st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fadd	%st(2)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fadds	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   faddl	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     faddp	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     faddp	%st(2)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   fiadds	(%ecx)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   fiaddl	(%ecx)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fbld	(%ecx)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fbstp	(%eax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fchs
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fnclex
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovb	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovbe	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmove	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovnb	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovnbe	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovne	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovnu	%st(1), %st(0)
-# CHECK-NEXT:  -      -     0.50    -      -     2.50    -      -     fcmovu	%st(1), %st(0)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fcom	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fcom	%st(3)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fcoms	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fcoml	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fcomp	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fcomp	%st(3)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fcomps	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fcompl	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fcompp
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     fcomi	%st(3)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     fcompi	%st(3)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fcos
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fdecstp
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdiv	%st(0), %st(1)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdiv	%st(2)
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   fdivs	(%ecx)
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   fdivl	(%eax)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdivp	%st(1)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdivp	%st(2)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   fidivs	(%ecx)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   fidivl	(%eax)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdivr	%st(0), %st(1)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdivr	%st(2)
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   fdivrs	(%ecx)
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   fdivrl	(%eax)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdivrp	%st(1)
-# CHECK-NEXT:  -     14.00  1.00    -      -      -      -      -     fdivrp	%st(2)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   fidivrs	(%ecx)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   fidivrl	(%eax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     ffree	%st(0)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   ficoms	(%ecx)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   ficoml	(%eax)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   ficomps	(%ecx)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   ficompl	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   filds	(%edx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fildl	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fildll	(%eax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fincstp
-# CHECK-NEXT:  -      -     1.00   1.00    -     2.00    -      -     fninit
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     1.00   1.00   fists	(%edx)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     1.00   1.00   fistl	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     1.00   1.00   fistps	(%edx)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     1.00   1.00   fistpl	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     1.00   1.00   fistpll	(%eax)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     0.50   0.50   fisttps	(%edx)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     0.50   0.50   fisttpl	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00   1.00    -     0.50   0.50   fisttpll	(%eax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fld	%st(0)
-# CHECK-NEXT:  -      -     0.50   0.50    -     1.00   0.50   0.50   flds	(%edx)
-# CHECK-NEXT:  -      -     0.50   0.50    -     1.00   0.50   0.50   fldl	(%ecx)
-# CHECK-NEXT:  -      -     0.50   0.50    -     1.00   0.50   0.50   fldt	(%eax)
-# CHECK-NEXT:  -      -      -      -     1.00   2.00   1.00   1.00   fldcw	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fldenv	(%eax)
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -      -     fld1
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     fldl2e
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     fldl2t
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     fldlg2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     fldln2
-# CHECK-NEXT:  -      -     1.00   1.00    -      -      -      -     fldpi
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fldz
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     fmul	%st(0), %st(1)
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     fmul	%st(2)
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   fmuls	(%ecx)
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   fmull	(%eax)
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     fmulp	%st(1)
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     fmulp	%st(2)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   fimuls	(%ecx)
-# CHECK-NEXT:  -      -     1.00   1.00    -      -     0.50   0.50   fimull	(%eax)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fnop
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fpatan
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fprem
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fprem1
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fptan
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     frndint
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     frstor	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fnsave	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fscale
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fsin
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fsincos
-# CHECK-NEXT:  -     24.00  1.00    -      -      -      -      -     fsqrt
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fst	%st(0)
-# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00   fsts	(%edx)
-# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00   fstl	(%ecx)
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     fstp	%st(0)
-# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00   fstpl	(%edx)
-# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00   fstpl	(%ecx)
-# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00   fstpt	(%eax)
-# CHECK-NEXT:  -      -      -      -     1.00   1.00   1.00   1.00   fnstcw	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fnstenv	(%eax)
-# CHECK-NEXT:  -      -     1.00    -     1.00    -     1.00   1.00   fnstsw	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     frstor	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     wait
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fnsave	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsub	%st(0), %st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsub	%st(2)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fsubs	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fsubl	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsubp	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsubp	%st(2)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   fisubs	(%ecx)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   fisubl	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsubr	%st(0), %st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsubr	%st(2)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fsubrs	(%ecx)
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   fsubrl	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsubrp	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fsubrp	%st(2)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   fisubrs	(%ecx)
-# CHECK-NEXT:  -      -      -     2.00    -      -     0.50   0.50   fisubrl	(%eax)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     ftst
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fucom	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fucom	%st(3)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fucomp	%st(1)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fucomp	%st(3)
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     fucompp
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     fucomi	%st(3)
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -     fucompi	%st(3)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     wait
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fxam
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fxch	%st(1)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fxch	%st(3)
-# CHECK-NEXT:  -      -     0.50   0.50   1.00   2.00   0.50   0.50   fxrstor	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fxsave	(%eax)
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fxtract
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fyl2x
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.33    -      -     fyl2xp1
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     f2xm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fabs
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fadd	%st(0), %st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fadd	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fadds	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     faddl	(%ecx)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     faddp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     faddp	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fiadds	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fiaddl	(%ecx)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fbld	(%ecx)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fbstp	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fchs
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnclex
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovb	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovbe	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmove	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovnb	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovnbe	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovne	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovnu	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcmovu	%st(1), %st(0)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcom	%st(1)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcom	%st(3)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fcoms	(%ecx)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fcoml	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcomp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcomp	%st(3)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fcomps	(%ecx)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fcompl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fcompp
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcomi	%st(3)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fcompi	%st(3)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fcos
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fdecstp
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdiv	%st(0), %st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdiv	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivp	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fidivs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fidivl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivr	%st(0), %st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivr	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivrs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivrl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivrp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fdivrp	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fidivrs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     9.50   9.50    -      -      -      -      -     1.00    -      -      -      -     fidivrl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     ffree	%st(0)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ficoms	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ficoml	(%eax)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ficomps	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     ficompl	(%eax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     filds	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fildl	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fildll	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fincstp
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fninit
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fists	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fistl	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fistps	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fistpl	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fistpll	(%eax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fisttps	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fisttpl	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fisttpll	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fld	%st(0)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     flds	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fldl	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fldt	(%eax)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fldcw	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fldenv	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fld1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fldl2e
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fldl2t
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fldlg2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fldln2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fldpi
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     fldz
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fmul	%st(0), %st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fmul	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fmuls	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fmull	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fmulp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fmulp	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fimuls	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fimull	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnop
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fpatan
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fprem
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fprem1
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fptan
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     frndint
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     frstor	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnsave	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fscale
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fsin
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fsincos
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     17.50  17.50   -      -      -      -      -     1.00    -      -      -      -     fsqrt
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fst	%st(0)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fsts	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fstl	(%ecx)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fstp	%st(0)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fstpl	(%edx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fstpl	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     fstpt	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnstcw	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnstenv	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnstsw	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     frstor	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     wait
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fnsave	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsub	%st(0), %st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsub	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubp	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fisubs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fisubl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubr	%st(0), %st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubr	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubrs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubrl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubrp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fsubrp	%st(2)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fisubrs	(%ecx)
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fisubrl	(%eax)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     ftst
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fucom	%st(1)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fucom	%st(3)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fucomp	%st(1)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fucomp	%st(3)
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     1.00    -      -      -      -     fucompp
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fucomi	%st(3)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     0.50   0.50    -      -      -      -     1.00    -      -      -      -      -     fucompi	%st(3)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     wait
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fxam
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fxch	%st(1)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fxch	%st(3)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fxrstor	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fxsave	(%eax)
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fxtract
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fyl2x
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     fyl2xp1
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-xop.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-xop.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/resources-xop.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -instruction-tables < %s | FileCheck %s
 
 vfrczpd %xmm0, %xmm3
 vfrczpd (%rax), %xmm3
@@ -221,314 +221,326 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vfrczpd	%xmm0, %xmm3
-# CHECK-NEXT:  2      9     1.00    *                   vfrczpd	(%rax), %xmm3
-# CHECK-NEXT:  1      3     1.00                        vfrczpd	%ymm0, %ymm3
-# CHECK-NEXT:  2      10    1.00    *                   vfrczpd	(%rax), %ymm3
-# CHECK-NEXT:  1      3     1.00                        vfrczps	%xmm0, %xmm3
-# CHECK-NEXT:  2      9     1.00    *                   vfrczps	(%rax), %xmm3
-# CHECK-NEXT:  1      3     1.00                        vfrczps	%ymm0, %ymm3
-# CHECK-NEXT:  2      10    1.00    *                   vfrczps	(%rax), %ymm3
-# CHECK-NEXT:  1      3     1.00                        vfrczsd	%xmm0, %xmm3
-# CHECK-NEXT:  2      9     1.00    *                   vfrczsd	(%rax), %xmm3
-# CHECK-NEXT:  1      3     1.00                        vfrczss	%xmm0, %xmm3
-# CHECK-NEXT:  2      9     1.00    *                   vfrczss	(%rax), %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcmov	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcmov	(%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcmov	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpcmov	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      8     1.00    *                   vpcmov	(%rax), %ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  2      8     1.00    *                   vpcmov	%ymm0, (%rax), %ymm1, %ymm3
-# CHECK-NEXT:  1      1     0.50                        vpcomb	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomb	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomd	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomd	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomq	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomq	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomub	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomub	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomud	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomud	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomuq	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomuq	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomuw	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomuw	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpcomw	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpcomw	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpermil2pd	$0, %xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpermil2pd	$0, (%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpermil2pd	$0, %xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpermil2pd	$0, %ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      8     1.00    *                   vpermil2pd	$0, (%rax), %ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  2      8     1.00    *                   vpermil2pd	$0, %ymm0, (%rax), %ymm1, %ymm3
-# CHECK-NEXT:  1      1     1.00                        vpermil2ps	$0, %xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpermil2ps	$0, (%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpermil2ps	$0, %xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpermil2ps	$0, %ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  2      8     1.00    *                   vpermil2ps	$0, (%rax), %ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  2      8     1.00    *                   vpermil2ps	$0, %ymm0, (%rax), %ymm1, %ymm3
-# CHECK-NEXT:  3      3     1.50                        vphaddbd	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddbd	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddbq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddbq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddbw	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddbw	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphadddq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphadddq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddubd	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddubd	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddubq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddubq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddubw	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddubw	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddudq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddudq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphadduwd	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphadduwd	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphadduwq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphadduwq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddwd	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddwd	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphaddwq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphaddwq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphsubbw	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphsubbw	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphsubdq	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphsubdq	(%rax), %xmm3
-# CHECK-NEXT:  3      3     1.50                        vphsubwd	%xmm0, %xmm3
-# CHECK-NEXT:  4      9     1.50    *                   vphsubwd	(%rax), %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacsdd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacsdd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacsdqh	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacsdqh	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacsdql	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacsdql	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacssdd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacssdd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacssdqh	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacssdqh	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacssdql	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacssdql	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacsswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacsswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacssww	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacssww	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmacsww	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmacsww	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmadcsswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmadcsswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      5     1.00                        vpmadcswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      11    1.00    *                   vpmadcswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpperm	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpperm	(%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     0.50    *                   vpperm	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotb	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotb	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotb	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotb	$0, %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotb	$0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotd	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotd	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotd	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotd	$0, %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotd	$0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotq	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotq	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotq	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotq	$0, %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotq	$0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotw	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotw	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotw	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vprotw	$0, %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vprotw	$0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshab	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshab	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshab	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshad	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshad	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshad	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshaq	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshaq	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshaq	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshaw	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshaw	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshaw	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshlb	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshlb	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshlb	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshld	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshld	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshld	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshlq	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshlq	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshlq	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  1      1     1.00                        vpshlw	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshlw	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  2      7     1.00    *                   vpshlw	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  2      10    1.00                        vfrczpd	%xmm0, %xmm3
+# CHECK-NEXT:  2      15    1.00    *                   vfrczpd	(%rax), %xmm3
+# CHECK-NEXT:  4      10    2.00                        vfrczpd	%ymm0, %ymm3
+# CHECK-NEXT:  8      15    2.00    *                   vfrczpd	(%rax), %ymm3
+# CHECK-NEXT:  2      10    1.00                        vfrczps	%xmm0, %xmm3
+# CHECK-NEXT:  2      15    1.00    *                   vfrczps	(%rax), %xmm3
+# CHECK-NEXT:  4      10    2.00                        vfrczps	%ymm0, %ymm3
+# CHECK-NEXT:  8      15    2.00    *                   vfrczps	(%rax), %ymm3
+# CHECK-NEXT:  2      10    1.00                        vfrczsd	%xmm0, %xmm3
+# CHECK-NEXT:  2      15    1.00    *                   vfrczsd	(%rax), %xmm3
+# CHECK-NEXT:  2      10    1.00                        vfrczss	%xmm0, %xmm3
+# CHECK-NEXT:  2      15    1.00    *                   vfrczss	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmov	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcmov	(%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcmov	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  2      2     0.50                        vpcmov	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      7     1.00    *                   vpcmov	(%rax), %ymm0, %ymm1, %ymm3
+# CHECK-NEXT:  2      7     1.00    *                   vpcmov	%ymm0, (%rax), %ymm1, %ymm3
+# CHECK-NEXT:  1      2     0.50                        vpcomb	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomb	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomd	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomd	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomq	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomq	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomub	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomub	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomud	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomud	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomuq	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomuq	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomuw	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomuw	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcomw	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vpcomw	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      3     2.00                        vpermil2pd	$0, %xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      8     2.00    *                   vpermil2pd	$0, (%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     2.00    *                   vpermil2pd	$0, %xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  2      3     3.00                        vpermil2pd	$0, %ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      8     3.00    *                   vpermil2pd	$0, (%rax), %ymm0, %ymm1, %ymm3
+# CHECK-NEXT:  2      8     3.00    *                   vpermil2pd	$0, %ymm0, (%rax), %ymm1, %ymm3
+# CHECK-NEXT:  1      3     2.00                        vpermil2ps	$0, %xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      8     2.00    *                   vpermil2ps	$0, (%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     2.00    *                   vpermil2ps	$0, %xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  2      3     3.00                        vpermil2ps	$0, %ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      8     3.00    *                   vpermil2ps	$0, (%rax), %ymm0, %ymm1, %ymm3
+# CHECK-NEXT:  2      8     3.00    *                   vpermil2ps	$0, %ymm0, (%rax), %ymm1, %ymm3
+# CHECK-NEXT:  1      2     0.50                        vphaddbd	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddbd	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddbq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddbq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddbw	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddbw	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphadddq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphadddq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddubd	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddubd	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddubq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddubq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddubw	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddubw	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddudq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddudq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphadduwd	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphadduwd	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphadduwq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphadduwq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddwd	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddwd	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphaddwq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphaddwq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphsubbw	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphsubbw	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphsubdq	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphsubdq	(%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vphsubwd	%xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vphsubwd	(%rax), %xmm3
+# CHECK-NEXT:  1      5     2.00                        vpmacsdd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    2.00    *                   vpmacsdd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     2.00                        vpmacsdqh	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    2.00    *                   vpmacsdqh	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     2.00                        vpmacsdql	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    2.00    *                   vpmacsdql	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      5     2.00                        vpmacssdd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    2.00    *                   vpmacssdd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     2.00                        vpmacssdqh	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    2.00    *                   vpmacssdqh	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     2.00                        vpmacssdql	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      10    2.00    *                   vpmacssdql	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmacsswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      9     1.00    *                   vpmacsswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmacssww	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      9     1.00    *                   vpmacssww	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmacswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      9     1.00    *                   vpmacswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmacsww	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      9     1.00    *                   vpmacsww	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmadcsswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      9     1.00    *                   vpmadcsswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      4     1.00                        vpmadcswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      9     1.00    *                   vpmadcswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      3     2.00                        vpperm	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT:  1      8     2.00    *                   vpperm	(%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     2.00    *                   vpperm	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  1      3     0.50                        vprotb	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotb	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotb	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vprotb	$0, %xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vprotb	$0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vprotd	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotd	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotd	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vprotd	$0, %xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vprotd	$0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vprotq	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotq	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotq	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vprotq	$0, %xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vprotq	$0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vprotw	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotw	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vprotw	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      2     0.50                        vprotw	$0, %xmm0, %xmm3
+# CHECK-NEXT:  1      7     0.50    *                   vprotw	$0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshab	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshab	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshab	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshad	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshad	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshad	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshaq	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshaq	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshaq	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshaw	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshaw	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshaw	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshlb	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshlb	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshlb	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshld	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshld	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshld	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshlq	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshlq	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshlq	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  1      3     0.50                        vpshlw	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshlw	(%rax), %xmm0, %xmm3
+# CHECK-NEXT:  1      8     0.50    *                   vpshlw	%xmm0, (%rax), %xmm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     68.00  68.00   -     71.00  41.50  41.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT: 41.50  41.50   -      -      -      -      -      -     30.00  30.00  60.00  60.00  36.00  12.00  100.50 80.50   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczpd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vfrczpd	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczpd	%ymm0, %ymm3
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vfrczpd	(%rax), %ymm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczps	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vfrczps	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczps	%ymm0, %ymm3
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vfrczps	(%rax), %ymm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczsd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vfrczsd	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczss	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -     0.50   0.50   vfrczss	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcmov	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmov	(%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcmov	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpcmov	%ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpcmov	(%rax), %ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpcmov	%ymm0, (%rax), %ymm1, %ymm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomb	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomb	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomd	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomd	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomq	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomq	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomub	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomub	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomud	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomud	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomuq	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomuq	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomuw	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomuw	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpcomw	$0, %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpcomw	$0, (%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermil2pd	$0, %xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2pd	$0, (%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2pd	$0, %xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermil2pd	$0, %ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2pd	$0, (%rax), %ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2pd	$0, %ymm0, (%rax), %ymm1, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermil2ps	$0, %xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2ps	$0, (%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2ps	$0, %xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermil2ps	$0, %ymm0, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2ps	$0, (%rax), %ymm0, %ymm1, %ymm3
-# CHECK-NEXT:  -      -      -      -      -     1.00   0.50   0.50   vpermil2ps	$0, %ymm0, (%rax), %ymm1, %ymm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddbd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddbd	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddbq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddbq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddbw	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddbw	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphadddq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphadddq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddubd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddubd	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddubq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddubq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddubw	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddubw	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddudq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddudq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphadduwd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphadduwd	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphadduwq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphadduwq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddwd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddwd	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphaddwq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphaddwq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphsubbw	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphsubbw	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphsubdq	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphsubdq	(%rax), %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50    -      -     vphsubwd	%xmm0, %xmm3
-# CHECK-NEXT:  -      -      -     1.50    -     1.50   0.50   0.50   vphsubwd	(%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacsdd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacsdd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacsdqh	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacsdqh	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacsdql	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacsdql	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacssdd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacssdd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacssdqh	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacssdqh	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacssdql	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacssdql	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacsswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacsswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacssww	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacssww	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmacsww	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmacsww	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmadcsswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmadcsswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpmadcswd	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpmadcswd	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50    -      -     vpperm	%xmm0, %xmm1, %xmm2, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpperm	(%rax), %xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -     0.50    -     0.50   0.50   0.50   vpperm	%xmm0, (%rax), %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotb	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotb	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotb	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotb	$0, %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotb	$0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotd	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotd	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotd	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotd	$0, %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotd	$0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotq	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotq	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotq	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotq	$0, %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotq	$0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotw	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotw	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotw	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vprotw	$0, %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vprotw	$0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshab	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshab	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshab	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshad	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshad	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshad	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshaq	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshaq	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshaq	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshaw	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshaw	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshaw	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshlb	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshlb	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshlb	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshld	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshld	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshld	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshlq	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshlq	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshlq	%xmm0, (%rax), %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpshlw	%xmm0, %xmm1, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshlw	(%rax), %xmm0, %xmm3
-# CHECK-NEXT:  -      -     1.00    -      -      -     0.50   0.50   vpshlw	%xmm0, (%rax), %xmm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczpd	%xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczpd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vfrczpd	%ymm0, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vfrczpd	(%rax), %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczps	%xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczps	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vfrczps	%ymm0, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     2.00    -      -      -      -     vfrczps	(%rax), %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczsd	%xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczsd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczss	%xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczss	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmov	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmov	(%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmov	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmov	%ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmov	(%rax), %ymm0, %ymm1, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcmov	%ymm0, (%rax), %ymm1, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomb	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomb	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomd	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomd	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomq	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomq	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomub	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomub	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomud	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomud	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomuq	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomuq	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomuw	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomuw	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomw	$0, %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpcomw	$0, (%rax), %xmm0, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermil2pd	$0, %xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermil2pd	$0, (%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermil2pd	$0, %xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermil2pd	$0, %ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermil2pd	$0, (%rax), %ymm0, %ymm1, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermil2pd	$0, %ymm0, (%rax), %ymm1, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermil2ps	$0, %xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermil2ps	$0, (%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.50   0.50    -      -      -      -     vpermil2ps	$0, %xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermil2ps	$0, %ymm0, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermil2ps	$0, (%rax), %ymm0, %ymm1, %ymm3
+# CHECK-NEXT: 1.00   1.00    -      -      -      -      -      -     3.00   3.00    -      -      -      -     1.00   1.00    -      -      -      -     vpermil2ps	$0, %ymm0, (%rax), %ymm1, %ymm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddbd	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddbd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddbq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddbq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddbw	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddbw	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphadddq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphadddq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddubd	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddubd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddubq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddubq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddubw	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddubw	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddudq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddudq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphadduwd	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphadduwd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphadduwq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphadduwq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddwd	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddwd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddwq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphaddwq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubbw	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubbw	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubdq	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubdq	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubwd	%xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vphsubwd	(%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacsdd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacsdd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacsdqh	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacsdqh	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacsdql	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacsdql	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacssdd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacssdd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacssdqh	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacssdqh	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacssdql	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50   2.00    -     2.50   0.50    -      -      -      -     vpmacssdql	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacsswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacsswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacssww	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacssww	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacsww	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmacsww	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmadcsswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmadcsswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmadcswd	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     vpmadcswd	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpperm	%xmm0, %xmm1, %xmm2, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpperm	(%rax), %xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     2.00   2.00    -      -     0.50   0.50    -      -      -      -     vpperm	%xmm0, (%rax), %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotb	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotb	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotb	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotb	$0, %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotb	$0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotd	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotd	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotd	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotd	$0, %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotd	$0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotq	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotq	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotq	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotq	$0, %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotq	$0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotw	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotw	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotw	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotw	$0, %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vprotw	$0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshab	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshab	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshab	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshad	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshad	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshad	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshaq	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshaq	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshaq	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshaw	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshaw	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshaw	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlb	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlb	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlb	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshld	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshld	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshld	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlq	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlq	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlq	%xmm0, (%rax), %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlw	%xmm0, %xmm1, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlw	(%rax), %xmm0, %xmm3
+# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -      -      -     0.50   0.50    -      -     0.50   0.50    -      -      -      -     vpshlw	%xmm0, (%rax), %xmm3
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/scheduler-queue-usage.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/scheduler-queue-usage.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/scheduler-queue-usage.s
@@ -1,17 +1,17 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -scheduler-stats < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -scheduler-stats < %s | FileCheck %s
 
 vmulps (%rsi), %xmm0, %xmm0
 add  %rsi, %rsi
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      14
-# CHECK-NEXT: Total uOps:        3
+# CHECK-NEXT: Total Cycles:      13
+# CHECK-NEXT: Total uOps:        2
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.21
-# CHECK-NEXT: IPC:               0.14
+# CHECK-NEXT: uOps Per Cycle:    0.15
+# CHECK-NEXT: IPC:               0.15
 # CHECK-NEXT: Block RThroughput: 1.0
 
 # CHECK:      Instruction Info:
@@ -23,13 +23,13 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      11    1.00    *                   vmulps	(%rsi), %xmm0, %xmm0
-# CHECK-NEXT:  1      1     0.33                        addq	%rsi, %rsi
+# CHECK-NEXT:  1      10    1.00    *                   vmulps	(%rsi), %xmm0, %xmm0
+# CHECK-NEXT:  1      1     0.50                        addq	%rsi, %rsi
 
 # CHECK:      Schedulers - number of cycles where we saw N instructions issued:
 # CHECK-NEXT: [# issued], [# cycles]
-# CHECK-NEXT:  0,          13  (92.9%)
-# CHECK-NEXT:  2,          1  (7.1%)
+# CHECK-NEXT:  0,          12  (92.3%)
+# CHECK-NEXT:  2,          1  (7.7%)
 
 # CHECK:      Scheduler's queue usage:
 # CHECK-NEXT: [1] Resource name.
@@ -38,23 +38,38 @@
 # CHECK-NEXT: [4] Total number of buffer entries.
 
 # CHECK:       [1]            [2]        [3]        [4]
-# CHECK-NEXT: SBPortAny        0          2          54
+# CHECK-NEXT: PdEX             0          2          40
+# CHECK-NEXT: PdFPU            0          1          64
+# CHECK-NEXT: PdLoad           0          1          40
+# CHECK-NEXT: PdStore          0          1          24
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00    -      -     1.00    -     1.00
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -     1.00    -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     1.00    -      -      -      -     1.00   vmulps	(%rsi), %xmm0, %xmm0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     addq	%rsi, %rsi
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -      -      -     vmulps	(%rsi), %xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -     addq	%rsi, %rsi
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/simple-test.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/simple-test.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/simple-test.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=100 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=100 < %s | FileCheck %s
 
 add %edi, %eax
 
@@ -11,7 +11,7 @@
 # CHECK:      Dispatch Width:    4
 # CHECK-NEXT: uOps Per Cycle:    0.97
 # CHECK-NEXT: IPC:               0.97
-# CHECK-NEXT: Block RThroughput: 0.3
+# CHECK-NEXT: Block RThroughput: 0.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -22,22 +22,34 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.33                        addl	%edi, %eax
+# CHECK-NEXT:  1      1     0.50                        addl	%edi, %eax
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.34    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.33   0.33    -     0.34    -      -     addl	%edi, %eax
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     addl	%edi, %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vbroadcast-operand-latency.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vbroadcast-operand-latency.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vbroadcast-operand-latency.s
@@ -1,18 +1,18 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 leaq 8(%rsp, %rdi, 2), %rax
 vbroadcastss (%rax), %ymm0
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      200
-# CHECK-NEXT: Total Cycles:      60
-# CHECK-NEXT: Total uOps:        200
+# CHECK-NEXT: Total Cycles:      206
+# CHECK-NEXT: Total uOps:        400
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    3.33
-# CHECK-NEXT: IPC:               3.33
-# CHECK-NEXT: Block RThroughput: 0.5
+# CHECK-NEXT: uOps Per Cycle:    1.94
+# CHECK-NEXT: IPC:               0.97
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -23,38 +23,50 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     0.50                        leaq	8(%rsp,%rdi,2), %rax
-# CHECK-NEXT:  1      7     0.50    *                   vbroadcastss	(%rax), %ymm0
+# CHECK-NEXT:  2      1     0.50                        leaq	8(%rsp,%rdi,2), %rax
+# CHECK-NEXT:  2      6     2.00    *                   vbroadcastss	(%rax), %ymm0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     0.50   0.50    -      -     0.50   0.50
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -     1.00    -      -      -     0.50   0.50    -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -     leaq	8(%rsp,%rdi,2), %rax
-# CHECK-NEXT:  -      -      -      -      -      -     0.50   0.50   vbroadcastss	(%rax), %ymm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	8(%rsp,%rdi,2), %rax
+# CHECK-NEXT:  -     1.00    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -      -     vbroadcastss	(%rax), %ymm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01
+# CHECK-NEXT:                     012
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeER .    ..   leaq	8(%rsp,%rdi,2), %rax
-# CHECK-NEXT: [0,1]     D=eeeeeeeER.   vbroadcastss	(%rax), %ymm0
-# CHECK-NEXT: [1,0]     DeE-------R.   leaq	8(%rsp,%rdi,2), %rax
-# CHECK-NEXT: [1,1]     D=eeeeeeeER.   vbroadcastss	(%rax), %ymm0
-# CHECK-NEXT: [2,0]     .DeE------R.   leaq	8(%rsp,%rdi,2), %rax
-# CHECK-NEXT: [2,1]     .D=eeeeeeeER   vbroadcastss	(%rax), %ymm0
+# CHECK:      [0,0]     DeER .    . .   leaq	8(%rsp,%rdi,2), %rax
+# CHECK-NEXT: [0,1]     DeeeeeeER . .   vbroadcastss	(%rax), %ymm0
+# CHECK-NEXT: [1,0]     .DeE----R . .   leaq	8(%rsp,%rdi,2), %rax
+# CHECK-NEXT: [1,1]     .DeeeeeeER. .   vbroadcastss	(%rax), %ymm0
+# CHECK-NEXT: [2,0]     . DeE----R. .   leaq	8(%rsp,%rdi,2), %rax
+# CHECK-NEXT: [2,1]     . D==eeeeeeER   vbroadcastss	(%rax), %ymm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -63,5 +75,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.0    1.0    4.3       leaq	8(%rsp,%rdi,2), %rax
-# CHECK-NEXT: 1.     3     2.0    0.0    0.0       vbroadcastss	(%rax), %ymm0
+# CHECK-NEXT: 0.     3     1.0    1.0    2.7       leaq	8(%rsp,%rdi,2), %rax
+# CHECK-NEXT: 1.     3     1.7    0.7    0.0       vbroadcastss	(%rax), %ymm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vec-logic-read-after-ld-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vec-logic-read-after-ld-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vec-logic-read-after-ld-1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
 
 vaddps %xmm0, %xmm0, %xmm1
 vandps (%rdi), %xmm1, %xmm2
@@ -7,10 +7,10 @@
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
 # CHECK-NEXT: Total Cycles:      10
-# CHECK-NEXT: Total uOps:        3
+# CHECK-NEXT: Total uOps:        2
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.30
+# CHECK-NEXT: uOps Per Cycle:    0.20
 # CHECK-NEXT: IPC:               0.20
 # CHECK-NEXT: Block RThroughput: 1.0
 
@@ -23,13 +23,13 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddps	%xmm0, %xmm0, %xmm1
-# CHECK-NEXT:  2      7     1.00    *                   vandps	(%rdi), %xmm1, %xmm2
+# CHECK-NEXT:  1      5     1.00                        vaddps	%xmm0, %xmm0, %xmm1
+# CHECK-NEXT:  1      7     0.50    *                   vandps	(%rdi), %xmm1, %xmm2
 
 # CHECK:      Timeline view:
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER   .   vaddps	%xmm0, %xmm0, %xmm1
+# CHECK:      [0,0]     DeeeeeER .   vaddps	%xmm0, %xmm0, %xmm1
 # CHECK-NEXT: [0,1]     DeeeeeeeER   vandps	(%rdi), %xmm1, %xmm2
 
 # CHECK:      Average Wait times (based on the timeline view):
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vec-logic-read-after-ld-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vec-logic-read-after-ld-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/vec-logic-read-after-ld-2.s
@@ -1,18 +1,18 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s
 
 vaddps %ymm0, %ymm0, %ymm1
 vandps (%rdi), %ymm1, %ymm2
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      2
-# CHECK-NEXT: Total Cycles:      11
-# CHECK-NEXT: Total uOps:        3
+# CHECK-NEXT: Total Cycles:      10
+# CHECK-NEXT: Total uOps:        4
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.27
-# CHECK-NEXT: IPC:               0.18
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    0.40
+# CHECK-NEXT: IPC:               0.20
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -23,15 +23,14 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  2      8     1.00    *                   vandps	(%rdi), %ymm1, %ymm2
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  2      7     1.00    *                   vandps	(%rdi), %ymm1, %ymm2
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER    .   vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [0,1]     DeeeeeeeeER   vandps	(%rdi), %ymm1, %ymm2
+# CHECK:      [0,0]     DeeeeeER .   vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [0,1]     DeeeeeeeER   vandps	(%rdi), %ymm1, %ymm2
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/xop-super-registers-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/xop-super-registers-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/xop-super-registers-1.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
 
   vmulps  %ymm0, %ymm1, %ymm2
   vfrczpd %xmm1, %xmm2
@@ -10,13 +10,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      600
-# CHECK-NEXT: Total Cycles:      318
-# CHECK-NEXT: Total uOps:        600
+# CHECK-NEXT: Total Cycles:      717
+# CHECK-NEXT: Total uOps:        1200
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.89
-# CHECK-NEXT: IPC:               1.89
-# CHECK-NEXT: Block RThroughput: 3.0
+# CHECK-NEXT: uOps Per Cycle:    1.67
+# CHECK-NEXT: IPC:               0.84
+# CHECK-NEXT: Block RThroughput: 7.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -27,52 +27,64 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  1      3     1.00                        vfrczpd	%xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      10    1.00                        vfrczpd	%xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm4, %ymm5, %ymm0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     3.00   3.00    -      -      -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.04   4.96    -      -      -     1.00   4.00   7.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vfrczpd	%xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -     1.00    -     1.00    -      -      -      -     vfrczpd	%xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.01   0.99    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.01   0.99    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.02   0.98    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm4, %ymm5, %ymm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789
-# CHECK-NEXT: Index     0123456789          0123456
+# CHECK-NEXT:                     0123456789          0123456789
+# CHECK-NEXT: Index     0123456789          0123456789          0
 
-# CHECK:      [0,0]     DeeeeeER  .    .    .    ..   vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [0,1]     DeeeE--R  .    .    .    ..   vfrczpd	%xmm1, %xmm2
-# CHECK-NEXT: [0,2]     D===eeeeeER    .    .    ..   vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT: [0,3]     D========eeeER .    .    ..   vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT: [0,4]     .D==========eeeeeER .    ..   vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT: [0,5]     .D===============eeeER   ..   vaddps	%ymm4, %ymm5, %ymm0
-# CHECK-NEXT: [1,0]     .D==================eeeeeER   vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [1,1]     .DeeeE--------------------R   vfrczpd	%xmm1, %xmm2
-# CHECK-NEXT: [1,2]     . D==eeeeeE---------------R   vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT: [1,3]     . D=======eeeE------------R   vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT: [1,4]     . D==========eeeeeE-------R   vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT: [1,5]     . D===============eeeE----R   vaddps	%ymm4, %ymm5, %ymm0
+# CHECK:      [0,0]     DeeeeeER  .    .    .    .    .    .    .   vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [0,1]     D==eeeeeeeeeeER.    .    .    .    .    .   vfrczpd	%xmm1, %xmm2
+# CHECK-NEXT: [0,2]     .D===========eeeeeER.    .    .    .    .   vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT: [0,3]     .D================eeeeeER.    .    .    .   vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT: [0,4]     . D====================eeeeeER.    .    .   vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT: [0,5]     . D=========================eeeeeER.    .   vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT: [1,0]     .  D=============================eeeeeER.   vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [1,1]     .  DeeeeeeeeeeE------------------------R.   vfrczpd	%xmm1, %xmm2
+# CHECK-NEXT: [1,2]     .   D==========eeeeeE------------------R.   vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT: [1,3]     .   D===============eeeeeE-------------R.   vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT: [1,4]     .    D===================eeeeeE---------R   vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT: [1,5]     .    D========================eeeeeE----R   vaddps	%ymm4, %ymm5, %ymm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -81,9 +93,9 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     10.0   0.5    0.0       vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: 1.     2     1.0    1.0    11.0      vfrczpd	%xmm1, %xmm2
-# CHECK-NEXT: 2.     2     3.5    0.0    7.5       vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT: 3.     2     8.5    0.0    6.0       vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT: 4.     2     11.0   0.0    3.5       vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT: 5.     2     16.0   0.0    2.0       vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT: 0.     2     15.5   0.5    0.0       vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.     2     2.0    2.0    12.0      vfrczpd	%xmm1, %xmm2
+# CHECK-NEXT: 2.     2     11.5   0.5    9.0       vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT: 3.     2     16.5   0.0    6.5       vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT: 4.     2     20.5   0.0    4.5       vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT: 5.     2     25.5   0.0    2.0       vaddps	%ymm4, %ymm5, %ymm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/xop-super-registers-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/xop-super-registers-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/xop-super-registers-2.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
 
   vmulps     %ymm0, %ymm1, %ymm2
   vpermil2pd $16, %xmm3, %xmm5, %xmm1, %xmm2
@@ -10,13 +10,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      600
-# CHECK-NEXT: Total Cycles:      316
-# CHECK-NEXT: Total uOps:        600
+# CHECK-NEXT: Total Cycles:      653
+# CHECK-NEXT: Total uOps:        1100
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.90
-# CHECK-NEXT: IPC:               1.90
-# CHECK-NEXT: Block RThroughput: 3.0
+# CHECK-NEXT: uOps Per Cycle:    1.68
+# CHECK-NEXT: IPC:               0.92
+# CHECK-NEXT: Block RThroughput: 6.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -27,52 +27,64 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT:  1      5     1.00                        vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  1      3     2.00                        vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT:  2      5     2.00                        vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm4, %ymm5, %ymm0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     3.00   2.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     4.54   4.46    -      -      -      -     4.99   6.01    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.52   0.48    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.00   2.00    -      -      -      -     0.99   0.01    -      -      -      -     vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -      -     2.00    -      -      -      -     vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.52   0.48    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm4, %ymm5, %ymm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789
-# CHECK-NEXT: Index     0123456789          01234
+# CHECK-NEXT:                     0123456789          012
+# CHECK-NEXT: Index     0123456789          0123456789
 
-# CHECK:      [0,0]     DeeeeeER  .    .    .   .   vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [0,1]     DeE----R  .    .    .   .   vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
-# CHECK-NEXT: [0,2]     D=eeeeeER .    .    .   .   vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT: [0,3]     D======eeeER   .    .   .   vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT: [0,4]     .D========eeeeeER   .   .   vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT: [0,5]     .D=============eeeER.   .   vaddps	%ymm4, %ymm5, %ymm0
-# CHECK-NEXT: [1,0]     .D================eeeeeER   vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [1,1]     .DeE--------------------R   vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
-# CHECK-NEXT: [1,2]     . DeeeeeE---------------R   vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT: [1,3]     . D=====eeeE------------R   vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT: [1,4]     . D========eeeeeE-------R   vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT: [1,5]     . D=============eeeE----R   vaddps	%ymm4, %ymm5, %ymm0
+# CHECK:      [0,0]     DeeeeeER  .    .    .    .    . .   vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [0,1]     DeeeE--R  .    .    .    .    . .   vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
+# CHECK-NEXT: [0,2]     .D==eeeeeER    .    .    .    . .   vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT: [0,3]     .D=======eeeeeER    .    .    . .   vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT: [0,4]     . D============eeeeeER   .    . .   vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT: [0,5]     . D=================eeeeeER   . .   vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT: [1,0]     .  D=====================eeeeeER.   vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [1,1]     .  D=eeeE----------------------R.   vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
+# CHECK-NEXT: [1,2]     .   D=====eeeeeE---------------R.   vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT: [1,3]     .   D===========eeeeeE---------R.   vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT: [1,4]     .    D===============eeeeeE-----R   vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT: [1,5]     .    D====================eeeeeER   vaddps	%ymm4, %ymm5, %ymm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -81,9 +93,9 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     2     9.0    0.5    0.0       vmulps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: 1.     2     1.0    1.0    12.0      vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
-# CHECK-NEXT: 2.     2     1.5    0.0    7.5       vmulps	%ymm2, %ymm3, %ymm4
-# CHECK-NEXT: 3.     2     6.5    0.0    6.0       vaddps	%ymm4, %ymm5, %ymm6
-# CHECK-NEXT: 4.     2     9.0    0.0    3.5       vmulps	%ymm6, %ymm3, %ymm4
-# CHECK-NEXT: 5.     2     14.0   0.0    2.0       vaddps	%ymm4, %ymm5, %ymm0
+# CHECK-NEXT: 0.     2     11.5   0.5    0.0       vmulps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.     2     1.5    1.5    12.0      vpermil2pd	$16, %xmm3, %xmm5, %xmm1, %xmm2
+# CHECK-NEXT: 2.     2     4.5    1.0    7.5       vmulps	%ymm2, %ymm3, %ymm4
+# CHECK-NEXT: 3.     2     10.0   0.5    4.5       vaddps	%ymm4, %ymm5, %ymm6
+# CHECK-NEXT: 4.     2     14.5   0.5    2.5       vmulps	%ymm6, %ymm3, %ymm4
+# CHECK-NEXT: 5.     2     19.5   0.0    0.0       vaddps	%ymm4, %ymm5, %ymm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/zero-idioms-avx-256.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/zero-idioms-avx-256.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/zero-idioms-avx-256.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
 
 # TODO: Fix the processor resource usage for zero-idiom YMM XOR instructions.
 #       Those vector XOR instructions should only consume 1cy of JFPU1 (instead
@@ -40,13 +40,13 @@
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      300
-# CHECK-NEXT: Total Cycles:      107
-# CHECK-NEXT: Total uOps:        300
+# CHECK-NEXT: Total Cycles:      305
+# CHECK-NEXT: Total uOps:        600
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    2.80
-# CHECK-NEXT: IPC:               2.80
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.97
+# CHECK-NEXT: IPC:               0.98
+# CHECK-NEXT: Block RThroughput: 2.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -57,42 +57,55 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  1      1     1.00                        vxorps	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT:  1      1     0.50                        vblendps	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  2      2     1.00                        vxorps	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT:  2      2     1.00                        vblendps	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.52   2.48    -      -      -      -     3.00   3.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vxorps	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vblendps	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     0.98   1.02    -      -      -      -     vxorps	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.02   0.98    -      -      -      -     0.02   1.98    -      -      -      -     vblendps	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      Timeline view:
+# CHECK-NEXT:                     012
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER   .   vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [0,1]     D===eER  .   vxorps	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: [0,2]     D====eER .   vblendps	$2, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT: [1,0]     D=eeeE-R .   vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [1,1]     .D===eER .   vxorps	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: [1,2]     .D====eER.   vblendps	$2, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT: [2,0]     .D=eeeE-R.   vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [2,1]     .D====eER.   vxorps	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: [2,2]     . D====eER   vblendps	$2, %ymm1, %ymm2, %ymm3
+# CHECK:      [0,0]     DeeeeeER  . .   vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [0,1]     DeeE---R  . .   vxorps	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: [0,2]     .D=eeE-R  . .   vblendps	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: [1,0]     .D=eeeeeER. .   vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [1,1]     . D==eeE-R. .   vxorps	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: [1,2]     . D====eeER .   vblendps	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: [2,0]     .  D=eeeeeER.   vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [2,1]     .  D===eeE-R.   vxorps	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: [2,2]     .   D====eeER   vblendps	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -101,21 +114,21 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.7    1.7    0.7       vaddps	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: 1.     3     4.3    0.0    0.0       vxorps	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: 2.     3     5.0    0.0    0.0       vblendps	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 0.     3     1.7    1.7    0.0       vaddps	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: 1.     3     2.7    2.7    1.7       vxorps	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: 2.     3     4.0    0.0    0.3       vblendps	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      [1] Code Region - ZERO-IDIOM-2
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      300
-# CHECK-NEXT: Total Cycles:      107
-# CHECK-NEXT: Total uOps:        300
+# CHECK-NEXT: Total Cycles:      305
+# CHECK-NEXT: Total uOps:        600
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    2.80
-# CHECK-NEXT: IPC:               2.80
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.97
+# CHECK-NEXT: IPC:               0.98
+# CHECK-NEXT: Block RThroughput: 2.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -126,42 +139,55 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddpd	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  1      1     1.00                        vxorpd	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT:  1      1     0.50                        vblendpd	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT:  2      5     2.00                        vaddpd	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  2      2     1.00                        vxorpd	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT:  2      2     1.00                        vblendpd	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     1.00   1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     2.52   2.48    -      -      -      -     3.00   3.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddpd	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vxorpd	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vblendpd	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddpd	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -     0.98   1.02    -      -      -      -     vxorpd	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.02   0.98    -      -      -      -     0.02   1.98    -      -      -      -     vblendpd	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      Timeline view:
+# CHECK-NEXT:                     012
 # CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER   .   vaddpd	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [0,1]     D===eER  .   vxorpd	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: [0,2]     D====eER .   vblendpd	$2, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT: [1,0]     D=eeeE-R .   vaddpd	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [1,1]     .D===eER .   vxorpd	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: [1,2]     .D====eER.   vblendpd	$2, %ymm1, %ymm2, %ymm3
-# CHECK-NEXT: [2,0]     .D=eeeE-R.   vaddpd	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [2,1]     .D====eER.   vxorpd	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: [2,2]     . D====eER   vblendpd	$2, %ymm1, %ymm2, %ymm3
+# CHECK:      [0,0]     DeeeeeER  . .   vaddpd	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [0,1]     DeeE---R  . .   vxorpd	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: [0,2]     .D=eeE-R  . .   vblendpd	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: [1,0]     .D=eeeeeER. .   vaddpd	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [1,1]     . D==eeE-R. .   vxorpd	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: [1,2]     . D====eeER .   vblendpd	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: [2,0]     .  D=eeeeeER.   vaddpd	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [2,1]     .  D===eeE-R.   vxorpd	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: [2,2]     .   D====eeER   vblendpd	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -170,21 +196,21 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.7    1.7    0.7       vaddpd	%ymm0, %ymm0, %ymm1
-# CHECK-NEXT: 1.     3     4.3    0.0    0.0       vxorpd	%ymm1, %ymm1, %ymm1
-# CHECK-NEXT: 2.     3     5.0    0.0    0.0       vblendpd	$2, %ymm1, %ymm2, %ymm3
+# CHECK-NEXT: 0.     3     1.7    1.7    0.0       vaddpd	%ymm0, %ymm0, %ymm1
+# CHECK-NEXT: 1.     3     2.7    2.7    1.7       vxorpd	%ymm1, %ymm1, %ymm1
+# CHECK-NEXT: 2.     3     4.0    0.0    0.3       vblendpd	$2, %ymm1, %ymm2, %ymm3
 
 # CHECK:      [2] Code Region - ZERO-IDIOM-3
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      200
-# CHECK-NEXT: Total Cycles:      106
-# CHECK-NEXT: Total uOps:        200
+# CHECK-NEXT: Total Cycles:      206
+# CHECK-NEXT: Total uOps:        400
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.89
-# CHECK-NEXT: IPC:               1.89
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.94
+# CHECK-NEXT: IPC:               0.97
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -195,37 +221,50 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      2     1.00                        vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.50   1.50    -      -      -      -     2.00   2.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -      -     2.00    -      -      -      -     vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     012345678
+# CHECK-NEXT:                     01
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER  .   vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [0,1]     D===eER .   vandnps	%ymm2, %ymm2, %ymm3
-# CHECK-NEXT: [1,0]     D=eeeER .   vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [1,1]     D====eER.   vandnps	%ymm2, %ymm2, %ymm3
-# CHECK-NEXT: [2,0]     .D=eeeER.   vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [2,1]     .D====eER   vandnps	%ymm2, %ymm2, %ymm3
+# CHECK:      [0,0]     DeeeeeER  ..   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [0,1]     DeeE---R  ..   vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: [1,0]     .D=eeeeeER..   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [1,1]     .D=eeE---R..   vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: [2,0]     . D==eeeeeER   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [2,1]     . D==eeE---R   vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -234,20 +273,20 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.7    1.7    0.0       vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: 1.     3     4.7    0.0    0.0       vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: 0.     3     2.0    2.0    0.0       vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.     3     2.0    2.0    3.0       vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      [3] Code Region - ZERO-IDIOM-4
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      200
-# CHECK-NEXT: Total Cycles:      106
-# CHECK-NEXT: Total uOps:        200
+# CHECK-NEXT: Total Cycles:      206
+# CHECK-NEXT: Total uOps:        400
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.89
-# CHECK-NEXT: IPC:               1.89
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.94
+# CHECK-NEXT: IPC:               0.97
+# CHECK-NEXT: Block RThroughput: 2.0
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -258,37 +297,50 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  1      1     1.00                        vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  2      2     1.00                        vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.50   1.50    -      -      -      -     2.00   2.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     0.50   0.50    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00   1.00    -      -      -      -      -     2.00    -      -      -      -     vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      Timeline view:
-# CHECK-NEXT: Index     012345678
+# CHECK-NEXT:                     01
+# CHECK-NEXT: Index     0123456789
 
-# CHECK:      [0,0]     DeeeER  .   vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [0,1]     D===eER .   vandnps	%ymm2, %ymm2, %ymm3
-# CHECK-NEXT: [1,0]     D=eeeER .   vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [1,1]     D====eER.   vandnps	%ymm2, %ymm2, %ymm3
-# CHECK-NEXT: [2,0]     .D=eeeER.   vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: [2,1]     .D====eER   vandnps	%ymm2, %ymm2, %ymm3
+# CHECK:      [0,0]     DeeeeeER  ..   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [0,1]     DeeE---R  ..   vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: [1,0]     .D=eeeeeER..   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [1,1]     .D=eeE---R..   vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: [2,0]     . D==eeeeeER   vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: [2,1]     . D==eeE---R   vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -297,20 +349,20 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     1.7    1.7    0.0       vaddps	%ymm0, %ymm1, %ymm2
-# CHECK-NEXT: 1.     3     4.7    0.0    0.0       vandnps	%ymm2, %ymm2, %ymm3
+# CHECK-NEXT: 0.     3     2.0    2.0    0.0       vaddps	%ymm0, %ymm1, %ymm2
+# CHECK-NEXT: 1.     3     2.0    2.0    3.0       vandnps	%ymm2, %ymm2, %ymm3
 
 # CHECK:      [4] Code Region - ZERO-IDIOM-5
 
 # CHECK:      Iterations:        100
 # CHECK-NEXT: Instructions:      200
-# CHECK-NEXT: Total Cycles:      403
-# CHECK-NEXT: Total uOps:        200
+# CHECK-NEXT: Total Cycles:      903
+# CHECK-NEXT: Total uOps:        1000
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    0.50
-# CHECK-NEXT: IPC:               0.50
-# CHECK-NEXT: Block RThroughput: 1.0
+# CHECK-NEXT: uOps Per Cycle:    1.11
+# CHECK-NEXT: IPC:               0.22
+# CHECK-NEXT: Block RThroughput: 2.5
 
 # CHECK:      Instruction Info:
 # CHECK-NEXT: [1]: #uOps
@@ -321,38 +373,50 @@
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  1      1     1.00                        vperm2f128	$136, %ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  1      3     1.00                        vaddps	%ymm1, %ymm1, %ymm0
+# CHECK-NEXT:  8      4     0.50                        vperm2f128	$136, %ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  2      5     2.00                        vaddps	%ymm1, %ymm1, %ymm0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     2.00    -      -      -      -     2.00   1.00    -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vperm2f128	$136, %ymm0, %ymm0, %ymm1
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vaddps	%ymm1, %ymm1, %ymm0
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -      -      -     vperm2f128	$136, %ymm0, %ymm0, %ymm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     2.00    -      -      -      -      -     vaddps	%ymm1, %ymm1, %ymm0
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     01234
-# CHECK-NEXT: Index     0123456789
+# CHECK-NEXT:                     0123456789
+# CHECK-NEXT: Index     0123456789          0123456789
 
-# CHECK:      [0,0]     DeER .    .   .   vperm2f128	$136, %ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [0,1]     D=eeeER   .   .   vaddps	%ymm1, %ymm1, %ymm0
-# CHECK-NEXT: [1,0]     D====eER  .   .   vperm2f128	$136, %ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [1,1]     D=====eeeER   .   vaddps	%ymm1, %ymm1, %ymm0
-# CHECK-NEXT: [2,0]     .D=======eER  .   vperm2f128	$136, %ymm0, %ymm0, %ymm1
-# CHECK-NEXT: [2,1]     .D========eeeER   vaddps	%ymm1, %ymm1, %ymm0
+# CHECK:      [0,0]     DeeeeER   .    .    .    .   .   vperm2f128	$136, %ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [0,1]     . D==eeeeeER   .    .    .   .   vaddps	%ymm1, %ymm1, %ymm0
+# CHECK-NEXT: [1,0]     .  D======eeeeER    .    .   .   vperm2f128	$136, %ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [1,1]     .    D========eeeeeER    .   .   vaddps	%ymm1, %ymm1, %ymm0
+# CHECK-NEXT: [2,0]     .    .D============eeeeER.   .   vperm2f128	$136, %ymm0, %ymm0, %ymm1
+# CHECK-NEXT: [2,1]     .    .  D==============eeeeeER   vaddps	%ymm1, %ymm1, %ymm0
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -361,5 +425,5 @@
 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 
 # CHECK:            [0]    [1]    [2]    [3]
-# CHECK-NEXT: 0.     3     4.7    0.3    0.0       vperm2f128	$136, %ymm0, %ymm0, %ymm1
-# CHECK-NEXT: 1.     3     5.7    0.0    0.0       vaddps	%ymm1, %ymm1, %ymm0
+# CHECK-NEXT: 0.     3     7.0    0.3    0.0       vperm2f128	$136, %ymm0, %ymm0, %ymm1
+# CHECK-NEXT: 1.     3     9.0    0.0    0.0       vaddps	%ymm1, %ymm1, %ymm0
Index: llvm/trunk/test/tools/llvm-mca/X86/BdVer2/zero-idioms.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/BdVer2/zero-idioms.s
+++ llvm/trunk/test/tools/llvm-mca/X86/BdVer2/zero-idioms.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -register-file-stats -iterations=1 < %s | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -register-file-stats -iterations=1 < %s | FileCheck %s
 
 subl  %eax, %eax
 subq  %rax, %rax
@@ -90,12 +90,12 @@
 
 # CHECK:      Iterations:        1
 # CHECK-NEXT: Instructions:      71
-# CHECK-NEXT: Total Cycles:      39
+# CHECK-NEXT: Total Cycles:      26
 # CHECK-NEXT: Total uOps:        71
 
 # CHECK:      Dispatch Width:    4
-# CHECK-NEXT: uOps Per Cycle:    1.82
-# CHECK-NEXT: IPC:               1.82
+# CHECK-NEXT: uOps Per Cycle:    2.73
+# CHECK-NEXT: IPC:               2.73
 # CHECK-NEXT: Block RThroughput: 17.8
 
 # CHECK:      Instruction Info:
@@ -111,25 +111,25 @@
 # CHECK-NEXT:  1      0     0.25                        subq	%rax, %rax
 # CHECK-NEXT:  1      0     0.25                        xorl	%eax, %eax
 # CHECK-NEXT:  1      0     0.25                        xorq	%rax, %rax
-# CHECK-NEXT:  1      3     1.00                        pcmpgtb	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpgtd	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        pcmpgtw	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        pcmpgtb	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        pcmpgtd	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        pcmpgtw	%mm2, %mm2
 # CHECK-NEXT:  1      0     0.25                        pcmpgtb	%xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        pcmpgtd	%xmm2, %xmm2
-# CHECK-NEXT:  1      0     0.25                        pcmpgtq	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        pcmpgtq	%xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        pcmpgtw	%xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtb	%xmm3, %xmm3, %xmm3
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      0     0.25                        vpcmpgtq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtq	%xmm3, %xmm3, %xmm3
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtw	%xmm3, %xmm3, %xmm3
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtb	%xmm3, %xmm3, %xmm5
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      0     0.25                        vpcmpgtq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vpcmpgtq	%xmm3, %xmm3, %xmm5
 # CHECK-NEXT:  1      0     0.25                        vpcmpgtw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      3     1.00                        psubb	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        psubd	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        psubq	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        psubw	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        psubb	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        psubd	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        psubq	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        psubw	%mm2, %mm2
 # CHECK-NEXT:  1      0     0.25                        psubb	%xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        psubd	%xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        psubq	%xmm2, %xmm2
@@ -142,37 +142,37 @@
 # CHECK-NEXT:  1      0     0.25                        vpsubd	%xmm3, %xmm3, %xmm5
 # CHECK-NEXT:  1      0     0.25                        vpsubq	%xmm3, %xmm3, %xmm5
 # CHECK-NEXT:  1      0     0.25                        vpsubw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      3     1.00                        psubsb	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        psubsw	%mm2, %mm2
-# CHECK-NEXT:  1      1     0.50                        psubsb	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubsw	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubsb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpsubsw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      1     0.50                        vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      3     1.00                        psubusb	%mm2, %mm2
-# CHECK-NEXT:  1      3     1.00                        psubusw	%mm2, %mm2
-# CHECK-NEXT:  1      1     0.50                        psubusb	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        psubusw	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.50                        vpsubusb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpsubusw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     0.50                        vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      1     0.50                        vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  1      1     1.00                        andnps	%xmm0, %xmm0
-# CHECK-NEXT:  1      1     1.00                        andnpd	%xmm1, %xmm1
-# CHECK-NEXT:  1      1     1.00                        vandnps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT:  1      1     1.00                        vandnpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT:  1      1     0.33                        pandn	%mm2, %mm2
-# CHECK-NEXT:  1      1     0.33                        pandn	%xmm2, %xmm2
-# CHECK-NEXT:  1      1     0.33                        vpandn	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  1      1     1.00                        vandnps	%xmm2, %xmm2, %xmm5
-# CHECK-NEXT:  1      1     1.00                        vandnpd	%xmm1, %xmm1, %xmm5
-# CHECK-NEXT:  1      1     0.33                        vpandn	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        psubsb	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        psubsw	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        psubsb	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubsw	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubsb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpsubsw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        psubusb	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        psubusw	%mm2, %mm2
+# CHECK-NEXT:  1      2     0.50                        psubusb	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        psubusw	%xmm2, %xmm2
+# CHECK-NEXT:  1      2     0.50                        vpsubusb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpsubusw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      2     0.50                        vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      2     0.50                        vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  1      0     0.25                        andnps	%xmm0, %xmm0
+# CHECK-NEXT:  1      0     0.25                        andnpd	%xmm1, %xmm1
+# CHECK-NEXT:  1      0     0.25                        vandnps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT:  1      0     0.25                        vandnpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT:  1      0     0.25                        pandn	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        pandn	%xmm2, %xmm2
+# CHECK-NEXT:  1      0     0.25                        vpandn	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  1      0     0.25                        vandnps	%xmm2, %xmm2, %xmm5
+# CHECK-NEXT:  1      0     0.25                        vandnpd	%xmm1, %xmm1, %xmm5
+# CHECK-NEXT:  1      0     0.25                        vpandn	%xmm3, %xmm3, %xmm5
 # CHECK-NEXT:  1      0     0.25                        xorps	%xmm0, %xmm0
 # CHECK-NEXT:  1      0     0.25                        xorpd	%xmm1, %xmm1
 # CHECK-NEXT:  1      0     0.25                        vxorps	%xmm2, %xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        vxorpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT:  1      1     0.33                        pxor	%mm2, %mm2
+# CHECK-NEXT:  1      0     0.25                        pxor	%mm2, %mm2
 # CHECK-NEXT:  1      0     0.25                        pxor	%xmm2, %xmm2
 # CHECK-NEXT:  1      0     0.25                        vpxor	%xmm3, %xmm3, %xmm3
 # CHECK-NEXT:  1      0     0.25                        vxorps	%xmm4, %xmm4, %xmm5
@@ -180,172 +180,194 @@
 # CHECK-NEXT:  1      0     0.25                        vpxor	%xmm3, %xmm3, %xmm5
 
 # CHECK:      Register File statistics:
-# CHECK-NEXT: Total number of mappings created:    75
-# CHECK-NEXT: Max number of mappings used:         51
+# CHECK-NEXT: Total number of mappings created:    1
+# CHECK-NEXT: Max number of mappings used:         1
+
+# CHECK:      *  Register File #1 -- PdFpuPRF:
+# CHECK-NEXT:    Number of physical registers:     160
+# CHECK-NEXT:    Total number of mappings created: 1
+# CHECK-NEXT:    Max number of mappings used:      1
+
+# CHECK:      *  Register File #2 -- PdIntegerPRF:
+# CHECK-NEXT:    Number of physical registers:     96
+# CHECK-NEXT:    Total number of mappings created: 0
+# CHECK-NEXT:    Max number of mappings used:      0
 
 # CHECK:      Resources:
-# CHECK-NEXT: [0]   - SBDivider
-# CHECK-NEXT: [1]   - SBFPDivider
-# CHECK-NEXT: [2]   - SBPort0
-# CHECK-NEXT: [3]   - SBPort1
-# CHECK-NEXT: [4]   - SBPort4
-# CHECK-NEXT: [5]   - SBPort5
-# CHECK-NEXT: [6.0] - SBPort23
-# CHECK-NEXT: [6.1] - SBPort23
+# CHECK-NEXT: [0.0] - PdAGLU01
+# CHECK-NEXT: [0.1] - PdAGLU01
+# CHECK-NEXT: [1]   - PdBranch
+# CHECK-NEXT: [2]   - PdCount
+# CHECK-NEXT: [3]   - PdDiv
+# CHECK-NEXT: [4]   - PdEX0
+# CHECK-NEXT: [5]   - PdEX1
+# CHECK-NEXT: [6]   - PdFPCVT
+# CHECK-NEXT: [7.0] - PdFPFMA
+# CHECK-NEXT: [7.1] - PdFPFMA
+# CHECK-NEXT: [8.0] - PdFPMAL
+# CHECK-NEXT: [8.1] - PdFPMAL
+# CHECK-NEXT: [9]   - PdFPMMA
+# CHECK-NEXT: [10]  - PdFPSTO
+# CHECK-NEXT: [11]  - PdFPU0
+# CHECK-NEXT: [12]  - PdFPU1
+# CHECK-NEXT: [13]  - PdFPU2
+# CHECK-NEXT: [14]  - PdFPU3
+# CHECK-NEXT: [15]  - PdFPXBR
+# CHECK-NEXT: [16]  - PdMul
 
 # CHECK:      Resource pressure per iteration:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
-# CHECK-NEXT:  -      -     5.00   16.00   -     13.00   -      -
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     8.00   11.00   -      -     9.00   10.00   -      -      -      -
 
 # CHECK:      Resource pressure by instruction:
-# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     subl	%eax, %eax
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     subq	%rax, %rax
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorl	%eax, %eax
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorq	%rax, %rax
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpgtb	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpgtd	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpgtw	%mm2, %mm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     pcmpgtb	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     pcmpgtd	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     pcmpgtq	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     pcmpgtw	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpcmpgtw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubb	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubd	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubq	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubw	%mm2, %mm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     psubb	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     psubd	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     psubq	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     psubw	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpsubw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubsb	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubsw	%mm2, %mm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psubsb	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psubsw	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpsubsb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpsubsw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubusb	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubusw	%mm2, %mm2
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     psubusb	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     psubusw	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpsubusb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpsubusw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -     1.00    -      -      -      -     vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     andnps	%xmm0, %xmm0
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     andnpd	%xmm1, %xmm1
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pandn	%mm2, %mm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pandn	%xmm2, %xmm2
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpandn	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnps	%xmm2, %xmm2, %xmm5
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -     vandnpd	%xmm1, %xmm1, %xmm5
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     vpandn	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorps	%xmm0, %xmm0
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     xorpd	%xmm1, %xmm1
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vxorps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vxorpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT:  -      -     1.00    -      -      -      -      -     pxor	%mm2, %mm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     pxor	%xmm2, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpxor	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vxorps	%xmm4, %xmm4, %xmm5
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vxorpd	%xmm1, %xmm1, %xmm3
-# CHECK-NEXT:  -      -      -      -      -      -      -      -     vpxor	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16]   Instructions:
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     subl	%eax, %eax
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     subq	%rax, %rax
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorl	%eax, %eax
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorq	%rax, %rax
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pcmpgtb	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pcmpgtd	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pcmpgtw	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pcmpgtb	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pcmpgtd	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     pcmpgtq	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pcmpgtw	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     vpcmpgtq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     vpcmpgtq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubb	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubd	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubq	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubw	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubb	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubd	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubq	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     psubw	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpsubw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     psubsb	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     psubsw	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -      -     psubsb	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -     psubsw	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     vpsubsb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     vpsubsw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -      -     vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -     vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     psubusb	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     psubusw	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -      -     psubusb	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -     psubusw	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -      -      -      -     vpsubusb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -     vpsubusw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -      -      -     vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -      -      -     vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     andnps	%xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     andnpd	%xmm1, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vandnps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vandnpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pandn	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pandn	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpandn	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vandnps	%xmm2, %xmm2, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vandnpd	%xmm1, %xmm1, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpandn	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorps	%xmm0, %xmm0
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     xorpd	%xmm1, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vxorps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vxorpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pxor	%mm2, %mm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     pxor	%xmm2, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpxor	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vxorps	%xmm4, %xmm4, %xmm5
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vxorpd	%xmm1, %xmm1, %xmm3
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpxor	%xmm3, %xmm3, %xmm5
 
 # CHECK:      Timeline view:
-# CHECK-NEXT:                     0123456789          012345678
-# CHECK-NEXT: Index     0123456789          0123456789
+# CHECK-NEXT:                     0123456789
+# CHECK-NEXT: Index     0123456789          012345
 
-# CHECK:      [0,0]     DR   .    .    .    .    .    .    .  .   subl	%eax, %eax
-# CHECK-NEXT: [0,1]     DR   .    .    .    .    .    .    .  .   subq	%rax, %rax
-# CHECK-NEXT: [0,2]     DR   .    .    .    .    .    .    .  .   xorl	%eax, %eax
-# CHECK-NEXT: [0,3]     DR   .    .    .    .    .    .    .  .   xorq	%rax, %rax
-# CHECK-NEXT: [0,4]     .DeeeER   .    .    .    .    .    .  .   pcmpgtb	%mm2, %mm2
-# CHECK-NEXT: [0,5]     .D===eeeER.    .    .    .    .    .  .   pcmpgtd	%mm2, %mm2
-# CHECK-NEXT: [0,6]     .D======eeeER  .    .    .    .    .  .   pcmpgtw	%mm2, %mm2
-# CHECK-NEXT: [0,7]     .D----------R  .    .    .    .    .  .   pcmpgtb	%xmm2, %xmm2
-# CHECK-NEXT: [0,8]     . D---------R  .    .    .    .    .  .   pcmpgtd	%xmm2, %xmm2
-# CHECK-NEXT: [0,9]     . D---------R  .    .    .    .    .  .   pcmpgtq	%xmm2, %xmm2
-# CHECK-NEXT: [0,10]    . D---------R  .    .    .    .    .  .   pcmpgtw	%xmm2, %xmm2
-# CHECK-NEXT: [0,11]    . D---------R  .    .    .    .    .  .   vpcmpgtb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,12]    .  D--------R  .    .    .    .    .  .   vpcmpgtd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,13]    .  D--------R  .    .    .    .    .  .   vpcmpgtq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,14]    .  D--------R  .    .    .    .    .  .   vpcmpgtw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,15]    .  D--------R  .    .    .    .    .  .   vpcmpgtb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,16]    .   D-------R  .    .    .    .    .  .   vpcmpgtd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,17]    .   D-------R  .    .    .    .    .  .   vpcmpgtq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,18]    .   D-------R  .    .    .    .    .  .   vpcmpgtw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,19]    .   D======eeeER    .    .    .    .  .   psubb	%mm2, %mm2
-# CHECK-NEXT: [0,20]    .    D========eeeER .    .    .    .  .   psubd	%mm2, %mm2
-# CHECK-NEXT: [0,21]    .    D===========eeeER   .    .    .  .   psubq	%mm2, %mm2
-# CHECK-NEXT: [0,22]    .    D==============eeeER.    .    .  .   psubw	%mm2, %mm2
-# CHECK-NEXT: [0,23]    .    D------------------R.    .    .  .   psubb	%xmm2, %xmm2
-# CHECK-NEXT: [0,24]    .    .D-----------------R.    .    .  .   psubd	%xmm2, %xmm2
-# CHECK-NEXT: [0,25]    .    .D-----------------R.    .    .  .   psubq	%xmm2, %xmm2
-# CHECK-NEXT: [0,26]    .    .D-----------------R.    .    .  .   psubw	%xmm2, %xmm2
-# CHECK-NEXT: [0,27]    .    .D-----------------R.    .    .  .   vpsubb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,28]    .    . D----------------R.    .    .  .   vpsubd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,29]    .    . D----------------R.    .    .  .   vpsubq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,30]    .    . D----------------R.    .    .  .   vpsubw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,31]    .    . D----------------R.    .    .  .   vpsubb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,32]    .    .  D---------------R.    .    .  .   vpsubd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,33]    .    .  D---------------R.    .    .  .   vpsubq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,34]    .    .  D---------------R.    .    .  .   vpsubw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,35]    .    .  D==============eeeER  .    .  .   psubsb	%mm2, %mm2
-# CHECK-NEXT: [0,36]    .    .   D================eeeER    .  .   psubsw	%mm2, %mm2
-# CHECK-NEXT: [0,37]    .    .   DeE------------------R    .  .   psubsb	%xmm2, %xmm2
-# CHECK-NEXT: [0,38]    .    .   D==eE----------------R    .  .   psubsw	%xmm2, %xmm2
-# CHECK-NEXT: [0,39]    .    .   DeE------------------R    .  .   vpsubsb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,40]    .    .    DeE-----------------R    .  .   vpsubsw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,41]    .    .    D=eE----------------R    .  .   vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,42]    .    .    D==eE---------------R    .  .   vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,43]    .    .    D==================eeeER .  .   psubusb	%mm2, %mm2
-# CHECK-NEXT: [0,44]    .    .    .D====================eeeER .   psubusw	%mm2, %mm2
-# CHECK-NEXT: [0,45]    .    .    .D=eE---------------------R .   psubusb	%xmm2, %xmm2
-# CHECK-NEXT: [0,46]    .    .    .D==eE--------------------R .   psubusw	%xmm2, %xmm2
-# CHECK-NEXT: [0,47]    .    .    .D===eE-------------------R .   vpsubusb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,48]    .    .    . D===eE------------------R .   vpsubusw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,49]    .    .    . D====eE-----------------R .   vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,50]    .    .    . D=====eE----------------R .   vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,51]    .    .    . D===eE------------------R .   andnps	%xmm0, %xmm0
-# CHECK-NEXT: [0,52]    .    .    .  D====eE----------------R .   andnpd	%xmm1, %xmm1
-# CHECK-NEXT: [0,53]    .    .    .  D======eE--------------R .   vandnps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT: [0,54]    .    .    .  D=====eE---------------R .   vandnpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT: [0,55]    .    .    .  D=====================eER.   pandn	%mm2, %mm2
-# CHECK-NEXT: [0,56]    .    .    .   D======eE--------------R.   pandn	%xmm2, %xmm2
-# CHECK-NEXT: [0,57]    .    .    .   D==eE------------------R.   vpandn	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,58]    .    .    .   D=======eE-------------R.   vandnps	%xmm2, %xmm2, %xmm5
-# CHECK-NEXT: [0,59]    .    .    .   D======eE--------------R.   vandnpd	%xmm1, %xmm1, %xmm5
-# CHECK-NEXT: [0,60]    .    .    .    D==eE-----------------R.   vpandn	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: [0,61]    .    .    .    D=E-------------------R.   xorps	%xmm0, %xmm0
-# CHECK-NEXT: [0,62]    .    .    .    D====E----------------R.   xorpd	%xmm1, %xmm1
-# CHECK-NEXT: [0,63]    .    .    .    D======E--------------R.   vxorps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT: [0,64]    .    .    .    .D===E----------------R.   vxorpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT: [0,65]    .    .    .    .D===================eER   pxor	%mm2, %mm2
-# CHECK-NEXT: [0,66]    .    .    .    .D=====E---------------R   pxor	%xmm2, %xmm2
-# CHECK-NEXT: [0,67]    .    .    .    .D=E-------------------R   vpxor	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: [0,68]    .    .    .    . D--------------------R   vxorps	%xmm4, %xmm4, %xmm5
-# CHECK-NEXT: [0,69]    .    .    .    . D==E-----------------R   vxorpd	%xmm1, %xmm1, %xmm3
-# CHECK-NEXT: [0,70]    .    .    .    . D==E-----------------R   vpxor	%xmm3, %xmm3, %xmm5
+# CHECK:      [0,0]     DR   .    .    .    .    .   subl	%eax, %eax
+# CHECK-NEXT: [0,1]     DR   .    .    .    .    .   subq	%rax, %rax
+# CHECK-NEXT: [0,2]     DR   .    .    .    .    .   xorl	%eax, %eax
+# CHECK-NEXT: [0,3]     DR   .    .    .    .    .   xorq	%rax, %rax
+# CHECK-NEXT: [0,4]     .DR  .    .    .    .    .   pcmpgtb	%mm2, %mm2
+# CHECK-NEXT: [0,5]     .DR  .    .    .    .    .   pcmpgtd	%mm2, %mm2
+# CHECK-NEXT: [0,6]     .DR  .    .    .    .    .   pcmpgtw	%mm2, %mm2
+# CHECK-NEXT: [0,7]     .DR  .    .    .    .    .   pcmpgtb	%xmm2, %xmm2
+# CHECK-NEXT: [0,8]     . DR .    .    .    .    .   pcmpgtd	%xmm2, %xmm2
+# CHECK-NEXT: [0,9]     . DeeER   .    .    .    .   pcmpgtq	%xmm2, %xmm2
+# CHECK-NEXT: [0,10]    . D---R   .    .    .    .   pcmpgtw	%xmm2, %xmm2
+# CHECK-NEXT: [0,11]    . D---R   .    .    .    .   vpcmpgtb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,12]    .  D--R   .    .    .    .   vpcmpgtd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,13]    .  DeeER  .    .    .    .   vpcmpgtq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,14]    .  D---R  .    .    .    .   vpcmpgtw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,15]    .  D---R  .    .    .    .   vpcmpgtb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,16]    .   D--R  .    .    .    .   vpcmpgtd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,17]    .   DeeER .    .    .    .   vpcmpgtq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,18]    .   D---R .    .    .    .   vpcmpgtw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,19]    .   D---R .    .    .    .   psubb	%mm2, %mm2
+# CHECK-NEXT: [0,20]    .    D--R .    .    .    .   psubd	%mm2, %mm2
+# CHECK-NEXT: [0,21]    .    D---R.    .    .    .   psubq	%mm2, %mm2
+# CHECK-NEXT: [0,22]    .    D---R.    .    .    .   psubw	%mm2, %mm2
+# CHECK-NEXT: [0,23]    .    D---R.    .    .    .   psubb	%xmm2, %xmm2
+# CHECK-NEXT: [0,24]    .    .D--R.    .    .    .   psubd	%xmm2, %xmm2
+# CHECK-NEXT: [0,25]    .    .D---R    .    .    .   psubq	%xmm2, %xmm2
+# CHECK-NEXT: [0,26]    .    .D---R    .    .    .   psubw	%xmm2, %xmm2
+# CHECK-NEXT: [0,27]    .    .D---R    .    .    .   vpsubb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,28]    .    . D--R    .    .    .   vpsubd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,29]    .    . D---R   .    .    .   vpsubq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,30]    .    . D---R   .    .    .   vpsubw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,31]    .    . D---R   .    .    .   vpsubb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,32]    .    .  D--R   .    .    .   vpsubd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,33]    .    .  D---R  .    .    .   vpsubq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,34]    .    .  D---R  .    .    .   vpsubw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,35]    .    .  DeeER  .    .    .   psubsb	%mm2, %mm2
+# CHECK-NEXT: [0,36]    .    .   DeeER .    .    .   psubsw	%mm2, %mm2
+# CHECK-NEXT: [0,37]    .    .   DeeER .    .    .   psubsb	%xmm2, %xmm2
+# CHECK-NEXT: [0,38]    .    .   D=eeER.    .    .   psubsw	%xmm2, %xmm2
+# CHECK-NEXT: [0,39]    .    .   D=eeER.    .    .   vpsubsb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,40]    .    .    D=eeER    .    .   vpsubsw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,41]    .    .    D=eeER    .    .   vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,42]    .    .    D==eeER   .    .   vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,43]    .    .    D==eeER   .    .   psubusb	%mm2, %mm2
+# CHECK-NEXT: [0,44]    .    .    .D==eeER  .    .   psubusw	%mm2, %mm2
+# CHECK-NEXT: [0,45]    .    .    .D==eeER  .    .   psubusb	%xmm2, %xmm2
+# CHECK-NEXT: [0,46]    .    .    .D===eeER .    .   psubusw	%xmm2, %xmm2
+# CHECK-NEXT: [0,47]    .    .    .D===eeER .    .   vpsubusb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,48]    .    .    . D===eeER.    .   vpsubusw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,49]    .    .    . D===eeER.    .   vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,50]    .    .    . D====eeER    .   vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,51]    .    .    . D-------R    .   andnps	%xmm0, %xmm0
+# CHECK-NEXT: [0,52]    .    .    .  D------R    .   andnpd	%xmm1, %xmm1
+# CHECK-NEXT: [0,53]    .    .    .  D------R    .   vandnps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT: [0,54]    .    .    .  D-------R   .   vandnpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT: [0,55]    .    .    .  D-------R   .   pandn	%mm2, %mm2
+# CHECK-NEXT: [0,56]    .    .    .   D------R   .   pandn	%xmm2, %xmm2
+# CHECK-NEXT: [0,57]    .    .    .   D------R   .   vpandn	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,58]    .    .    .   D-------R  .   vandnps	%xmm2, %xmm2, %xmm5
+# CHECK-NEXT: [0,59]    .    .    .   D-------R  .   vandnpd	%xmm1, %xmm1, %xmm5
+# CHECK-NEXT: [0,60]    .    .    .    D------R  .   vpandn	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: [0,61]    .    .    .    D------R  .   xorps	%xmm0, %xmm0
+# CHECK-NEXT: [0,62]    .    .    .    D-------R .   xorpd	%xmm1, %xmm1
+# CHECK-NEXT: [0,63]    .    .    .    D-------R .   vxorps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT: [0,64]    .    .    .    .D------R .   vxorpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT: [0,65]    .    .    .    .D------R .   pxor	%mm2, %mm2
+# CHECK-NEXT: [0,66]    .    .    .    .D-------R.   pxor	%xmm2, %xmm2
+# CHECK-NEXT: [0,67]    .    .    .    .D-------R.   vpxor	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: [0,68]    .    .    .    . D------R.   vxorps	%xmm4, %xmm4, %xmm5
+# CHECK-NEXT: [0,69]    .    .    .    . D------R.   vxorpd	%xmm1, %xmm1, %xmm3
+# CHECK-NEXT: [0,70]    .    .    .    . D-------R   vpxor	%xmm3, %xmm3, %xmm5
 
 # CHECK:      Average Wait times (based on the timeline view):
 # CHECK-NEXT: [0]: Executions
@@ -358,70 +380,70 @@
 # CHECK-NEXT: 1.     1     0.0    0.0    0.0       subq	%rax, %rax
 # CHECK-NEXT: 2.     1     0.0    0.0    0.0       xorl	%eax, %eax
 # CHECK-NEXT: 3.     1     0.0    0.0    0.0       xorq	%rax, %rax
-# CHECK-NEXT: 4.     1     1.0    1.0    0.0       pcmpgtb	%mm2, %mm2
-# CHECK-NEXT: 5.     1     4.0    0.0    0.0       pcmpgtd	%mm2, %mm2
-# CHECK-NEXT: 6.     1     7.0    0.0    0.0       pcmpgtw	%mm2, %mm2
-# CHECK-NEXT: 7.     1     0.0    0.0    10.0      pcmpgtb	%xmm2, %xmm2
-# CHECK-NEXT: 8.     1     0.0    0.0    9.0       pcmpgtd	%xmm2, %xmm2
-# CHECK-NEXT: 9.     1     0.0    0.0    9.0       pcmpgtq	%xmm2, %xmm2
-# CHECK-NEXT: 10.    1     0.0    0.0    9.0       pcmpgtw	%xmm2, %xmm2
-# CHECK-NEXT: 11.    1     0.0    0.0    9.0       vpcmpgtb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 12.    1     0.0    0.0    8.0       vpcmpgtd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 13.    1     0.0    0.0    8.0       vpcmpgtq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 14.    1     0.0    0.0    8.0       vpcmpgtw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 15.    1     0.0    0.0    8.0       vpcmpgtb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 16.    1     0.0    0.0    7.0       vpcmpgtd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 17.    1     0.0    0.0    7.0       vpcmpgtq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 18.    1     0.0    0.0    7.0       vpcmpgtw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 19.    1     7.0    0.0    0.0       psubb	%mm2, %mm2
-# CHECK-NEXT: 20.    1     9.0    0.0    0.0       psubd	%mm2, %mm2
-# CHECK-NEXT: 21.    1     12.0   0.0    0.0       psubq	%mm2, %mm2
-# CHECK-NEXT: 22.    1     15.0   0.0    0.0       psubw	%mm2, %mm2
-# CHECK-NEXT: 23.    1     0.0    0.0    18.0      psubb	%xmm2, %xmm2
-# CHECK-NEXT: 24.    1     0.0    0.0    17.0      psubd	%xmm2, %xmm2
-# CHECK-NEXT: 25.    1     0.0    0.0    17.0      psubq	%xmm2, %xmm2
-# CHECK-NEXT: 26.    1     0.0    0.0    17.0      psubw	%xmm2, %xmm2
-# CHECK-NEXT: 27.    1     0.0    0.0    17.0      vpsubb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 28.    1     0.0    0.0    16.0      vpsubd	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 29.    1     0.0    0.0    16.0      vpsubq	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 30.    1     0.0    0.0    16.0      vpsubw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 31.    1     0.0    0.0    16.0      vpsubb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 32.    1     0.0    0.0    15.0      vpsubd	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 33.    1     0.0    0.0    15.0      vpsubq	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 34.    1     0.0    0.0    15.0      vpsubw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 35.    1     15.0   0.0    0.0       psubsb	%mm2, %mm2
-# CHECK-NEXT: 36.    1     17.0   0.0    0.0       psubsw	%mm2, %mm2
-# CHECK-NEXT: 37.    1     1.0    1.0    18.0      psubsb	%xmm2, %xmm2
-# CHECK-NEXT: 38.    1     3.0    1.0    16.0      psubsw	%xmm2, %xmm2
-# CHECK-NEXT: 39.    1     1.0    1.0    18.0      vpsubsb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 40.    1     1.0    0.0    17.0      vpsubsw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 41.    1     2.0    0.0    16.0      vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 42.    1     3.0    1.0    15.0      vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 43.    1     19.0   0.0    0.0       psubusb	%mm2, %mm2
-# CHECK-NEXT: 44.    1     21.0   0.0    0.0       psubusw	%mm2, %mm2
-# CHECK-NEXT: 45.    1     2.0    0.0    21.0      psubusb	%xmm2, %xmm2
-# CHECK-NEXT: 46.    1     3.0    0.0    20.0      psubusw	%xmm2, %xmm2
-# CHECK-NEXT: 47.    1     4.0    3.0    19.0      vpsubusb	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 48.    1     4.0    0.0    18.0      vpsubusw	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 49.    1     5.0    0.0    17.0      vpsubsb	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 50.    1     6.0    1.0    16.0      vpsubsw	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 51.    1     4.0    4.0    18.0      andnps	%xmm0, %xmm0
-# CHECK-NEXT: 52.    1     5.0    5.0    16.0      andnpd	%xmm1, %xmm1
-# CHECK-NEXT: 53.    1     7.0    5.0    14.0      vandnps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT: 54.    1     6.0    0.0    15.0      vandnpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT: 55.    1     22.0   0.0    0.0       pandn	%mm2, %mm2
-# CHECK-NEXT: 56.    1     7.0    0.0    14.0      pandn	%xmm2, %xmm2
-# CHECK-NEXT: 57.    1     3.0    0.0    18.0      vpandn	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 58.    1     8.0    0.0    13.0      vandnps	%xmm2, %xmm2, %xmm5
-# CHECK-NEXT: 59.    1     7.0    1.0    14.0      vandnpd	%xmm1, %xmm1, %xmm5
-# CHECK-NEXT: 60.    1     3.0    0.0    17.0      vpandn	%xmm3, %xmm3, %xmm5
-# CHECK-NEXT: 61.    1     2.0    0.0    19.0      xorps	%xmm0, %xmm0
-# CHECK-NEXT: 62.    1     5.0    0.0    16.0      xorpd	%xmm1, %xmm1
-# CHECK-NEXT: 63.    1     7.0    0.0    14.0      vxorps	%xmm2, %xmm2, %xmm2
-# CHECK-NEXT: 64.    1     4.0    0.0    16.0      vxorpd	%xmm1, %xmm1, %xmm1
-# CHECK-NEXT: 65.    1     20.0   0.0    0.0       pxor	%mm2, %mm2
-# CHECK-NEXT: 66.    1     6.0    0.0    15.0      pxor	%xmm2, %xmm2
-# CHECK-NEXT: 67.    1     2.0    0.0    19.0      vpxor	%xmm3, %xmm3, %xmm3
-# CHECK-NEXT: 68.    1     0.0    0.0    20.0      vxorps	%xmm4, %xmm4, %xmm5
-# CHECK-NEXT: 69.    1     3.0    0.0    17.0      vxorpd	%xmm1, %xmm1, %xmm3
-# CHECK-NEXT: 70.    1     3.0    0.0    17.0      vpxor	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 4.     1     0.0    0.0    0.0       pcmpgtb	%mm2, %mm2
+# CHECK-NEXT: 5.     1     0.0    0.0    0.0       pcmpgtd	%mm2, %mm2
+# CHECK-NEXT: 6.     1     0.0    0.0    0.0       pcmpgtw	%mm2, %mm2
+# CHECK-NEXT: 7.     1     0.0    0.0    0.0       pcmpgtb	%xmm2, %xmm2
+# CHECK-NEXT: 8.     1     0.0    0.0    0.0       pcmpgtd	%xmm2, %xmm2
+# CHECK-NEXT: 9.     1     1.0    1.0    0.0       pcmpgtq	%xmm2, %xmm2
+# CHECK-NEXT: 10.    1     0.0    0.0    3.0       pcmpgtw	%xmm2, %xmm2
+# CHECK-NEXT: 11.    1     0.0    0.0    3.0       vpcmpgtb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 12.    1     0.0    0.0    2.0       vpcmpgtd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 13.    1     1.0    1.0    0.0       vpcmpgtq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 14.    1     0.0    0.0    3.0       vpcmpgtw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 15.    1     0.0    0.0    3.0       vpcmpgtb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 16.    1     0.0    0.0    2.0       vpcmpgtd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 17.    1     1.0    1.0    0.0       vpcmpgtq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 18.    1     0.0    0.0    3.0       vpcmpgtw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 19.    1     0.0    0.0    3.0       psubb	%mm2, %mm2
+# CHECK-NEXT: 20.    1     0.0    0.0    2.0       psubd	%mm2, %mm2
+# CHECK-NEXT: 21.    1     0.0    0.0    3.0       psubq	%mm2, %mm2
+# CHECK-NEXT: 22.    1     0.0    0.0    3.0       psubw	%mm2, %mm2
+# CHECK-NEXT: 23.    1     0.0    0.0    3.0       psubb	%xmm2, %xmm2
+# CHECK-NEXT: 24.    1     0.0    0.0    2.0       psubd	%xmm2, %xmm2
+# CHECK-NEXT: 25.    1     0.0    0.0    3.0       psubq	%xmm2, %xmm2
+# CHECK-NEXT: 26.    1     0.0    0.0    3.0       psubw	%xmm2, %xmm2
+# CHECK-NEXT: 27.    1     0.0    0.0    3.0       vpsubb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 28.    1     0.0    0.0    2.0       vpsubd	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 29.    1     0.0    0.0    3.0       vpsubq	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 30.    1     0.0    0.0    3.0       vpsubw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 31.    1     0.0    0.0    3.0       vpsubb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 32.    1     0.0    0.0    2.0       vpsubd	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 33.    1     0.0    0.0    3.0       vpsubq	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 34.    1     0.0    0.0    3.0       vpsubw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 35.    1     1.0    1.0    0.0       psubsb	%mm2, %mm2
+# CHECK-NEXT: 36.    1     1.0    1.0    0.0       psubsw	%mm2, %mm2
+# CHECK-NEXT: 37.    1     1.0    1.0    0.0       psubsb	%xmm2, %xmm2
+# CHECK-NEXT: 38.    1     2.0    2.0    0.0       psubsw	%xmm2, %xmm2
+# CHECK-NEXT: 39.    1     2.0    2.0    0.0       vpsubsb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 40.    1     2.0    2.0    0.0       vpsubsw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 41.    1     2.0    2.0    0.0       vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 42.    1     3.0    3.0    0.0       vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 43.    1     3.0    3.0    0.0       psubusb	%mm2, %mm2
+# CHECK-NEXT: 44.    1     3.0    3.0    0.0       psubusw	%mm2, %mm2
+# CHECK-NEXT: 45.    1     3.0    3.0    0.0       psubusb	%xmm2, %xmm2
+# CHECK-NEXT: 46.    1     4.0    4.0    0.0       psubusw	%xmm2, %xmm2
+# CHECK-NEXT: 47.    1     4.0    4.0    0.0       vpsubusb	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 48.    1     4.0    4.0    0.0       vpsubusw	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 49.    1     4.0    4.0    0.0       vpsubsb	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 50.    1     5.0    5.0    0.0       vpsubsw	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 51.    1     0.0    0.0    7.0       andnps	%xmm0, %xmm0
+# CHECK-NEXT: 52.    1     0.0    0.0    6.0       andnpd	%xmm1, %xmm1
+# CHECK-NEXT: 53.    1     0.0    0.0    6.0       vandnps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT: 54.    1     0.0    0.0    7.0       vandnpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT: 55.    1     0.0    0.0    7.0       pandn	%mm2, %mm2
+# CHECK-NEXT: 56.    1     0.0    0.0    6.0       pandn	%xmm2, %xmm2
+# CHECK-NEXT: 57.    1     0.0    0.0    6.0       vpandn	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 58.    1     0.0    0.0    7.0       vandnps	%xmm2, %xmm2, %xmm5
+# CHECK-NEXT: 59.    1     0.0    0.0    7.0       vandnpd	%xmm1, %xmm1, %xmm5
+# CHECK-NEXT: 60.    1     0.0    0.0    6.0       vpandn	%xmm3, %xmm3, %xmm5
+# CHECK-NEXT: 61.    1     0.0    0.0    6.0       xorps	%xmm0, %xmm0
+# CHECK-NEXT: 62.    1     0.0    0.0    7.0       xorpd	%xmm1, %xmm1
+# CHECK-NEXT: 63.    1     0.0    0.0    7.0       vxorps	%xmm2, %xmm2, %xmm2
+# CHECK-NEXT: 64.    1     0.0    0.0    6.0       vxorpd	%xmm1, %xmm1, %xmm1
+# CHECK-NEXT: 65.    1     0.0    0.0    6.0       pxor	%mm2, %mm2
+# CHECK-NEXT: 66.    1     0.0    0.0    7.0       pxor	%xmm2, %xmm2
+# CHECK-NEXT: 67.    1     0.0    0.0    7.0       vpxor	%xmm3, %xmm3, %xmm3
+# CHECK-NEXT: 68.    1     0.0    0.0    6.0       vxorps	%xmm4, %xmm4, %xmm5
+# CHECK-NEXT: 69.    1     0.0    0.0    6.0       vxorpd	%xmm1, %xmm1, %xmm3
+# CHECK-NEXT: 70.    1     0.0    0.0    7.0       vpxor	%xmm3, %xmm3, %xmm5
Index: llvm/trunk/test/tools/llvm-mca/X86/bextr-read-after-ld.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/bextr-read-after-ld.s
+++ llvm/trunk/test/tools/llvm-mca/X86/bextr-read-after-ld.s
@@ -2,7 +2,7 @@
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=haswell -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=HASWELL
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=broadwell -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDWELL
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=SKYLAKE
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BTVER2
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -timeline -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=ZNVER1
 
@@ -12,8 +12,8 @@
 # ALL:          Iterations:        1
 # ALL-NEXT:     Instructions:      2
 
-# BDVER2-NEXT:  Total Cycles:      10
-# BDVER2-NEXT:  Total uOps:        4
+# BDVER2-NEXT:  Total Cycles:      9
+# BDVER2-NEXT:  Total uOps:        3
 
 # BDWELL-NEXT:  Total Cycles:      10
 # BDWELL-NEXT:  Total uOps:        4
@@ -31,8 +31,8 @@
 # ZNVER1-NEXT:  Total uOps:        3
 
 # BDVER2:       Dispatch Width:    4
-# BDVER2-NEXT:  uOps Per Cycle:    0.40
-# BDVER2-NEXT:  IPC:               0.20
+# BDVER2-NEXT:  uOps Per Cycle:    0.33
+# BDVER2-NEXT:  IPC:               0.22
 # BDVER2-NEXT:  Block RThroughput: 1.0
 
 # BDWELL:       Dispatch Width:    4
@@ -70,8 +70,8 @@
 
 # ALL:          [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 
-# BDVER2-NEXT:   1      1     0.33                        addl	%edi, %esi
-# BDVER2-NEXT:   3      7     1.00    *                   bextrl	%esi, (%rdi), %eax
+# BDVER2-NEXT:   1      1     0.50                        addl	%edi, %esi
+# BDVER2-NEXT:   2      6     0.50    *                   bextrl	%esi, (%rdi), %eax
 
 # BDWELL-NEXT:   1      1     0.25                        addl	%edi, %esi
 # BDWELL-NEXT:   3      7     0.50    *                   bextrl	%esi, (%rdi), %eax
@@ -90,15 +90,15 @@
 
 # ALL:          Timeline view:
 
-# BDVER2-NEXT:  Index     0123456789
+# BDVER2-NEXT:  Index     012345678
 # BDWELL-NEXT:  Index     0123456789
 # BTVER2-NEXT:  Index     0123456
 # HASWELL-NEXT: Index     0123456789
 # SKYLAKE-NEXT: Index     0123456789
 # ZNVER1-NEXT:  Index     01234567
 
-# BDVER2:       [0,0]     DeER .   .   addl	%edi, %esi
-# BDVER2-NEXT:  [0,1]     DeeeeeeeER   bextrl	%esi, (%rdi), %eax
+# BDVER2:       [0,0]     DeER .  .   addl	%edi, %esi
+# BDVER2-NEXT:  [0,1]     DeeeeeeER   bextrl	%esi, (%rdi), %eax
 
 # BDWELL:       [0,0]     DeER .   .   addl	%edi, %esi
 # BDWELL-NEXT:  [0,1]     DeeeeeeeER   bextrl	%esi, (%rdi), %eax
Index: llvm/trunk/test/tools/llvm-mca/X86/cpus.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/cpus.s
+++ llvm/trunk/test/tools/llvm-mca/X86/cpus.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -resource-pressure=false -instruction-info=false < %s | FileCheck --check-prefix=ALL --check-prefix=BDVER2 %s
+# RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -resource-pressure=false -instruction-info=false < %s | FileCheck --check-prefix=ALL --check-prefix=BDVER2 %s
 # RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=btver2 -resource-pressure=false -instruction-info=false < %s | FileCheck --check-prefix=ALL --check-prefix=BTVER2 %s
 # RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=znver1 -resource-pressure=false -instruction-info=false < %s | FileCheck --check-prefix=ALL --check-prefix=ZNVER1 %s
 # RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -resource-pressure=false -instruction-info=false < %s | FileCheck --check-prefix=ALL --check-prefix=SANDYBRIDGE %s
@@ -21,7 +21,7 @@
 # BDVER2:           Dispatch Width:    4
 # BDVER2-NEXT:      uOps Per Cycle:    0.97
 # BDVER2-NEXT:      IPC:               0.97
-# BDVER2-NEXT:      Block RThroughput: 0.3
+# BDVER2-NEXT:      Block RThroughput: 0.5
 
 # BROADWELL:        Dispatch Width:    4
 # BROADWELL-NEXT:   uOps Per Cycle:    0.97
Index: llvm/trunk/test/tools/llvm-mca/X86/read-after-ld-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/read-after-ld-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/read-after-ld-1.s
@@ -3,7 +3,7 @@
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=haswell -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=HASWELL
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=broadwell -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDWELL
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=SKYLAKE
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BTVER2
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=ZNVER1
 
@@ -13,8 +13,8 @@
 # ALL:          Iterations:        1
 # ALL-NEXT:     Instructions:      2
 
-# BDVER2-NEXT:  Total Cycles:      20
-# BDVER2-NEXT:  Total uOps:        3
+# BDVER2-NEXT:  Total Cycles:      17
+# BDVER2-NEXT:  Total uOps:        2
 
 # BDWELL-NEXT:  Total Cycles:      17
 # BDWELL-NEXT:  Total uOps:        3
@@ -35,9 +35,9 @@
 # ZNVER1-NEXT:  Total uOps:        2
 
 # BDVER2:       Dispatch Width:    4
-# BDVER2-NEXT:  uOps Per Cycle:    0.15
-# BDVER2-NEXT:  IPC:               0.10
-# BDVER2-NEXT:  Block RThroughput: 14.0
+# BDVER2-NEXT:  uOps Per Cycle:    0.12
+# BDVER2-NEXT:  IPC:               0.12
+# BDVER2-NEXT:  Block RThroughput: 10.0
 
 # BDWELL:       Dispatch Width:    4
 # BDWELL-NEXT:  uOps Per Cycle:    0.18
@@ -71,7 +71,7 @@
 
 # ALL:          Timeline view:
 
-# BDVER2-NEXT:                      0123456789
+# BDVER2-NEXT:                      0123456
 # BDVER2-NEXT:  Index     0123456789
 
 # BDWELL-NEXT:                      0123456
@@ -92,8 +92,8 @@
 # ZNVER1-NEXT:                      0123456789
 # ZNVER1-NEXT:  Index     0123456789
 
-# BDVER2:       [0,0]     DeeeeeeeeeeeeeeER  .   vdivps	%xmm0, %xmm1, %xmm1
-# BDVER2-NEXT:  [0,1]     D========eeeeeeeeeER   vaddps	(%rax), %xmm1, %xmm1
+# BDVER2:       [0,0]     DeeeeeeeeeER   ..   vdivps	%xmm0, %xmm1, %xmm1
+# BDVER2-NEXT:  [0,1]     D====eeeeeeeeeeER   vaddps	(%rax), %xmm1, %xmm1
 
 # BDWELL:       [0,0]     DeeeeeeeeeeeER ..   vdivps	%xmm0, %xmm1, %xmm1
 # BDWELL-NEXT:  [0,1]     D======eeeeeeeeER   vaddps	(%rax), %xmm1, %xmm1
@@ -122,7 +122,7 @@
 # ALL:                [0]    [1]    [2]    [3]
 # ALL-NEXT:     0.     1     1.0    1.0    0.0       vdivps	%xmm0, %xmm1, %xmm1
 
-# BDVER2-NEXT:  1.     1     9.0    0.0    0.0       vaddps	(%rax), %xmm1, %xmm1
+# BDVER2-NEXT:  1.     1     5.0    0.0    0.0       vaddps	(%rax), %xmm1, %xmm1
 # BDWELL-NEXT:  1.     1     7.0    0.0    0.0       vaddps	(%rax), %xmm1, %xmm1
 # BTVER2-NEXT:  1.     1     15.0   0.0    0.0       vaddps	(%rax), %xmm1, %xmm1
 # HASWELL-NEXT: 1.     1     8.0    0.0    0.0       vaddps	(%rax), %xmm1, %xmm1
Index: llvm/trunk/test/tools/llvm-mca/X86/scheduler-queue-usage.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/scheduler-queue-usage.s
+++ llvm/trunk/test/tools/llvm-mca/X86/scheduler-queue-usage.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -all-stats=false -all-views=false -scheduler-stats < %s | FileCheck --check-prefixes=ALL,BDVER2 %s
+# RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -all-stats=false -all-views=false -scheduler-stats < %s | FileCheck --check-prefixes=ALL,BDVER2 %s
 # RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -all-stats=false -all-views=false -scheduler-stats < %s | FileCheck --check-prefixes=ALL,BTVER2 %s
 # RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -all-stats=false -all-views=false -scheduler-stats < %s | FileCheck --check-prefixes=ALL,ZNVER1 %s
 # RUN: llvm-mca %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -iterations=1 -all-stats=false -all-views=false -scheduler-stats < %s | FileCheck --check-prefixes=ALL,SNB %s
@@ -82,7 +82,10 @@
 # ZNVER1-NEXT:     [4] Total number of buffer entries.
 
 # BDVER2:           [1]            [2]        [3]        [4]
-# BDVER2-NEXT:     SBPortAny        0          1          54
+# BDVER2-NEXT:     PdEX             0          1          40
+# BDVER2-NEXT:     PdFPU            0          0          64
+# BDVER2-NEXT:     PdLoad           0          0          40
+# BDVER2-NEXT:     PdStore          0          0          24
 
 # BDW:              [1]            [2]        [3]        [4]
 # BDW-NEXT:        BWPortAny        0          1          60
Index: llvm/trunk/test/tools/llvm-mca/X86/sqrt-rsqrt-rcp-memop.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/sqrt-rsqrt-rcp-memop.s
+++ llvm/trunk/test/tools/llvm-mca/X86/sqrt-rsqrt-rcp-memop.s
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
-# ZZZ: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BTVER2
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=ZNVER1
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=haswell -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=HASWELL
@@ -30,6 +30,9 @@
 
 # ALL:            Timeline view:
 
+# BDVER2-NEXT:                        01234567
+# BDVER2-NEXT:    Index     0123456789
+
 # BROADWELL-NEXT:                     0123456789
 # BROADWELL-NEXT: Index     0123456789
 
@@ -45,6 +48,9 @@
 # ZNVER1-NEXT:                        0123456789          0
 # ZNVER1-NEXT:    Index     0123456789          0123456789
 
+# BDVER2:         [0,0]     DeER .    .    . .   leaq	8(%rsp,%rdi,2), %rax
+# BDVER2-NEXT:    [0,1]     D=eeeeeeeeeeeeeeER   sqrtss	(%rax), %xmm1
+
 # BROADWELL:      [0,0]     DeER .    .    .   .   leaq	8(%rsp,%rdi,2), %rax
 # BROADWELL-NEXT: [0,1]     D=eeeeeeeeeeeeeeeeER   sqrtss	(%rax), %xmm1
 
@@ -69,6 +75,7 @@
 # ALL:                  [0]    [1]    [2]    [3]
 # ALL-NEXT:       0.     1     1.0    1.0    0.0       leaq	8(%rsp,%rdi,2), %rax
 
+# BDVER2-NEXT:    1.     1     2.0    0.0    0.0       sqrtss	(%rax), %xmm1
 # BROADWELL-NEXT: 1.     1     2.0    0.0    0.0       sqrtss	(%rax), %xmm1
 # BTVER2-NEXT:    1.     1     3.0    0.0    0.0       sqrtss	(%rax), %xmm1
 # HASWELL-NEXT:   1.     1     2.0    0.0    0.0       sqrtss	(%rax), %xmm1
@@ -79,6 +86,9 @@
 
 # ALL:            Timeline view:
 
+# BDVER2-NEXT:                        01234567
+# BDVER2-NEXT:    Index     0123456789
+
 # BROADWELL-NEXT:                     0123456789
 # BROADWELL-NEXT: Index     0123456789          01234
 
@@ -94,6 +104,9 @@
 # ZNVER1-NEXT:                        0123456789          0
 # ZNVER1-NEXT:    Index     0123456789          0123456789
 
+# BDVER2:         [0,0]     DeER .    .    . .   leaq	8(%rsp,%rdi,2), %rax
+# BDVER2-NEXT:    [0,1]     D=eeeeeeeeeeeeeeER   sqrtsd	(%rax), %xmm1
+
 # BROADWELL:      [0,0]     DeER .    .    .    .   .   leaq	8(%rsp,%rdi,2), %rax
 # BROADWELL-NEXT: [0,1]     D=eeeeeeeeeeeeeeeeeeeeeER   sqrtsd	(%rax), %xmm1
 
@@ -118,6 +131,7 @@
 # ALL:                  [0]    [1]    [2]    [3]
 # ALL-NEXT:       0.     1     1.0    1.0    0.0       leaq	8(%rsp,%rdi,2), %rax
 
+# BDVER2-NEXT:    1.     1     2.0    0.0    0.0       sqrtsd	(%rax), %xmm1
 # BROADWELL-NEXT: 1.     1     2.0    0.0    0.0       sqrtsd	(%rax), %xmm1
 # BTVER2-NEXT:    1.     1     3.0    0.0    0.0       sqrtsd	(%rax), %xmm1
 # HASWELL-NEXT:   1.     1     2.0    0.0    0.0       sqrtsd	(%rax), %xmm1
@@ -128,6 +142,7 @@
 
 # ALL:            Timeline view:
 
+# BDVER2-NEXT:                        0123
 # BROADWELL-NEXT:                     0123
 # BTVER2-NEXT:                        01
 # HASWELL-NEXT:                       0123
@@ -136,6 +151,9 @@
 
 # ALL-NEXT:       Index     0123456789
 
+# BDVER2:         [0,0]     DeER .    .  .   leaq	8(%rsp,%rdi,2), %rax
+# BDVER2-NEXT:    [0,1]     D=eeeeeeeeeeER   rsqrtss	(%rax), %xmm1
+
 # BROADWELL:      [0,0]     DeER .    .  .   leaq	8(%rsp,%rdi,2), %rax
 # BROADWELL-NEXT: [0,1]     D=eeeeeeeeeeER   rsqrtss	(%rax), %xmm1
 
@@ -160,6 +178,7 @@
 # ALL:                  [0]    [1]    [2]    [3]
 # ALL-NEXT:       0.     1     1.0    1.0    0.0       leaq	8(%rsp,%rdi,2), %rax
 
+# BDVER2-NEXT:    1.     1     2.0    0.0    0.0       rsqrtss	(%rax), %xmm1
 # BROADWELL-NEXT: 1.     1     2.0    0.0    0.0       rsqrtss	(%rax), %xmm1
 # BTVER2-NEXT:    1.     1     3.0    0.0    0.0       rsqrtss	(%rax), %xmm1
 # HASWELL-NEXT:   1.     1     2.0    0.0    0.0       rsqrtss	(%rax), %xmm1
@@ -170,6 +189,7 @@
 
 # ALL:            Timeline view:
 
+# BDVER2-NEXT:                        0123
 # BROADWELL-NEXT:                     0123
 # BTVER2-NEXT:                        01
 # HASWELL-NEXT:                       0123
@@ -178,6 +198,9 @@
 
 # ALL-NEXT:       Index     0123456789
 
+# BDVER2:         [0,0]     DeER .    .  .   leaq	8(%rsp,%rdi,2), %rax
+# BDVER2-NEXT:    [0,1]     D=eeeeeeeeeeER   rcpss	(%rax), %xmm1
+
 # BROADWELL:      [0,0]     DeER .    .  .   leaq	8(%rsp,%rdi,2), %rax
 # BROADWELL-NEXT: [0,1]     D=eeeeeeeeeeER   rcpss	(%rax), %xmm1
 
@@ -202,6 +225,7 @@
 # ALL:                  [0]    [1]    [2]    [3]
 # ALL-NEXT:       0.     1     1.0    1.0    0.0       leaq	8(%rsp,%rdi,2), %rax
 
+# BDVER2-NEXT:    1.     1     2.0    0.0    0.0       rcpss	(%rax), %xmm1
 # BROADWELL-NEXT: 1.     1     2.0    0.0    0.0       rcpss	(%rax), %xmm1
 # BTVER2-NEXT:    1.     1     3.0    0.0    0.0       rcpss	(%rax), %xmm1
 # HASWELL-NEXT:   1.     1     2.0    0.0    0.0       rcpss	(%rax), %xmm1
Index: llvm/trunk/test/tools/llvm-mca/X86/variable-blend-read-after-ld-1.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/variable-blend-read-after-ld-1.s
+++ llvm/trunk/test/tools/llvm-mca/X86/variable-blend-read-after-ld-1.s
@@ -9,7 +9,7 @@
 
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=SKYLAKE
 
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
 
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BTVER2
 
@@ -21,8 +21,8 @@
 # ALL:          Iterations:        1
 # ALL-NEXT:     Instructions:      2
 
-# BDVER2-NEXT:  Total Cycles:      11
-# BDVER2-NEXT:  Total uOps:        4
+# BDVER2-NEXT:  Total Cycles:      10
+# BDVER2-NEXT:  Total uOps:        2
 
 # BDWELL-NEXT:  Total Cycles:      10
 # BDWELL-NEXT:  Total uOps:        4
@@ -46,9 +46,9 @@
 # ZNVER1-NEXT:  Total uOps:        2
 
 # BDVER2:       Dispatch Width:    4
-# BDVER2-NEXT:  uOps Per Cycle:    0.36
-# BDVER2-NEXT:  IPC:               0.18
-# BDVER2-NEXT:  Block RThroughput: 1.0
+# BDVER2-NEXT:  uOps Per Cycle:    0.20
+# BDVER2-NEXT:  IPC:               0.20
+# BDVER2-NEXT:  Block RThroughput: 2.5
 
 # BDWELL:       Dispatch Width:    4
 # BDWELL-NEXT:  uOps Per Cycle:    0.40
@@ -86,7 +86,6 @@
 # ZNVER1-NEXT:  Block RThroughput: 1.0
 
 # BDVER2:       Timeline view:
-# BDVER2-NEXT:                      0
 # BDVER2-NEXT:  Index     0123456789
 
 # BDWELL:       Timeline view:
@@ -116,8 +115,8 @@
 # ZNVER1-NEXT:                      0
 # ZNVER1-NEXT:  Index     0123456789
 
-# BDVER2:       [0,0]     DeeeER    .   vaddps	%xmm0, %xmm0, %xmm1
-# BDVER2-NEXT:  [0,1]     DeeeeeeeeER   vblendvps	%xmm1, (%rdi), %xmm2, %xmm3
+# BDVER2:       [0,0]     DeeeeeER .   vaddps	%xmm0, %xmm0, %xmm1
+# BDVER2-NEXT:  [0,1]     DeeeeeeeER   vblendvps	%xmm1, (%rdi), %xmm2, %xmm3
 
 # BDWELL:       [0,0]     DeeeER   .   vaddps	%xmm0, %xmm0, %xmm1
 # BDWELL-NEXT:  [0,1]     DeeeeeeeER   vblendvps	%xmm1, (%rdi), %xmm2, %xmm3
Index: llvm/trunk/test/tools/llvm-mca/X86/variable-blend-read-after-ld-2.s
===================================================================
--- llvm/trunk/test/tools/llvm-mca/X86/variable-blend-read-after-ld-2.s
+++ llvm/trunk/test/tools/llvm-mca/X86/variable-blend-read-after-ld-2.s
@@ -9,7 +9,7 @@
 
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=SKYLAKE
 
-# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
 
 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefix=ALL -check-prefix=BTVER2
 
@@ -21,8 +21,8 @@
 # ALL:          Iterations:        1
 # ALL-NEXT:     Instructions:      2
 
-# BDVER2-NEXT:  Total Cycles:      11
-# BDVER2-NEXT:  Total uOps:        4
+# BDVER2-NEXT:  Total Cycles:      10
+# BDVER2-NEXT:  Total uOps:        2
 
 # BDWELL-NEXT:  Total Cycles:      10
 # BDWELL-NEXT:  Total uOps:        4
@@ -46,9 +46,9 @@
 # ZNVER1-NEXT:  Total uOps:        2
 
 # BDVER2:       Dispatch Width:    4
-# BDVER2-NEXT:  uOps Per Cycle:    0.36
-# BDVER2-NEXT:  IPC:               0.18
-# BDVER2-NEXT:  Block RThroughput: 1.0
+# BDVER2-NEXT:  uOps Per Cycle:    0.20
+# BDVER2-NEXT:  IPC:               0.20
+# BDVER2-NEXT:  Block RThroughput: 2.5
 
 # BDWELL:       Dispatch Width:    4
 # BDWELL-NEXT:  uOps Per Cycle:    0.40
@@ -86,7 +86,6 @@
 # ZNVER1-NEXT:  Block RThroughput: 1.0
 
 # BDVER2:       Timeline view:
-# BDVER2-NEXT:                      0
 # BDVER2-NEXT:  Index     0123456789
 
 # BDWELL:       Timeline view:
@@ -116,8 +115,8 @@
 # ZNVER1-NEXT:                      0
 # ZNVER1-NEXT:  Index     0123456789
 
-# BDVER2:       [0,0]     DeeeER    .   vaddps	%xmm0, %xmm0, %xmm2
-# BDVER2-NEXT:  [0,1]     DeeeeeeeeER   vblendvps	%xmm1, (%rdi), %xmm2, %xmm3
+# BDVER2:       [0,0]     DeeeeeER .   vaddps	%xmm0, %xmm0, %xmm2
+# BDVER2-NEXT:  [0,1]     DeeeeeeeER   vblendvps	%xmm1, (%rdi), %xmm2, %xmm3
 
 # BDWELL:       [0,0]     DeeeER   .   vaddps	%xmm0, %xmm0, %xmm2
 # BDWELL-NEXT:  [0,1]     DeeeeeeeER   vblendvps	%xmm1, (%rdi), %xmm2, %xmm3