[X86] Fix VZEROUPPER scheduling info on SNB,HSW,BDW,SXL,SKX.

[courbet]

Summary

Starting from SNB, VZEROUPPER is handled by the renamer and uses no proc resources.
After HSW, it also has zero latency.

This fixes PR35606.

To reproduce:
Uops:

llvm-exegesis -mode=uops -opcode-name=VZEROUPPER

Latency:

echo -e '#LLVM-EXEGESIS-DEFREG XMM0 1\n#LLVM-EXEGESIS-DEFREG XMM1 1\nvzeroupper' | /tmp/llvm-exegesis -mode=latency -snippets-file=-
echo -e '#LLVM-EXEGESIS-DEFREG XMM0 1\n#LLVM-EXEGESIS-DEFREG XMM1 1\nvzeroupper\naddps %xmm0, %xmm1' | /tmp/llvm-exegesis -mode=latency -snippets-file=-

[RKSimon]

tools/llvm-mca/lib/Stages/Decode.cpp

1 ↗

(On Diff #172595)

Typo?

[courbet]

remove unrelated file

[courbet]

tools/llvm-mca/lib/Stages/Decode.cpp

1 ↗

(On Diff #172595)

oops, yes.

[RKSimon]

LGTM

llvm/trunk/lib/Target/X86/X86SchedBroadwell.td

 
 def BWWriteResGroup45 : SchedWriteRes<[BWPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [4];
 }
 def: InstRW<[BWWriteResGroup45], (instrs FNCLEX)>;
 
-def BWWriteResGroup46 : SchedWriteRes<[BWPort015,BWPort0156]> {
-  let Latency = 4;
+def BWWriteResGroup46 : SchedWriteRes<[]> {
+  let Latency = 0;
   let NumMicroOps = 4;
-  let ResourceCycles = [1,3];
+  let ResourceCycles = [];
 }
 def: InstRW<[BWWriteResGroup46], (instrs VZEROUPPER)>;
 
 def BWWriteResGroup47 : SchedWriteRes<[BWPort0]> {
   let Latency = 5;
   let NumMicroOps = 1;
   let ResourceCycles = [1];
 }

llvm/trunk/lib/Target/X86/X86SchedHaswell.td

 
 def HWWriteResGroup81 : SchedWriteRes<[HWPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [4];
 }
 def: InstRW<[HWWriteResGroup81], (instrs FNCLEX)>;
 
-def HWWriteResGroup82 : SchedWriteRes<[HWPort015,HWPort0156]> {
-  let Latency = 4;
+def HWWriteResGroup82 : SchedWriteRes<[]> {
+  let Latency = 0;
   let NumMicroOps = 4;
-  let ResourceCycles = [1,3];
+  let ResourceCycles = [];
 }
 def: InstRW<[HWWriteResGroup82], (instrs VZEROUPPER)>;
 
 def HWWriteResGroup83 : SchedWriteRes<[HWPort1,HWPort6,HWPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [1,1,2];
 }

llvm/trunk/lib/Target/X86/X86SchedSandyBridge.td

 
 def SBWriteResGroupVzeroall : SchedWriteRes<[SBPort5]> {
   let Latency = 9;
   let NumMicroOps = 20;
   let ResourceCycles = [2];
 }
 def: InstRW<[SBWriteResGroupVzeroall], (instrs VZEROALL)>;
 
+def SBWriteResGroupVzeroupper : SchedWriteRes<[]> {
+  let Latency = 1;
+  let NumMicroOps = 4;
+  let ResourceCycles = [];
+}
+def: InstRW<[SBWriteResGroupVzeroupper], (instrs VZEROUPPER)>;
+
 def: InstRW<[WriteZero], (instrs CLC)>;
 
 // Intruction variants handled by the renamer. These might not need execution
 // ports in certain conditions.
 // See Agner's Fog "The microarchitecture of Intel, AMD and VIA CPUs",
 // section "Sandy Bridge and Ivy Bridge Pipeline" > "Register allocation and
 // renaming".
 // These can be investigated with llvm-exegesis, e.g.

llvm/trunk/lib/Target/X86/X86SchedSkylakeClient.td

 
 def SKLWriteResGroup55 : SchedWriteRes<[SKLPort6,SKLPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [1,3];
 }
 def: InstRW<[SKLWriteResGroup55], (instrs PAUSE)>;
 
-def SKLWriteResGroup56 : SchedWriteRes<[SKLPort015,SKLPort0156]> {
-  let Latency = 4;
+def SKLWriteResGroup56 : SchedWriteRes<[]> {
+  let Latency = 0;
   let NumMicroOps = 4;
-  let ResourceCycles = [1,3];
+  let ResourceCycles = [];
 }
 def: InstRW<[SKLWriteResGroup56], (instrs VZEROUPPER)>;
 
 def SKLWriteResGroup57 : SchedWriteRes<[SKLPort1,SKLPort6,SKLPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [1,1,2];
 }

llvm/trunk/lib/Target/X86/X86SchedSkylakeServer.td

 
 def SKXWriteResGroup55 : SchedWriteRes<[SKXPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [4];
 }
 def: InstRW<[SKXWriteResGroup55], (instrs FNCLEX)>;
 
-def SKXWriteResGroup56 : SchedWriteRes<[SKXPort015,SKXPort0156]> {
-  let Latency = 4;
+def SKXWriteResGroup56 : SchedWriteRes<[]> {
+  let Latency = 0;
   let NumMicroOps = 4;
-  let ResourceCycles = [1,3];
+  let ResourceCycles = [];
 }
 def: InstRW<[SKXWriteResGroup56], (instrs VZEROUPPER)>;
 
 def SKXWriteResGroup57 : SchedWriteRes<[SKXPort1,SKXPort6,SKXPort0156]> {
   let Latency = 4;
   let NumMicroOps = 4;
   let ResourceCycles = [1,1,2];
 }

llvm/trunk/test/CodeGen/X86/avx-schedule.ll

 }
 declare <8 x float> @llvm.x86.avx.dp.ps.256(<8 x float>, <8 x float>, i8) nounwind readnone
 
 define <4 x float> @test_extractf128(<8 x float> %a0, <8 x float> %a1, <4 x float> *%a2) {
 ; GENERIC-LABEL: test_extractf128:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_extractf128:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [1:1.00]
 ; SANDY-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_extractf128:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [3:1.00]
 ; HASWELL-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_extractf128:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [3:1.00]
 ; BROADWELL-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_extractf128:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [3:1.00]
 ; SKYLAKE-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_extractf128:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vextractf128 $1, %ymm0, %xmm0 # sched: [3:1.00]
 ; SKX-NEXT:    vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_extractf128:
 ; BDVER2:       # %bb.0:
 ; 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]
; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %4 = fadd <4 x double> %1, %3
   ret <4 x double> %4
 }
 
 define i32 @test_movmskpd(<4 x double> %a0) {
 ; GENERIC-LABEL: test_movmskpd:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovmskpd %ymm0, %eax # sched: [2:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_movmskpd:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    vmovmskpd %ymm0, %eax # sched: [2:1.00]
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_movmskpd:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vmovmskpd %ymm0, %eax # sched: [3:1.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_movmskpd:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vmovmskpd %ymm0, %eax # sched: [3:1.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_movmskpd:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vmovmskpd %ymm0, %eax # sched: [2:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_movmskpd:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovmskpd %ymm0, %eax # sched: [2:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_movmskpd:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmovmskpd %ymm0, %eax # sched: [10:1.00]
 ; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
; ZNVER1-NEXT:    retq # sched: [1:0.50]
   ret i32 %1
 }
 declare i32 @llvm.x86.avx.movmsk.pd.256(<4 x double>) nounwind readnone
 
 define i32 @test_movmskps(<8 x float> %a0) {
 ; GENERIC-LABEL: test_movmskps:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovmskps %ymm0, %eax # sched: [2:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_movmskps:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    vmovmskps %ymm0, %eax # sched: [2:1.00]
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_movmskps:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vmovmskps %ymm0, %eax # sched: [3:1.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_movmskps:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vmovmskps %ymm0, %eax # sched: [3:1.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_movmskps:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vmovmskps %ymm0, %eax # sched: [2:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_movmskps:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovmskps %ymm0, %eax # sched: [2:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_movmskps:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vmovmskps %ymm0, %eax # sched: [10:1.00]
 ; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;

 declare i32 @llvm.x86.avx.movmsk.ps.256(<8 x float>) nounwind readnone
 
 define void @test_movntdq(<4 x i64> %a0, <4 x i64> *%a1) {
 ; GENERIC-LABEL: test_movntdq:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_movntdq:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    #APP
 ; SANDY-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
 ; SANDY-NEXT:    #NO_APP
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_movntdq:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_movntdq:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_movntdq:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_movntdq:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [1:1.00]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_movntdq:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; BDVER2-NEXT:    vmovntdq %ymm0, (%rdi) # sched: [2:2.00]
 ; BDVER2-NEXT:    #NO_APP
 ; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]

 define i32 @test_testpd_ymm(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
 ; GENERIC-LABEL: test_testpd_ymm:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; GENERIC-NEXT:    vtestpd %ymm1, %ymm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    setb %al # sched: [1:0.50]
 ; GENERIC-NEXT:    vtestpd (%rdi), %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_testpd_ymm:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; SANDY-NEXT:    vtestpd %ymm1, %ymm0 # sched: [1:1.00]
 ; SANDY-NEXT:    setb %al # sched: [1:0.50]
 ; SANDY-NEXT:    vtestpd (%rdi), %ymm0 # sched: [8:1.00]
 ; SANDY-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_testpd_ymm:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; HASWELL-NEXT:    vtestpd %ymm1, %ymm0 # sched: [1:1.00]
 ; HASWELL-NEXT:    setb %al # sched: [1:0.50]
 ; HASWELL-NEXT:    vtestpd (%rdi), %ymm0 # sched: [8:1.00]
 ; HASWELL-NEXT:    adcl $0, %eax # sched: [2:0.50]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_testpd_ymm:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; BROADWELL-NEXT:    vtestpd %ymm1, %ymm0 # sched: [1:1.00]
 ; BROADWELL-NEXT:    setb %al # sched: [1:0.50]
 ; BROADWELL-NEXT:    vtestpd (%rdi), %ymm0 # sched: [7:1.00]
 ; BROADWELL-NEXT:    adcl $0, %eax # sched: [1:0.50]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_testpd_ymm:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; SKYLAKE-NEXT:    vtestpd %ymm1, %ymm0 # sched: [2:1.00]
 ; SKYLAKE-NEXT:    setb %al # sched: [1:0.50]
 ; SKYLAKE-NEXT:    vtestpd (%rdi), %ymm0 # sched: [9:1.00]
 ; SKYLAKE-NEXT:    adcl $0, %eax # sched: [1:0.50]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_testpd_ymm:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; SKX-NEXT:    vtestpd %ymm1, %ymm0 # sched: [2:1.00]
 ; SKX-NEXT:    setb %al # sched: [1:0.50]
 ; SKX-NEXT:    vtestpd (%rdi), %ymm0 # sched: [9:1.00]
 ; SKX-NEXT:    adcl $0, %eax # sched: [1:0.50]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_testpd_ymm:
 ; BDVER2:       # %bb.0:
 ; 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: [6:1.00]

 define i32 @test_testps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
 ; GENERIC-LABEL: test_testps_ymm:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; GENERIC-NEXT:    vtestps %ymm1, %ymm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    setb %al # sched: [1:0.50]
 ; GENERIC-NEXT:    vtestps (%rdi), %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_testps_ymm:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    xorl %eax, %eax # sched: [0:0.25]
 ; SANDY-NEXT:    vtestps %ymm1, %ymm0 # sched: [1:1.00]
 ; SANDY-NEXT:    setb %al # sched: [1:0.50]
 ; SANDY-NEXT:    vtestps (%rdi), %ymm0 # sched: [8:1.00]
 ; SANDY-NEXT:    adcl $0, %eax # sched: [2:0.67]
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_testps_ymm:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; HASWELL-NEXT:    vtestps %ymm1, %ymm0 # sched: [1:1.00]
 ; HASWELL-NEXT:    setb %al # sched: [1:0.50]
 ; HASWELL-NEXT:    vtestps (%rdi), %ymm0 # sched: [8:1.00]
 ; HASWELL-NEXT:    adcl $0, %eax # sched: [2:0.50]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_testps_ymm:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; BROADWELL-NEXT:    vtestps %ymm1, %ymm0 # sched: [1:1.00]
 ; BROADWELL-NEXT:    setb %al # sched: [1:0.50]
 ; BROADWELL-NEXT:    vtestps (%rdi), %ymm0 # sched: [7:1.00]
 ; BROADWELL-NEXT:    adcl $0, %eax # sched: [1:0.50]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_testps_ymm:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; SKYLAKE-NEXT:    vtestps %ymm1, %ymm0 # sched: [2:1.00]
 ; SKYLAKE-NEXT:    setb %al # sched: [1:0.50]
 ; SKYLAKE-NEXT:    vtestps (%rdi), %ymm0 # sched: [9:1.00]
 ; SKYLAKE-NEXT:    adcl $0, %eax # sched: [1:0.50]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_testps_ymm:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    xorl %eax, %eax # sched: [1:0.25]
 ; SKX-NEXT:    vtestps %ymm1, %ymm0 # sched: [2:1.00]
 ; SKX-NEXT:    setb %al # sched: [1:0.50]
 ; SKX-NEXT:    vtestps (%rdi), %ymm0 # sched: [9:1.00]
 ; SKX-NEXT:    adcl $0, %eax # sched: [1:0.50]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_testps_ymm:
 ; BDVER2:       # %bb.0:
 ; 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: [6:1.00]
; ZNVER1-NEXT:    retq # sched: [1:0.50]
   call void @llvm.x86.avx.vzeroall()
   ret void
 }
 declare void @llvm.x86.avx.vzeroall() nounwind
 
 define void @test_zeroupper() {
 ; GENERIC-LABEL: test_zeroupper:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_zeroupper:
 ; SANDY:       # %bb.0:
-; SANDY-NEXT:    vzeroupper # sched: [100:0.33]
+; SANDY-NEXT:    vzeroupper # sched: [1:1.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_zeroupper:
 ; HASWELL:       # %bb.0:
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_zeroupper:
 ; BROADWELL:       # %bb.0:
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_zeroupper:
 ; SKYLAKE:       # %bb.0:
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_zeroupper:
 ; SKX:       # %bb.0:
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_zeroupper:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    vzeroupper # sched: [46:4.00]
 ; BDVER2-NEXT:    retq # sched: [5:1.00]
 ;
 ; BTVER2-LABEL: test_zeroupper:

llvm/trunk/test/CodeGen/X86/avx2-schedule.ll

 }
 
 define <4 x i32> @test_extracti128(<8 x i16> %a0, <4 x i32> *%a1) {
 ; GENERIC-LABEL: test_extracti128:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpmovzxwd {{.*#+}} ymm1 = 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: [1:1.00]
 ; GENERIC-NEXT:    vextracti128 $1, %ymm1, %xmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vextracti128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_extracti128:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = 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: [3:1.00]
 ; HASWELL-NEXT:    vextracti128 $1, %ymm1, %xmm0 # sched: [3:1.00]
 ; HASWELL-NEXT:    vextracti128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_extracti128:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = 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: [3:1.00]
 ; BROADWELL-NEXT:    vextracti128 $1, %ymm1, %xmm0 # sched: [3:1.00]
 ; BROADWELL-NEXT:    vextracti128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_extracti128:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vpmovzxwd {{.*#+}} ymm1 = 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: [3:1.00]
 ; SKYLAKE-NEXT:    vextracti128 $1, %ymm1, %xmm0 # sched: [3:1.00]
 ; SKYLAKE-NEXT:    vextracti128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_extracti128:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpmovzxwd {{.*#+}} ymm1 = 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: [3:1.00]
 ; SKX-NEXT:    vextracti128 $1, %ymm1, %xmm0 # sched: [3:1.00]
 ; SKX-NEXT:    vextracti128 $1, %ymm1, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_extracti128:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpmovzxwd {{.*#+}} ymm1 = 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]
 ; ZNVER1-NEXT:    vextracti128 $1, %ymm1, %xmm0 # sched: [2:0.25]
 ; ZNVER1-NEXT:    vextracti128 $1, %ymm1, (%rdi) # sched: [1:0.50]
 ; ZNVER1-NEXT:    vzeroupper # sched: [100:0.25]
; ZNVER1-NEXT:    retq # sched: [1:0.50]
   ret <4 x float> %1
 }
 declare <4 x float> @llvm.x86.avx2.gather.q.ps(<4 x float>, i8*, <2 x i64>, <4 x float>, i8) nounwind readonly
 
 define <4 x float> @test_gatherqps_ymm(<4 x float> %a0, i8* %a1, <4 x i64> %a2, <4 x float> %a3) {
 ; GENERIC-LABEL: test_gatherqps_ymm:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vgatherqps %xmm2, (%rdi,%ymm1,4), %xmm0 # sched: [5:0.50]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_gatherqps_ymm:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vgatherqps %xmm2, (%rdi,%ymm1,4), %xmm0 # sched: [28:3.67]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_gatherqps_ymm:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vgatherqps %xmm2, (%rdi,%ymm1,4), %xmm0 # sched: [24:5.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_gatherqps_ymm:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vgatherqps %xmm2, (%rdi,%ymm1,4), %xmm0 # sched: [25:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_gatherqps_ymm:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vgatherqps %xmm2, (%rdi,%ymm1,4), %xmm0 # sched: [25:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_gatherqps_ymm:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vgatherqps %xmm2, (%rdi,%ymm1,4), %xmm0 # sched: [100:0.25]
 ; ZNVER1-NEXT:    vzeroupper # sched: [100:0.25]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call <4 x float> @llvm.x86.avx2.gather.q.ps.256(<4 x float> %a0, i8* %a1, <4 x i64> %a2, <4 x float> %a3, i8 4)
; ZNVER1-NEXT:    retq # sched: [1:0.50]
   ret <4 x i32> %1
 }
 declare <4 x i32> @llvm.x86.avx2.gather.q.d(<4 x i32>, i8*, <2 x i64>, <4 x i32>, i8) nounwind readonly
 
 define <4 x i32> @test_pgatherqd_ymm(<4 x i32> %a0, i8* %a1, <4 x i64> %a2, <4 x i32> %a3) {
 ; GENERIC-LABEL: test_pgatherqd_ymm:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpgatherqd %xmm2, (%rdi,%ymm1,2), %xmm0 # sched: [5:0.50]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_pgatherqd_ymm:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vpgatherqd %xmm2, (%rdi,%ymm1,2), %xmm0 # sched: [28:5.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_pgatherqd_ymm:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vpgatherqd %xmm2, (%rdi,%ymm1,2), %xmm0 # sched: [5:0.50]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_pgatherqd_ymm:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vpgatherqd %xmm2, (%rdi,%ymm1,2), %xmm0 # sched: [25:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_pgatherqd_ymm:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpgatherqd %xmm2, (%rdi,%ymm1,2), %xmm0 # sched: [25:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_pgatherqd_ymm:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpgatherqd %xmm2, (%rdi,%ymm1,2), %xmm0 # sched: [100:0.25]
 ; ZNVER1-NEXT:    vzeroupper # sched: [100:0.25]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call <4 x i32> @llvm.x86.avx2.gather.q.d.256(<4 x i32> %a0, i8* %a1, <4 x i64> %a2, <4 x i32> %a3, i8 2)
; ZNVER1-NEXT:    retq # sched: [1:0.50]
   ret <16 x i16> %3
 }
 declare <16 x i16> @llvm.x86.avx2.pminu.w(<16 x i16>, <16 x i16>) nounwind readnone
 
 define i32 @test_pmovmskb(<32 x i8> %a0) {
 ; GENERIC-LABEL: test_pmovmskb:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpmovmskb %ymm0, %eax # sched: [2:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_pmovmskb:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vpmovmskb %ymm0, %eax # sched: [3:1.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_pmovmskb:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vpmovmskb %ymm0, %eax # sched: [3:1.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_pmovmskb:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vpmovmskb %ymm0, %eax # sched: [2:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_pmovmskb:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpmovmskb %ymm0, %eax # sched: [2:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_pmovmskb:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    vpmovmskb %ymm0, %eax # sched: [2:2.00]
 ; ZNVER1-NEXT:    vzeroupper # sched: [100:0.25]
 ; ZNVER1-NEXT:    retq # sched: [1:0.50]
   %1 = call i32 @llvm.x86.avx2.pmovmskb(<32 x i8> %a0)

llvm/trunk/test/CodeGen/X86/avx512-schedule.ll

; SKX-NEXT:    retq # sched: [7:1.00]
   %b = fptosi <4 x float> %a to <4 x i64>
   ret <4 x i64> %b
 }
 
 define <4 x float> @slto4f32(<4 x i64> %a) {
 ; GENERIC-LABEL: slto4f32:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvtqq2ps %ymm0, %xmm0 # sched: [3:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: slto4f32:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvtqq2ps %ymm0, %xmm0 # sched: [7:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %b = sitofp <4 x i64> %a to <4 x float>
   ret <4 x float> %b
 }
 
 define <4 x float> @ulto4f32(<4 x i64> %a) {
 ; GENERIC-LABEL: ulto4f32:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvtuqq2ps %ymm0, %xmm0 # sched: [3:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: ulto4f32:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvtuqq2ps %ymm0, %xmm0 # sched: [7:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %b = uitofp <4 x i64> %a to <4 x float>
   ret <4 x float> %b
 }
 
 define <8 x double> @ulto8f64(<8 x i64> %a) {
 ; GENERIC-LABEL: ulto8f64:
 ; GENERIC:       # %bb.0:
; SKX-NEXT:    retq # sched: [7:1.00]
   ret <16 x i32> %b
 }
 
 define <16 x i8> @f32to16uc(<16 x float> %f) {
 ; GENERIC-LABEL: f32to16uc:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvttps2dq %zmm0, %zmm0 # sched: [3:1.00]
 ; GENERIC-NEXT:    vpmovdb %zmm0, %xmm0 # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f32to16uc:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvttps2dq %zmm0, %zmm0 # sched: [4:0.50]
 ; SKX-NEXT:    vpmovdb %zmm0, %xmm0 # sched: [4:2.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %res = fptoui <16 x float> %f to <16 x i8>
   ret <16 x i8> %res
 }
 
 define <16 x i16> @f32to16us(<16 x float> %f) {
 ; GENERIC-LABEL: f32to16us:
 ; GENERIC:       # %bb.0:
; SKX-NEXT:    retq # sched: [7:1.00]
   ret <8 x i32> %b
 }
 
 define <8 x i16> @f64to8us(<8 x double> %f) {
 ; GENERIC-LABEL: f64to8us:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvttpd2dq %zmm0, %ymm0 # sched: [4:1.00]
 ; GENERIC-NEXT:    vpmovdw %ymm0, %xmm0 # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f64to8us:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvttpd2dq %zmm0, %ymm0 # sched: [7:1.00]
 ; SKX-NEXT:    vpmovdw %ymm0, %xmm0 # sched: [4:2.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %res = fptoui <8 x double> %f to <8 x i16>
   ret <8 x i16> %res
 }
 
 define <8 x i8> @f64to8uc(<8 x double> %f) {
 ; GENERIC-LABEL: f64to8uc:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvttpd2dq %zmm0, %ymm0 # sched: [4:1.00]
 ; GENERIC-NEXT:    vpmovdw %ymm0, %xmm0 # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f64to8uc:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvttpd2dq %zmm0, %ymm0 # sched: [7:1.00]
 ; SKX-NEXT:    vpmovdw %ymm0, %xmm0 # sched: [4:2.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %res = fptoui <8 x double> %f to <8 x i8>
   ret <8 x i8> %res
 }
 
 define <4 x i32> @f64to4ui(<4 x double> %a) nounwind {
 ; GENERIC-LABEL: f64to4ui:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvttpd2udq %ymm0, %xmm0 # sched: [4:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f64to4ui:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvttpd2udq %ymm0, %xmm0 # sched: [7:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %b = fptoui <4 x double> %a to <4 x i32>
   ret <4 x i32> %b
 }
 
 define <8 x double> @sito8f64(<8 x i32> %a) {
 ; GENERIC-LABEL: sito8f64:
 ; GENERIC:       # %bb.0:
; SKX-NEXT:    retq # sched: [7:1.00]
   %b = fptosi <8 x double> %a to <8 x i32>
   ret <8 x i32> %b
 }
 
 define <4 x i32> @f64to4si(<4 x double> %a) {
 ; GENERIC-LABEL: f64to4si:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvttpd2dq %ymm0, %xmm0 # sched: [4:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f64to4si:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvttpd2dq %ymm0, %xmm0 # sched: [7:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %b = fptosi <4 x double> %a to <4 x i32>
   ret <4 x i32> %b
 }
 
 define <16 x float> @f64to16f32(<16 x double> %b) nounwind {
 ; GENERIC-LABEL: f64to16f32:
 ; GENERIC:       # %bb.0:
; SKX-NEXT:    retq # sched: [7:1.00]
   %a = fptrunc <16 x double> %b to <16 x float>
   ret <16 x float> %a
 }
 
 define <4 x float> @f64to4f32(<4 x double> %b) {
 ; GENERIC-LABEL: f64to4f32:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvtpd2ps %ymm0, %xmm0 # sched: [4:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f64to4f32:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vcvtpd2ps %ymm0, %xmm0 # sched: [7:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %a = fptrunc <4 x double> %b to <4 x float>
   ret <4 x float> %a
 }
 
 define <4 x float> @f64to4f32_mask(<4 x double> %b, <4 x i1> %mask) {
 ; GENERIC-LABEL: f64to4f32_mask:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpslld $31, %xmm1, %xmm1 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpmovd2m %xmm1, %k1 # sched: [1:0.33]
 ; GENERIC-NEXT:    vcvtpd2ps %ymm0, %xmm0 {%k1} {z} # sched: [4:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: f64to4f32_mask:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpslld $31, %xmm1, %xmm1 # sched: [1:0.50]
 ; SKX-NEXT:    vpmovd2m %xmm1, %k1 # sched: [1:1.00]
 ; SKX-NEXT:    vcvtpd2ps %ymm0, %xmm0 {%k1} {z} # sched: [7:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %a = fptrunc <4 x double> %b to <4 x float>
   %c = select <4 x i1>%mask, <4 x float>%a, <4 x float> zeroinitializer
   ret <4 x float> %c
 }
 
 define <4 x float> @f64tof32_inreg(<2 x double> %a0, <4 x float> %a1) nounwind {
 ; GENERIC-LABEL: f64tof32_inreg:

 
 define i16 @trunc_16i32_to_16i1(<16 x i32> %a) {
 ; GENERIC-LABEL: trunc_16i32_to_16i1:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpslld $31, %zmm0, %zmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpmovd2m %zmm0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    kmovd %k0, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    # kill: def $ax killed $ax killed $eax
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: trunc_16i32_to_16i1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpslld $31, %zmm0, %zmm0 # sched: [1:1.00]
 ; SKX-NEXT:    vpmovd2m %zmm0, %k0 # sched: [1:1.00]
 ; SKX-NEXT:    kmovd %k0, %eax # sched: [3:1.00]
 ; SKX-NEXT:    # kill: def $ax killed $ax killed $eax
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %mask_b = trunc <16 x i32>%a to <16 x i1>
   %mask = bitcast <16 x i1> %mask_b to i16
   ret i16 %mask
 }
 
 define <4 x i32> @trunc_4i32_to_4i1(<4 x i32> %a, <4 x i32> %b) {
 ; GENERIC-LABEL: trunc_4i32_to_4i1:
; SKX-NEXT:    retq # sched: [7:1.00]
   ret i16 %res
 }
 
 define <8 x i16> @sext_8i1_8i16(<8 x i32> %a1, <8 x i32> %a2) nounwind {
 ; GENERIC-LABEL: sext_8i1_8i16:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpcmpgtd %ymm0, %ymm1, %k0 # sched: [1:0.50]
 ; GENERIC-NEXT:    vpmovm2w %k0, %xmm0 # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: sext_8i1_8i16:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpcmpgtd %ymm0, %ymm1, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    vpmovm2w %k0, %xmm0 # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %x = icmp slt <8 x i32> %a1, %a2
   %y = sext <8 x i1> %x to <8 x i16>
   ret <8 x i16> %y
 }
 
 define <16 x i32> @sext_16i1_16i32(<16 x i32> %a1, <16 x i32> %a2) nounwind {
 ; GENERIC-LABEL: sext_16i1_16i32:
; SKX-NEXT:    retq # sched: [7:1.00]
   ret <8 x i64> %y
 }
 
 define void @extload_v8i64(<8 x i8>* %a, <8 x i64>* %res) {
 ; GENERIC-LABEL: extload_v8i64:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpmovsxbq (%rdi), %zmm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    vmovdqa64 %zmm0, (%rsi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: extload_v8i64:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpmovsxbq (%rdi), %zmm0 # sched: [10:1.00]
 ; SKX-NEXT:    vmovdqa64 %zmm0, (%rsi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %sign_load = load <8 x i8>, <8 x i8>* %a
   %c = sext <8 x i8> %sign_load to <8 x i64>
   store <8 x i64> %c, <8 x i64>* %res
   ret void
 }
 
 define <64 x i16> @test21(<64 x i16> %x , <64 x i1> %mask) nounwind readnone {
; SKX-NEXT:    retq # sched: [7:1.00]
   %res = load <16 x i32>, <16 x i32>* %vaddr, align 64
   ret <16 x i32>%res
 }
 
 define void @mov_test18(i8 * %addr, <8 x i64> %data) {
 ; GENERIC-LABEL: mov_test18:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test18:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <8 x i64>*
   store <8 x i64>%data, <8 x i64>* %vaddr, align 64
   ret void
 }
 
 define void @mov_test19(i8 * %addr, <16 x i32> %data) {
 ; GENERIC-LABEL: mov_test19:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test19:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <16 x i32>*
   store <16 x i32>%data, <16 x i32>* %vaddr, align 1
   ret void
 }
 
 define void @mov_test20(i8 * %addr, <16 x i32> %data) {
 ; GENERIC-LABEL: mov_test20:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test20:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <16 x i32>*
   store <16 x i32>%data, <16 x i32>* %vaddr, align 64
   ret void
 }
 
 define  <8 x i64> @mov_test21(i8 * %addr) {
 ; GENERIC-LABEL: mov_test21:
; SKX-NEXT:    retq # sched: [7:1.00]
   %res = load <8 x i64>, <8 x i64>* %vaddr, align 64
   ret <8 x i64>%res
 }
 
 define void @mov_test22(i8 * %addr, <8 x i64> %data) {
 ; GENERIC-LABEL: mov_test22:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test22:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <8 x i64>*
   store <8 x i64>%data, <8 x i64>* %vaddr, align 1
   ret void
 }
 
 define <8 x i64> @mov_test23(i8 * %addr) {
 ; GENERIC-LABEL: mov_test23:
; SKX-NEXT:    retq # sched: [7:1.00]
   %res = load <8 x i64>, <8 x i64>* %vaddr, align 1
   ret <8 x i64>%res
 }
 
 define void @mov_test24(i8 * %addr, <8 x double> %data) {
 ; GENERIC-LABEL: mov_test24:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test24:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <8 x double>*
   store <8 x double>%data, <8 x double>* %vaddr, align 64
   ret void
 }
 
 define <8 x double> @mov_test25(i8 * %addr) {
 ; GENERIC-LABEL: mov_test25:
; SKX-NEXT:    retq # sched: [7:1.00]
   %res = load <8 x double>, <8 x double>* %vaddr, align 64
   ret <8 x double>%res
 }
 
 define void @mov_test26(i8 * %addr, <16 x float> %data) {
 ; GENERIC-LABEL: mov_test26:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test26:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <16 x float>*
   store <16 x float>%data, <16 x float>* %vaddr, align 64
   ret void
 }
 
 define <16 x float> @mov_test27(i8 * %addr) {
 ; GENERIC-LABEL: mov_test27:
; SKX-NEXT:    retq # sched: [7:1.00]
   %res = load <16 x float>, <16 x float>* %vaddr, align 64
   ret <16 x float>%res
 }
 
 define void @mov_test28(i8 * %addr, <8 x double> %data) {
 ; GENERIC-LABEL: mov_test28:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test28:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <8 x double>*
   store <8 x double>%data, <8 x double>* %vaddr, align 1
   ret void
 }
 
 define <8 x double> @mov_test29(i8 * %addr) {
 ; GENERIC-LABEL: mov_test29:
; SKX-NEXT:    retq # sched: [7:1.00]
   %res = load <8 x double>, <8 x double>* %vaddr, align 1
   ret <8 x double>%res
 }
 
 define void @mov_test30(i8 * %addr, <16 x float> %data) {
 ; GENERIC-LABEL: mov_test30:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: mov_test30:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovups %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %vaddr = bitcast i8* %addr to <16 x float>*
   store <16 x float>%data, <16 x float>* %vaddr, align 1
   ret void
 }
 
 define <16 x float> @mov_test31(i8 * %addr) {
 ; GENERIC-LABEL: mov_test31:

 
 define i32 @zext_test1(<16 x i32> %a, <16 x i32> %b) {
 ; GENERIC-LABEL: zext_test1:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0 # sched: [1:0.50]
 ; GENERIC-NEXT:    kshiftrw $5, %k0, %k0 # sched: [1:1.00]
 ; GENERIC-NEXT:    kmovd %k0, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    andl $1, %eax # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: zext_test1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kshiftrw $5, %k0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kmovd %k0, %eax # sched: [3:1.00]
 ; SKX-NEXT:    andl $1, %eax # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %cmp_res = icmp ugt <16 x i32> %a, %b
   %cmp_res.i1 = extractelement <16 x i1> %cmp_res, i32 5
   %res = zext i1 %cmp_res.i1 to i32
   ret i32 %res
 }
 
 define i16 @zext_test2(<16 x i32> %a, <16 x i32> %b) {
 ; GENERIC-LABEL: zext_test2:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0 # sched: [1:0.50]
 ; GENERIC-NEXT:    kshiftrw $5, %k0, %k0 # sched: [1:1.00]
 ; GENERIC-NEXT:    kmovd %k0, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    andl $1, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    # kill: def $ax killed $ax killed $eax
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: zext_test2:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kshiftrw $5, %k0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kmovd %k0, %eax # sched: [3:1.00]
 ; SKX-NEXT:    andl $1, %eax # sched: [1:0.25]
 ; SKX-NEXT:    # kill: def $ax killed $ax killed $eax
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %cmp_res = icmp ugt <16 x i32> %a, %b
   %cmp_res.i1 = extractelement <16 x i1> %cmp_res, i32 5
   %res = zext i1 %cmp_res.i1 to i16
   ret i16 %res
 }
 
 define i8 @zext_test3(<16 x i32> %a, <16 x i32> %b) {
 ; GENERIC-LABEL: zext_test3:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0 # sched: [1:0.50]
 ; GENERIC-NEXT:    kshiftrw $5, %k0, %k0 # sched: [1:1.00]
 ; GENERIC-NEXT:    kmovd %k0, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    andb $1, %al # sched: [1:0.33]
 ; GENERIC-NEXT:    # kill: def $al killed $al killed $eax
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: zext_test3:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kshiftrw $5, %k0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kmovd %k0, %eax # sched: [3:1.00]
 ; SKX-NEXT:    andb $1, %al # sched: [1:0.25]
 ; SKX-NEXT:    # kill: def $al killed $al killed $eax
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %cmp_res = icmp ugt <16 x i32> %a, %b
   %cmp_res.i1 = extractelement <16 x i1> %cmp_res, i32 5
   %res = zext i1 %cmp_res.i1 to i8
   ret i8 %res
 }
 
 define i8 @conv1(<8 x i1>* %R) {

 }
 
 define <4 x i32> @test4(<4 x i64> %x, <4 x i64> %y, <4 x i64> %x1, <4 x i64> %y1) {
 ; GENERIC-LABEL: test4:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpcmpleq %ymm1, %ymm0, %k1 # sched: [1:0.50]
 ; GENERIC-NEXT:    vpcmpgtq %ymm3, %ymm2, %k0 {%k1} # sched: [1:0.50]
 ; GENERIC-NEXT:    vpmovm2d %k0, %xmm0 # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: test4:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpcmpleq %ymm1, %ymm0, %k1 # sched: [3:1.00]
 ; SKX-NEXT:    vpcmpgtq %ymm3, %ymm2, %k0 {%k1} # sched: [3:1.00]
 ; SKX-NEXT:    vpmovm2d %k0, %xmm0 # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %x_gt_y = icmp sgt <4 x i64> %x, %y
   %x1_gt_y1 = icmp sgt <4 x i64> %x1, %y1
   %res = icmp sgt <4 x i1>%x_gt_y, %x1_gt_y1
   %resse = sext <4 x i1>%res to <4 x i32>
   ret <4 x i32> %resse
 }
 

 define <16 x i8> @vcmp_test8(<16 x i32>%a, <16 x i32>%b, i32 %a1, i32 %b1) {
 ; GENERIC-LABEL: vcmp_test8:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    cmpl %esi, %edi # sched: [1:0.33]
 ; GENERIC-NEXT:    jg .LBB386_1 # sched: [1:1.00]
 ; GENERIC-NEXT:  # %bb.2:
 ; GENERIC-NEXT:    kxorw %k0, %k0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    vpmovm2b %k0, %xmm0 # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ; GENERIC-NEXT:  .LBB386_1:
 ; GENERIC-NEXT:    vpxor %xmm1, %xmm1, %xmm1 # sched: [0:0.25]
 ; GENERIC-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 # sched: [1:0.50]
 ; GENERIC-NEXT:    vpmovm2b %k0, %xmm0 # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: vcmp_test8:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    cmpl %esi, %edi # sched: [1:0.25]
 ; SKX-NEXT:    jg .LBB386_1 # sched: [1:0.50]
 ; SKX-NEXT:  # %bb.2:
 ; SKX-NEXT:    kxorw %k0, %k0, %k0 # sched: [1:1.00]
 ; SKX-NEXT:    vpmovm2b %k0, %xmm0 # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ; SKX-NEXT:  .LBB386_1:
 ; SKX-NEXT:    vpxor %xmm1, %xmm1, %xmm1 # sched: [1:0.33]
 ; SKX-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    vpmovm2b %k0, %xmm0 # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %cond = icmp sgt i32 %a1, %b1
   %cmp1 = icmp sgt <16 x i32> %a, zeroinitializer
   %cmp2 = icmp ult <16 x i32> %b, zeroinitializer
   %mix = select i1 %cond, <16 x i1> %cmp1, <16 x i1> %cmp2
   %res = sext <16 x i1> %mix to <16 x i8>
   ret <16 x i8> %res
 }

 ; GENERIC-NEXT:    vmovupd (%rdi), %zmm1 # sched: [7:0.50]
 ; GENERIC-NEXT:    vcmpltpd %zmm0, %zmm1, %k1 # sched: [3:1.00]
 ; GENERIC-NEXT:    vmovupd 8(%rdi), %zmm1 {%k1} {z} # sched: [7:0.50]
 ; GENERIC-NEXT:    vcmpltpd %zmm1, %zmm0, %k0 {%k1} # sched: [3:1.00]
 ; GENERIC-NEXT:    kortestb %k0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    je .LBB410_2 # sched: [1:1.00]
 ; GENERIC-NEXT:  # %bb.1: # %L1
 ; GENERIC-NEXT:    vmovapd %zmm0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ; GENERIC-NEXT:  .LBB410_2: # %L2
 ; GENERIC-NEXT:    vmovapd %zmm0, 8(%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: ktest_1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovupd (%rdi), %zmm1 # sched: [8:0.50]
 ; SKX-NEXT:    vcmpltpd %zmm0, %zmm1, %k1 # sched: [3:1.00]
 ; SKX-NEXT:    vmovupd 8(%rdi), %zmm1 {%k1} {z} # sched: [8:0.50]
 ; SKX-NEXT:    vcmpltpd %zmm1, %zmm0, %k0 {%k1} # sched: [3:1.00]
 ; SKX-NEXT:    kortestb %k0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    je .LBB410_2 # sched: [1:0.50]
 ; SKX-NEXT:  # %bb.1: # %L1
 ; SKX-NEXT:    vmovapd %zmm0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ; SKX-NEXT:  .LBB410_2: # %L2
 ; SKX-NEXT:    vmovapd %zmm0, 8(%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %addr1 = getelementptr double, double * %base, i64 0
   %addr2 = getelementptr double, double * %base, i64 1
 
   %vaddr1 = bitcast double* %addr1 to <8 x double>*
   %vaddr2 = bitcast double* %addr2 to <8 x double>*
 
   %val1 = load <8 x double>, <8 x double> *%vaddr1, align 1

 ; GENERIC-NEXT:    vcmpltps %zmm3, %zmm0, %k1 # sched: [3:1.00]
 ; GENERIC-NEXT:    vcmpltps %zmm2, %zmm1, %k2 # sched: [3:1.00]
 ; GENERIC-NEXT:    kunpckwd %k1, %k2, %k1 # sched: [1:1.00]
 ; GENERIC-NEXT:    kortestd %k1, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    je .LBB411_2 # sched: [1:1.00]
 ; GENERIC-NEXT:  # %bb.1: # %L1
 ; GENERIC-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
 ; GENERIC-NEXT:    vmovaps %zmm1, 64(%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ; GENERIC-NEXT:  .LBB411_2: # %L2
 ; GENERIC-NEXT:    vmovaps %zmm0, 4(%rdi) # sched: [1:1.00]
 ; GENERIC-NEXT:    vmovaps %zmm1, 68(%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: ktest_2:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vmovups (%rdi), %zmm2 # sched: [8:0.50]
 ; SKX-NEXT:    vmovups 64(%rdi), %zmm3 # sched: [8:0.50]
 ; SKX-NEXT:    vcmpltps %zmm0, %zmm2, %k1 # sched: [3:1.00]
 ; SKX-NEXT:    vcmpltps %zmm1, %zmm3, %k2 # sched: [3:1.00]
 ; SKX-NEXT:    kunpckwd %k1, %k2, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    vmovups 68(%rdi), %zmm2 {%k2} {z} # sched: [8:0.50]
 ; SKX-NEXT:    vmovups 4(%rdi), %zmm3 {%k1} {z} # sched: [8:0.50]
 ; SKX-NEXT:    vcmpltps %zmm3, %zmm0, %k1 # sched: [3:1.00]
 ; SKX-NEXT:    vcmpltps %zmm2, %zmm1, %k2 # sched: [3:1.00]
 ; SKX-NEXT:    kunpckwd %k1, %k2, %k1 # sched: [3:1.00]
 ; SKX-NEXT:    kortestd %k1, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    je .LBB411_2 # sched: [1:0.50]
 ; SKX-NEXT:  # %bb.1: # %L1
 ; SKX-NEXT:    vmovaps %zmm0, (%rdi) # sched: [1:1.00]
 ; SKX-NEXT:    vmovaps %zmm1, 64(%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ; SKX-NEXT:  .LBB411_2: # %L2
 ; SKX-NEXT:    vmovaps %zmm0, 4(%rdi) # sched: [1:1.00]
 ; SKX-NEXT:    vmovaps %zmm1, 68(%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %addr1 = getelementptr float, float * %base, i64 0
   %addr2 = getelementptr float, float * %base, i64 1
 
   %vaddr1 = bitcast float* %addr1 to <32 x float>*
   %vaddr2 = bitcast float* %addr2 to <32 x float>*
 
   %val1 = load <32 x float>, <32 x float> *%vaddr1, align 1

 }
 
 define void @store_32i1(<32 x i1>* %a, <32 x i1> %v) {
 ; GENERIC-LABEL: store_32i1:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpsllw $7, %ymm0, %ymm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpmovb2m %ymm0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    kmovd %k0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: store_32i1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %ymm0, %ymm0 # sched: [1:0.50]
 ; SKX-NEXT:    vpmovb2m %ymm0, %k0 # sched: [1:1.00]
 ; SKX-NEXT:    kmovd %k0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   store <32 x i1> %v, <32 x i1>* %a
   ret void
 }
 
 define void @store_32i1_1(<32 x i1>* %a, <32 x i16> %v) {
 ; GENERIC-LABEL: store_32i1_1:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpsllw $15, %zmm0, %zmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpmovw2m %zmm0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    kmovd %k0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: store_32i1_1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $15, %zmm0, %zmm0 # sched: [1:1.00]
 ; SKX-NEXT:    vpmovw2m %zmm0, %k0 # sched: [1:1.00]
 ; SKX-NEXT:    kmovd %k0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %v1 = trunc <32 x i16> %v to <32 x i1>
   store <32 x i1> %v1, <32 x i1>* %a
   ret void
 }
 
 
 define void @store_64i1(<64 x i1>* %a, <64 x i1> %v) {
 ;
 ; GENERIC-LABEL: store_64i1:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpsllw $7, %zmm0, %zmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpmovb2m %zmm0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    kmovq %k0, (%rdi) # sched: [1:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: store_64i1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %zmm0, %zmm0 # sched: [1:1.00]
 ; SKX-NEXT:    vpmovb2m %zmm0, %k0 # sched: [1:1.00]
 ; SKX-NEXT:    kmovq %k0, (%rdi) # sched: [1:1.00]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   store <64 x i1> %v, <64 x i1>* %a
   ret void
 }
 
 define i32 @test_bitcast_v8i1_zext(<16 x i32> %a) {
 ; GENERIC-LABEL: test_bitcast_v8i1_zext:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vptestnmd %zmm0, %zmm0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    kmovb %k0, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    addl %eax, %eax # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: test_bitcast_v8i1_zext:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vptestnmd %zmm0, %zmm0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kmovb %k0, %eax # sched: [3:1.00]
 ; SKX-NEXT:    addl %eax, %eax # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
    %v1 = icmp eq <16 x i32> %a, zeroinitializer
    %mask = shufflevector <16 x i1> %v1, <16 x i1> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
    %mask1 = bitcast <8 x i1> %mask to i8
    %val = zext i8 %mask1 to i32
    %val1 = add i32 %val, %val
    ret i32 %val1
 }
 
 define i32 @test_bitcast_v16i1_zext(<16 x i32> %a) {
 ; GENERIC-LABEL: test_bitcast_v16i1_zext:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vptestnmd %zmm0, %zmm0, %k0 # sched: [1:0.33]
 ; GENERIC-NEXT:    kmovw %k0, %eax # sched: [1:0.33]
 ; GENERIC-NEXT:    addl %eax, %eax # sched: [1:0.33]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: test_bitcast_v16i1_zext:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vptestnmd %zmm0, %zmm0, %k0 # sched: [3:1.00]
 ; SKX-NEXT:    kmovw %k0, %eax # sched: [3:1.00]
 ; SKX-NEXT:    addl %eax, %eax # sched: [1:0.25]
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
    %v1 = icmp eq <16 x i32> %a, zeroinitializer
    %mask1 = bitcast <16 x i1> %v1 to i16
    %val = zext i16 %mask1 to i32
    %val1 = add i32 %val, %val
    ret i32 %val1
 }
 

llvm/trunk/test/CodeGen/X86/avx512vpopcntdq-schedule.ll

 ; GENERIC-NEXT:    vpopcntd %zmm1, %zmm0 {%k1} {z} # sched: [1:0.50]
 ; GENERIC-NEXT:    vpopcntd (%rdi), %zmm0 # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntd (%rdi), %zmm0 {%k1} # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntd (%rdi), %zmm0 {%k1} {z} # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntd (%rdi){1to16}, %zmm0 # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntd (%rdi){1to16}, %zmm0 {%k1} # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntd (%rdi){1to16}, %zmm0 {%k1} {z} # sched: [8:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; ICELAKE-LABEL: test_vpopcntd:
 ; ICELAKE:       # %bb.0:
 ; ICELAKE-NEXT:    kmovd %esi, %k1 # sched: [1:1.00]
 ; ICELAKE-NEXT:    #APP
 ; ICELAKE-NEXT:    vpopcntd %zmm1, %zmm0 # sched: [1:1.00]
 ; ICELAKE-NEXT:    vpopcntd %zmm1, %zmm0 {%k1} # sched: [1:1.00]
 ; ICELAKE-NEXT:    vpopcntd %zmm1, %zmm0 {%k1} {z} # sched: [1:1.00]
 ; ICELAKE-NEXT:    vpopcntd (%rdi), %zmm0 # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntd (%rdi), %zmm0 {%k1} # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntd (%rdi), %zmm0 {%k1} {z} # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntd (%rdi){1to16}, %zmm0 # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntd (%rdi){1to16}, %zmm0 {%k1} # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntd (%rdi){1to16}, %zmm0 {%k1} {z} # sched: [8:1.00]
 ; ICELAKE-NEXT:    #NO_APP
-; ICELAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; ICELAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; ICELAKE-NEXT:    retq # sched: [7:1.00]
   tail call void asm "vpopcntd $1, $0 \0A\09 vpopcntd $1, $0 {$3} \0A\09 vpopcntd $1, $0 {$3} {z} \0A\09 vpopcntd $2, $0 \0A\09 vpopcntd $2, $0 {$3} \0A\09 vpopcntd $2, $0 {$3} {z} \0A\09 vpopcntd $2{1to16}, $0 \0A\09 vpopcntd $2{1to16}, $0 {$3} \0A\09 vpopcntd $2{1to16}, $0 {$3} {z}", "v,v,*m,^Yk"(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> *%a2, i16 %a3) nounwind
   ret void
 }
 
 define void @test_vpopcntq(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> *%a2, i8 %a3) {
 ; GENERIC-LABEL: test_vpopcntq:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    kmovw %esi, %k1 # sched: [1:0.33]
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vpopcntq %zmm1, %zmm0 # sched: [1:0.50]
 ; GENERIC-NEXT:    vpopcntq %zmm1, %zmm0 {%k1} # sched: [1:0.50]
 ; GENERIC-NEXT:    vpopcntq %zmm1, %zmm0 {%k1} {z} # sched: [1:0.50]
 ; GENERIC-NEXT:    vpopcntq (%rdi), %zmm0 # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntq (%rdi), %zmm0 {%k1} # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntq (%rdi), %zmm0 {%k1} {z} # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntq (%rdi){1to8}, %zmm0 # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntq (%rdi){1to8}, %zmm0 {%k1} # sched: [8:0.50]
 ; GENERIC-NEXT:    vpopcntq (%rdi){1to8}, %zmm0 {%k1} {z} # sched: [8:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; ICELAKE-LABEL: test_vpopcntq:
 ; ICELAKE:       # %bb.0:
 ; ICELAKE-NEXT:    kmovd %esi, %k1 # sched: [1:1.00]
 ; ICELAKE-NEXT:    #APP
 ; ICELAKE-NEXT:    vpopcntq %zmm1, %zmm0 # sched: [1:1.00]
 ; ICELAKE-NEXT:    vpopcntq %zmm1, %zmm0 {%k1} # sched: [1:1.00]
 ; ICELAKE-NEXT:    vpopcntq %zmm1, %zmm0 {%k1} {z} # sched: [1:1.00]
 ; ICELAKE-NEXT:    vpopcntq (%rdi), %zmm0 # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntq (%rdi), %zmm0 {%k1} # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntq (%rdi), %zmm0 {%k1} {z} # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntq (%rdi){1to8}, %zmm0 # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntq (%rdi){1to8}, %zmm0 {%k1} # sched: [8:1.00]
 ; ICELAKE-NEXT:    vpopcntq (%rdi){1to8}, %zmm0 {%k1} {z} # sched: [8:1.00]
 ; ICELAKE-NEXT:    #NO_APP
-; ICELAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; ICELAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; ICELAKE-NEXT:    retq # sched: [7:1.00]
   tail call void asm "vpopcntq $1, $0 \0A\09 vpopcntq $1, $0 {$3} \0A\09 vpopcntq $1, $0 {$3} {z} \0A\09 vpopcntq $2, $0 \0A\09 vpopcntq $2, $0 {$3} \0A\09 vpopcntq $2, $0 {$3} {z} \0A\09 vpopcntq $2{1to8}, $0 \0A\09 vpopcntq $2{1to8}, $0 {$3} \0A\09 vpopcntq $2{1to8}, $0 {$3} {z}", "v,v,*m,^Yk"(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> *%a2, i8 %a3) nounwind
   ret void
 }

llvm/trunk/test/CodeGen/X86/f16c-schedule.ll

 }
 declare <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float>, i32)
 
 define <8 x i16> @test_vcvtps2ph_256(<8 x float> %a0, <8 x float> %a1, <8 x i16> *%a2) {
 ; GENERIC-LABEL: test_vcvtps2ph_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [3:1.00]
 ; GENERIC-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [4:1.00]
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; IVY-LABEL: test_vcvtps2ph_256:
 ; IVY:       # %bb.0:
 ; IVY-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [3:1.00]
 ; IVY-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [4:1.00]
-; IVY-NEXT:    vzeroupper # sched: [100:0.33]
+; IVY-NEXT:    vzeroupper # sched: [1:1.00]
 ; IVY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_vcvtps2ph_256:
 ; HASWELL:       # %bb.0:
 ; HASWELL-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [6:1.00]
 ; HASWELL-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [7:1.00]
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vcvtps2ph_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [6:1.00]
 ; BROADWELL-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [7:1.00]
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vcvtps2ph_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    vcvtps2ph $0, %ymm0, %xmm0 # sched: [7:1.00]
 ; SKYLAKE-NEXT:    vcvtps2ph $0, %ymm1, (%rdi) # sched: [8:1.00]
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; BDVER2-LABEL: test_vcvtps2ph_256:
 ; BDVER2:       # %bb.0:
 ; 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]

llvm/trunk/test/CodeGen/X86/fma-schedule.ll

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmaddpd_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmaddpd_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmaddpd_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmaddpd_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [12:0.50]
 ; KNL-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmaddpd_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKX-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmaddpd_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmaddps_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmaddps_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmaddps_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmaddps_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [12:0.50]
 ; KNL-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmaddps_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKX-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmaddps_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) + ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmaddsubpd_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmaddsubpd_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmaddsubpd_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmaddsubpd_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [12:0.50]
 ; KNL-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmaddsubpd_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [11:0.50]
 ; SKX-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmaddsubpd_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmaddsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmaddsubps_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmaddsubps_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmaddsubps_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmaddsubps_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [12:0.50]
 ; KNL-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmaddsubps_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * mem) +/- ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- mem sched: [11:0.50]
 ; SKX-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * mem) +/- ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmaddsubps_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) +/- ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmaddsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) +/- ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmsubaddpd_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmsubaddpd_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmsubaddpd_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmsubaddpd_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [12:0.50]
 ; KNL-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmsubaddpd_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [11:0.50]
 ; SKX-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmsubaddpd_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsubadd231pd {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmsubaddps_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmsubaddps_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmsubaddps_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmsubaddps_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [12:0.50]
 ; KNL-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmsubaddps_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * mem) -/+ ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ mem sched: [11:0.50]
 ; SKX-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * mem) -/+ ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmsubaddps_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm2) -/+ ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsubadd231ps {{.*#+}} ymm0 = (ymm1 * ymm2) -/+ ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmsubpd_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmsubpd_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmsubpd_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmsubpd_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [12:0.50]
 ; KNL-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmsubpd_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKX-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmsubpd_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsub231pd {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfmsubps_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfmsubps_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfmsubps_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfmsubps_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [12:0.50]
 ; KNL-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfmsubps_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKX-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfmsubps_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfmsub231ps {{.*#+}} ymm0 = (ymm1 * ymm2) - ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfnmaddpd_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfnmaddpd_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfnmaddpd_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfnmaddpd_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [12:0.50]
 ; KNL-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfnmaddpd_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKX-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfnmaddpd_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmadd231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfnmaddps_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfnmaddps_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfnmaddps_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfnmaddps_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [12:0.50]
 ; KNL-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfnmaddps_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * mem) + ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + mem sched: [11:0.50]
 ; SKX-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * mem) + ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfnmaddps_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) + ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) + ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfnmsubpd_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfnmsubpd_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfnmsubpd_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfnmsubpd_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [12:0.50]
 ; KNL-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfnmsubpd_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKX-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfnmsubpd_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmsub231pd {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]

 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER2-LABEL: test_vfnmsubps_256:
 ; BDVER2:       # %bb.0:
 ; BDVER2-NEXT:    #APP
 ; 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]

 ; HASWELL-NEXT:    #APP
 ; HASWELL-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; HASWELL-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [12:0.50]
 ; HASWELL-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [12:0.50]
 ; HASWELL-NEXT:    #NO_APP
-; HASWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; HASWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; HASWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; BROADWELL-LABEL: test_vfnmsubps_256:
 ; BROADWELL:       # %bb.0:
 ; BROADWELL-NEXT:    #APP
 ; BROADWELL-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; BROADWELL-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [11:0.50]
 ; BROADWELL-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [11:0.50]
 ; BROADWELL-NEXT:    #NO_APP
-; BROADWELL-NEXT:    vzeroupper # sched: [4:1.00]
+; BROADWELL-NEXT:    vzeroupper # sched: [0:1.00]
 ; BROADWELL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKYLAKE-LABEL: test_vfnmsubps_256:
 ; SKYLAKE:       # %bb.0:
 ; SKYLAKE-NEXT:    #APP
 ; SKYLAKE-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKYLAKE-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKYLAKE-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKYLAKE-NEXT:    #NO_APP
-; SKYLAKE-NEXT:    vzeroupper # sched: [4:1.00]
+; SKYLAKE-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKYLAKE-NEXT:    retq # sched: [7:1.00]
 ;
 ; KNL-LABEL: test_vfnmsubps_256:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    #APP
 ; KNL-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; KNL-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; KNL-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]
 ; KNL-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [12:0.50]
 ; KNL-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [12:0.50]
 ; KNL-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [12:0.50]
 ; KNL-NEXT:    #NO_APP
 ; KNL-NEXT:    retq # sched: [7:1.00]
 ;
 ; SKX-LABEL: test_vfnmsubps_256:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    #APP
 ; SKX-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [4:0.50]
 ; SKX-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [4:0.50]
 ; SKX-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [4:0.50]
 ; SKX-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * mem) - ymm1 sched: [11:0.50]
 ; SKX-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - mem sched: [11:0.50]
 ; SKX-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * mem) - ymm0 sched: [11:0.50]
 ; SKX-NEXT:    #NO_APP
-; SKX-NEXT:    vzeroupper # sched: [4:1.00]
+; SKX-NEXT:    vzeroupper # sched: [0:0.67]
 ; SKX-NEXT:    retq # sched: [7:1.00]
 ;
 ; ZNVER1-LABEL: test_vfnmsubps_256:
 ; ZNVER1:       # %bb.0:
 ; ZNVER1-NEXT:    #APP
 ; ZNVER1-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm2) - ymm1 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2 sched: [5:0.50]
 ; ZNVER1-NEXT:    vfnmsub231ps {{.*#+}} ymm0 = -(ymm1 * ymm2) - ymm0 sched: [5:0.50]

llvm/trunk/test/CodeGen/X86/fma4-schedule.ll

 define void @test_vfmaddpd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmaddpd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmaddpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmaddpd_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmaddps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmaddps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmaddps_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmaddsubpd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmaddsubpd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmaddsubpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmaddsubpd_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmaddsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmaddsubps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmaddsubps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmaddsubps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmaddsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmaddsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmaddsubps_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmaddsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmsubaddpd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmsubaddpd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsubaddpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmsubaddpd_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmsubaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmsubaddps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmsubaddps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsubaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmsubaddps_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmsubaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmsubpd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmsubpd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsubpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmsubpd_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfmsubps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) optsize {
 ; GENERIC-LABEL: test_vfmsubps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfmsubps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfmsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfmsubps_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfnmaddpd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) optsize {
 ; GENERIC-LABEL: test_vfnmaddpd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmaddpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmaddpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfnmaddpd_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfnmaddpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfnmaddps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) optsize {
 ; GENERIC-LABEL: test_vfnmaddps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmaddps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfnmaddps_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfnmaddps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfnmsubpd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) optsize {
 ; GENERIC-LABEL: test_vfnmsubpd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmsubpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsubpd (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfnmsubpd_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfnmsubpd %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

 define void @test_vfnmsubps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) optsize {
 ; GENERIC-LABEL: test_vfnmsubps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfnmsubps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
 ; GENERIC-NEXT:    vfnmsubps (%rdi), %ymm1, %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    vfnmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:0.50]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfnmsubps_256:
 ; 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: [10:1.00]
 ; BDVER12-NEXT:    vfnmsubps %ymm1, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]

llvm/trunk/test/CodeGen/X86/xop-schedule.ll

 ; 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=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
 
 define void @test_vfrczpd(<2 x double> %a0, <4 x double> %a1, <2 x double> *%a2, <4 x double> *%a3) {
 ; GENERIC-LABEL: test_vfrczpd:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfrczpd %xmm0, %xmm0 # sched: [3:1.00]
 ; GENERIC-NEXT:    vfrczpd %ymm1, %ymm1 # sched: [3:1.00]
 ; GENERIC-NEXT:    vfrczpd (%rdi), %xmm0 # sched: [9:1.00]
 ; GENERIC-NEXT:    vfrczpd (%rsi), %ymm1 # sched: [10:1.00]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfrczpd:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; 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]

 ; GENERIC-LABEL: test_vfrczps:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vfrczps %xmm0, %xmm0 # sched: [3:1.00]
 ; GENERIC-NEXT:    vfrczps %ymm1, %ymm1 # sched: [3:1.00]
 ; GENERIC-NEXT:    vfrczps (%rdi), %xmm0 # sched: [9:1.00]
 ; GENERIC-NEXT:    vfrczps (%rsi), %ymm1 # sched: [10:1.00]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vfrczps:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; 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]

 define void @test_vpcmov_256(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> %a2, <4 x i64> *%a3) {
 ; GENERIC-LABEL: test_vpcmov_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpcmov (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    vpcmov %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vpcmov_256:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; 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]

 define void @test_vpermil2pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) {
 ; GENERIC-LABEL: test_vpermil2pd_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vpermil2pd $3, %ymm2, %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpermil2pd $3, %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    vpermil2pd $3, (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vpermil2pd_256:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; 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]

 define void @test_vpermil2ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) {
 ; GENERIC-LABEL: test_vpermil2ps_256:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    #APP
 ; GENERIC-NEXT:    vpermil2ps $3, %ymm2, %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    vpermil2ps $3, %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    vpermil2ps $3, (%rdi), %ymm1, %ymm0, %ymm0 # sched: [8:1.00]
 ; GENERIC-NEXT:    #NO_APP
-; GENERIC-NEXT:    vzeroupper # sched: [100:0.33]
+; GENERIC-NEXT:    vzeroupper # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; BDVER12-LABEL: test_vpermil2ps_256:
 ; BDVER12:       # %bb.0:
 ; BDVER12-NEXT:    #APP
 ; 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]

llvm/trunk/test/tools/llvm-mca/X86/Broadwell/resources-avx1.s

 # CHECK-NEXT:  2      6     1.00    *                   vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      1     1.00                        vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      7     1.00    *                   vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     1.00                        vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  2      6     1.00    *                   vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      1     1.00                        vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      7     1.00    *                   vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  20     8     1.00    *      *      U     vzeroall
-# CHECK-NEXT:  4      4     1.00    *      *      U     vzeroupper
+# CHECK-NEXT:  4      0     1.00    *      *      U     vzeroupper
 
 # CHECK:      Resources:
 # CHECK-NEXT: [0]   - BWDivider
 # CHECK-NEXT: [1]   - BWFPDivider
 # CHECK-NEXT: [2]   - BWPort0
 # CHECK-NEXT: [3]   - BWPort1
 # CHECK-NEXT: [4]   - BWPort2
 # CHECK-NEXT: [5]   - BWPort3
 # CHECK-NEXT: [6]   - BWPort4
 # CHECK-NEXT: [7]   - BWPort5
 # CHECK-NEXT: [8]   - BWPort6
 # CHECK-NEXT: [9]   - BWPort7
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
-# CHECK-NEXT:  -     257.00 216.33 236.33 176.17 176.17 38.00  427.33 3.00   12.67
+# CHECK-NEXT:  -     257.00 215.25 235.25 176.17 176.17 38.00  426.25 2.25   12.67
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    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:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00   1.00    -     vzeroall
-# CHECK-NEXT:  -      -     1.08   1.08    -      -      -     1.08   0.75    -     vzeroupper
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     vzeroupper

llvm/trunk/test/tools/llvm-mca/X86/Generic/resources-avx1.s

 # 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:  4      1     1.00    *      *      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:      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:  -     572.00 246.50 317.00 39.00  365.50 179.50 179.50
 
 # 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   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:  -      -      -      -      -      -      -      -     vzeroupper

llvm/trunk/test/tools/llvm-mca/X86/Haswell/resources-avx1.s

 # 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     8     1.00    *      *      U     vzeroall
-# CHECK-NEXT:  4      4     1.00    *      *      U     vzeroupper
+# CHECK-NEXT:  4      0     1.00    *      *      U     vzeroupper
 
 # CHECK:      Resources:
 # CHECK-NEXT: [0]   - HWDivider
 # CHECK-NEXT: [1]   - HWFPDivider
 # CHECK-NEXT: [2]   - HWPort0
 # CHECK-NEXT: [3]   - HWPort1
 # CHECK-NEXT: [4]   - HWPort2
 # CHECK-NEXT: [5]   - HWPort3
 # CHECK-NEXT: [6]   - HWPort4
 # CHECK-NEXT: [7]   - HWPort5
 # CHECK-NEXT: [8]   - HWPort6
 # CHECK-NEXT: [9]   - HWPort7
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
-# CHECK-NEXT:  -     336.00 215.67 237.67 176.17 176.17 38.00  430.67 3.00   12.67
+# CHECK-NEXT:  -     336.00 214.58 236.58 176.17 176.17 38.00  429.58 2.25   12.67
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    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:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00   1.00    -     vzeroall
-# CHECK-NEXT:  -      -     1.08   1.08    -      -      -     1.08   0.75    -     vzeroupper
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     vzeroupper

llvm/trunk/test/tools/llvm-mca/X86/SandyBridge/resources-avx1.s

 # 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:  4      1     1.00    *      *      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:      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:  -     572.00 246.50 317.00 39.00  365.50 179.50 179.50
 
 # 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   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:  -      -      -      -      -      -      -      -     vzeroupper

llvm/trunk/test/tools/llvm-mca/X86/SkylakeClient/resources-avx1.s

 # CHECK-NEXT:  2      7     0.50    *                   vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      1     0.33                        vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      8     0.50    *                   vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     0.33                        vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  2      7     0.50    *                   vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      1     0.33                        vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      8     0.50    *                   vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  16     16    4.00    *      *      U     vzeroall
-# CHECK-NEXT:  4      4     1.00    *      *      U     vzeroupper
+# CHECK-NEXT:  4      0     0.67    *      *      U     vzeroupper
 
 # CHECK:      Resources:
 # CHECK-NEXT: [0]   - SKLDivider
 # CHECK-NEXT: [1]   - SKLFPDivider
 # CHECK-NEXT: [2]   - SKLPort0
 # CHECK-NEXT: [3]   - SKLPort1
 # CHECK-NEXT: [4]   - SKLPort2
 # CHECK-NEXT: [5]   - SKLPort3
 # CHECK-NEXT: [6]   - SKLPort4
 # CHECK-NEXT: [7]   - SKLPort5
 # CHECK-NEXT: [8]   - SKLPort6
 # CHECK-NEXT: [9]   - SKLPort7
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
-# CHECK-NEXT:  -     123.00 333.17 204.17 173.17 173.17 34.00  324.67 6.00   12.67
+# CHECK-NEXT:  -     123.00 332.08 203.08 173.17 173.17 34.00  323.58 5.25   12.67
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vaddpd  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.50   0.50   0.50   0.50    -      -      -      -     vaddpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vaddpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.50   0.50   0.50   0.50    -      -      -      -     vaddpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vaddps  %xmm0, %xmm1, %xmm2

 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -     4.00   4.00    -      -      -     4.00   4.00    -     vzeroall
-# CHECK-NEXT:  -      -     1.08   1.08    -      -      -     1.08   0.75    -     vzeroupper
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     vzeroupper

llvm/trunk/test/tools/llvm-mca/X86/SkylakeServer/resources-avx1.s

 # CHECK-NEXT:  2      7     0.50    *                   vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      1     0.33                        vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      8     0.50    *                   vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     0.33                        vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  2      7     0.50    *                   vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  1      1     0.33                        vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      8     0.50    *                   vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  34     12    5.00    *      *      U     vzeroall
-# CHECK-NEXT:  4      4     1.00    *      *      U     vzeroupper
+# CHECK-NEXT:  4      0     0.67    *      *      U     vzeroupper
 
 # CHECK:      Resources:
 # CHECK-NEXT: [0]   - SKXDivider
 # CHECK-NEXT: [1]   - SKXFPDivider
 # CHECK-NEXT: [2]   - SKXPort0
 # CHECK-NEXT: [3]   - SKXPort1
 # CHECK-NEXT: [4]   - SKXPort2
 # CHECK-NEXT: [5]   - SKXPort3
 # CHECK-NEXT: [6]   - SKXPort4
 # CHECK-NEXT: [7]   - SKXPort5
 # CHECK-NEXT: [8]   - SKXPort6
 # CHECK-NEXT: [9]   - SKXPort7
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
-# CHECK-NEXT:  -     123.00 318.50 197.50 173.17 173.17 34.00  339.00 7.00   12.67
+# CHECK-NEXT:  -     123.00 317.42 196.42 173.17 173.17 34.00  337.92 6.25   12.67
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vaddpd  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.50   0.50   0.50   0.50    -      -      -      -     vaddpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vaddpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.50   0.50   0.50   0.50    -      -      -      -     vaddpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vaddps  %xmm0, %xmm1, %xmm2

 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorpd  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorpd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorpd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorps  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorps  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorps  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorps  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -     2.00   1.00    -      -      -     2.00   5.00    -     vzeroall
-# CHECK-NEXT:  -      -     1.08   1.08    -      -      -     1.08   0.75    -     vzeroupper
+# CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     vzeroupper