[AMDGPU] Divergence driven instruction selection. Part 1.

[alex-t]

Summary

This change is the first part of the AMDGPU target description change. The aim of it is the effective splitting the vector and scalar flows at the selection stage. Selection uses predicate functions based on the framework implemented earlier: https://reviews.llvm.org/D35267

Tests: CodeGen/AMDGPU on Win32

make check-llvm on lnx

[rampitec]

lib/Target/AMDGPU/SIInstrInfo.td

1762

You can create enum for the mode in SIInstrInfo.td. Look at SIEncodingFamily as an example.
Then I might be missing something, but I do not see anywhere a value 1 used for mode.

lib/Target/AMDGPU/VOP2Instructions.td

364

Is there a reason to move it from its original place? This seems to grow the patch unnecessary and makes review harder.

424

It would be nice to swap these instructions' blocks back to minimize the diff. Order change is not needed for this patch.

502

Again, yhou probably do not need to move adde and sube patterns.

515

Why is this reg_sequence and extract_subregs are needed? Why not just use i64 operands?

lib/Target/AMDGPU/VOPInstructions.td

576

Where do you set NeedPatGen = 1?

test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll

450

It is the same, isn't it? Why not to keep it GCN-DAG?

479

Same here.

[alex-t]

Source cleanup.

lib/Target/AMDGPU/VOP2Instructions.td

364

In fact the reason was that Tablegen gen-asm-matcher fails with the empty prefix strings

in a def like this the empty string "" overrides the default VOP2_Pseido "prefix" argument that is prefix = "_e32"
VOP2_Pseudo <"v_madmk_f32", VOP_MADMK_F32, [], "">

As a result asm-matcher-gen complained about mnemonic alias with the same string.

For some miracle reason this expressed only with Predicates[] explicitly set.
I don't know how it is related. Some another one Tablegen black magic.

I moved this 4 defs back but removed empty prefix "". Since nothing failed after this I conclude that the empty prefix string was a covered bug.

515

How can I use i64 operands with V_AND_B32 that operates i32?

lib/Target/AMDGPU/VOPInstructions.td

576

Окау. All the stuff related this flag is going to be used later on when I add the patterns for extended encoding forms. I agree that it should be removed from this patch.

[rampitec]

lib/Target/AMDGPU/VOP2Instructions.td

364

I think it was supposed to suppress printing of _e32 suffix in a command like:

llvm-mc -arch=amdgcn -mcpu=fiji <<< 'v_madmk_f32 v0, v0, 1.0, v0'

Although we have tests for it in MC/Disassembler/AMDGPU/, so as long as they pass it should be fine.

515

OK, thanks.

test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll

20

This must be a leftover from some experiments, there is no such check.

test/CodeGen/AMDGPU/shift-i128.ll

7–24

Was it done with update_llc_test_checks.py? It is strange to see indents changed. I suspect the file was really changed manually instead.

[alex-t]

MC/Disassembler/AMDGPU passed
Tests fixed

[rampitec]

LGTM. Thanks!

[alex-t]

Committed r342719.

[hakzsam]

Hi there,

This change introduces a regression with RADV and Doom 2016, see https://bugs.freedesktop.org/show_bug.cgi?id=110636
We just discovered this, so LLVM 8 and master are affected. :(
I'm trying to find the root cause, I will let you know.

[arsenm]

In D52019#1496145, @hakzsam wrote:

Hi there,

This change introduces a regression with RADV and Doom 2016, see https://bugs.freedesktop.org/show_bug.cgi?id=110636
We just discovered this, so LLVM 8 and master are affected. :(
I'm trying to find the root cause, I will let you know.

Did r360293 fix it?

[hakzsam]

In D52019#1496159, @arsenm wrote:

In D52019#1496145, @hakzsam wrote:

Hi there,

This change introduces a regression with RADV and Doom 2016, see https://bugs.freedesktop.org/show_bug.cgi?id=110636
We just discovered this, so LLVM 8 and master are affected. :(
I'm trying to find the root cause, I will let you know.

Did r360293 fix it?

Yes, it did fix the problem.
Do you plan to backport to LLVM 8?

In D52019#1496159, @arsenm wrote:

In D52019#1496145, @hakzsam wrote:

Hi there,

This change introduces a regression with RADV and Doom 2016, see https://bugs.freedesktop.org/show_bug.cgi?id=110636
We just discovered this, so LLVM 8 and master are affected. :(
I'm trying to find the root cause, I will let you know.

Did r360293 fix it?

[arsenm]

In D52019#1496188, @hakzsam wrote:

In D52019#1496159, @arsenm wrote:

In D52019#1496145, @hakzsam wrote:

Hi there,

This change introduces a regression with RADV and Doom 2016, see https://bugs.freedesktop.org/show_bug.cgi?id=110636
We just discovered this, so LLVM 8 and master are affected. :(
I'm trying to find the root cause, I will let you know.

Did r360293 fix it?

Yes, it did fix the problem.
Do you plan to backport to LLVM 8?

https://bugs.llvm.org/show_bug.cgi?id=41811

lib/Target/AMDGPU/AMDGPU.td

def D16PreservesUnusedBits : Predicate<"Subtarget->d16PreservesUnusedBits()">,
   AssemblerPredicate<"FeatureD16PreservesUnusedBits">;
 
 def LDSRequiresM0Init : Predicate<"Subtarget->ldsRequiresM0Init()">;
 def NotLDSRequiresM0Init : Predicate<"!Subtarget->ldsRequiresM0Init()">;
 
 def HasDSAddTid : Predicate<"Subtarget->getGeneration() >= AMDGPUSubtarget::GFX9">,
   AssemblerPredicate<"FeatureGFX9Insts">;
 
-def HasAddNoCarryInsts : Predicate<"Subtarget->hasAddNoCarryInsts()">,
+def HasAddNoCarryInsts : Predicate<"Subtarget->hasAddNoCarry()">,
   AssemblerPredicate<"FeatureAddNoCarryInsts">;
 
-def NotHasAddNoCarryInsts : Predicate<"!Subtarget->hasAddNoCarryInsts()">,
+def NotHasAddNoCarryInsts : Predicate<"!Subtarget->hasAddNoCarry()">,
   AssemblerPredicate<"!FeatureAddNoCarryInsts">;
 
 def Has16BitInsts : Predicate<"Subtarget->has16BitInsts()">,
   AssemblerPredicate<"Feature16BitInsts">;
 def HasVOP3PInsts : Predicate<"Subtarget->hasVOP3PInsts()">,
   AssemblerPredicate<"FeatureVOP3P">;
 
 def NotHasVOP3PInsts : Predicate<"!Subtarget->hasVOP3PInsts()">,

lib/Target/AMDGPU/SIInstrInfo.td

class VOPProfile <list<ValueType> _ArgVT> {
 
   field bit IsPacked = isPackedType<Src0VT>.ret;
   field bit HasOpSel = IsPacked;
   field bit HasOMod = !if(HasOpSel, 0, isFloatType<DstVT>.ret);
   field bit HasSDWAOMod = isFloatType<DstVT>.ret;
 
   field bit HasExt = getHasExt<NumSrcArgs, DstVT, Src0VT, Src1VT>.ret;
   field bit HasSDWA9 = HasExt;
+  field bits<2> NeedPatGen = 0;
 
   field Operand Src0PackedMod = !if(HasSrc0FloatMods, PackedF16InputMods, PackedI16InputMods);
   field Operand Src1PackedMod = !if(HasSrc1FloatMods, PackedF16InputMods, PackedI16InputMods);
   field Operand Src2PackedMod = !if(HasSrc2FloatMods, PackedF16InputMods, PackedI16InputMods);
 
   field dag Outs = !if(HasDst,(outs DstRC:$vdst),(outs));
 
   // VOP3b instructions are a special case with a second explicit
class VOPProfile <list<ValueType> _ArgVT> {
   field string AsmSDWA9 = getAsmSDWA9<HasDst, HasSDWAOMod, NumSrcArgs, DstVT>.ret;
 }
 
 class VOP_NO_EXT <VOPProfile p> : VOPProfile <p.ArgVT> {
   let HasExt = 0;
   let HasSDWA9 = 0;
 }
 
+class VOP_PAT_GEN <VOPProfile p, bits<2> mode=2> : VOPProfile <p.ArgVT> {
+  let NeedPatGen = mode;
+}
+
 def VOP_F16_F16 : VOPProfile <[f16, f16, untyped, untyped]>;
 def VOP_F16_I16 : VOPProfile <[f16, i16, untyped, untyped]>;
 def VOP_I16_F16 : VOPProfile <[i16, f16, untyped, untyped]>;
 
 def VOP_F16_F16_F16 : VOPProfile <[f16, f16, f16, untyped]>;
 def VOP_F16_F16_I16 : VOPProfile <[f16, f16, i16, untyped]>;
 def VOP_F16_F16_I32 : VOPProfile <[f16, f16, i32, untyped]>;
 def VOP_I16_I16_I16 : VOPProfile <[i16, i16, i16, untyped]>;

lib/Target/AMDGPU/SIInstructions.td

 // all the instruction definitions were originally commented out.  Instructions
 // that are not yet supported remain commented out.
 //===----------------------------------------------------------------------===//
 
 class GCNPat<dag pattern, dag result> : Pat<pattern, result>, GCNPredicateControl {
   let SubtargetPredicate = isGCN;
 }
 
-include "VOPInstructions.td"
 include "SOPInstructions.td"
+include "VOPInstructions.td"
 include "SMInstructions.td"
 include "FLATInstructions.td"
 include "BUFInstructions.td"
 
 //===----------------------------------------------------------------------===//
 // EXP Instructions
 //===----------------------------------------------------------------------===//
 
multiclass SelectPat <ValueType vt, Instruction inst> {
   >;
 }
 
 defm : SelectPat <i16, V_CNDMASK_B32_e64>;
 defm : SelectPat <i32, V_CNDMASK_B32_e64>;
 defm : SelectPat <f16, V_CNDMASK_B32_e64>;
 defm : SelectPat <f32, V_CNDMASK_B32_e64>;
 
+let AddedComplexity = 1 in {
 def : GCNPat <
-  (i32 (add (i32 (ctpop i32:$popcnt)), i32:$val)),
+  (i32 (add (i32 (getDivergentFrag<ctpop>.ret i32:$popcnt)), i32:$val)),
   (V_BCNT_U32_B32_e64 $popcnt, $val)
 >;
+}
 def : GCNPat <
-  (i16 (add (i16 (trunc (ctpop i32:$popcnt))), i16:$val)),
+  (i16 (add (i16 (trunc (getDivergentFrag<ctpop>.ret i32:$popcnt))), i16:$val)),
   (V_BCNT_U32_B32_e64 $popcnt, $val)
 >;
 
 /********** ============================================ **********/
 /********** Extraction, Insertion, Building and Casting  **********/
 /********** ============================================ **********/
 
 foreach Index = 0-2 in {

lib/Target/AMDGPU/SOPInstructions.td

class SOP2_64_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
   "$sdst, $src0, $src1", pattern
 >;
 
 class SOP2_64_32_32 <string opName, list<dag> pattern=[]> : SOP2_Pseudo <
   opName, (outs SReg_64:$sdst), (ins SSrc_b32:$src0, SSrc_b32:$src1),
   "$sdst, $src0, $src1", pattern
 >;
 
+class UniformBinFrag<SDPatternOperator Op> : PatFrag <
+  (ops node:$src0, node:$src1),
+  (Op $src0, $src1),
+  [{ return !N->isDivergent(); }]
+>;
+
 let Defs = [SCC] in { // Carry out goes to SCC
 let isCommutable = 1 in {
 def S_ADD_U32 : SOP2_32 <"s_add_u32">;
 def S_ADD_I32 : SOP2_32 <"s_add_i32",
-  [(set i32:$sdst, (add SSrc_b32:$src0, SSrc_b32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<add> SSrc_b32:$src0, SSrc_b32:$src1))]
 >;
 } // End isCommutable = 1
 
 def S_SUB_U32 : SOP2_32 <"s_sub_u32">;
 def S_SUB_I32 : SOP2_32 <"s_sub_i32",
-  [(set i32:$sdst, (sub SSrc_b32:$src0, SSrc_b32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<sub> SSrc_b32:$src0, SSrc_b32:$src1))]
 >;
 
 let Uses = [SCC] in { // Carry in comes from SCC
 let isCommutable = 1 in {
 def S_ADDC_U32 : SOP2_32 <"s_addc_u32",
-  [(set i32:$sdst, (adde (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>;
+  [(set i32:$sdst, (UniformBinFrag<adde> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>;
 } // End isCommutable = 1
 
 def S_SUBB_U32 : SOP2_32 <"s_subb_u32",
-  [(set i32:$sdst, (sube (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>;
+  [(set i32:$sdst, (UniformBinFrag<sube> (i32 SSrc_b32:$src0), (i32 SSrc_b32:$src1)))]>;
 } // End Uses = [SCC]
 
 
 let isCommutable = 1 in {
 def S_MIN_I32 : SOP2_32 <"s_min_i32",
-  [(set i32:$sdst, (smin i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<smin> i32:$src0, i32:$src1))]
 >;
 def S_MIN_U32 : SOP2_32 <"s_min_u32",
-  [(set i32:$sdst, (umin i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<umin> i32:$src0, i32:$src1))]
 >;
 def S_MAX_I32 : SOP2_32 <"s_max_i32",
-  [(set i32:$sdst, (smax i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<smax> i32:$src0, i32:$src1))]
 >;
 def S_MAX_U32 : SOP2_32 <"s_max_u32",
-  [(set i32:$sdst, (umax i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<umax> i32:$src0, i32:$src1))]
 >;
 } // End isCommutable = 1
 } // End Defs = [SCC]
 
 
 let Uses = [SCC] in {
   def S_CSELECT_B32 : SOP2_32 <"s_cselect_b32">;
   def S_CSELECT_B64 : SOP2_64 <"s_cselect_b64">;
 } // End Uses = [SCC]
 
 let Defs = [SCC] in {
 let isCommutable = 1 in {
 def S_AND_B32 : SOP2_32 <"s_and_b32",
-  [(set i32:$sdst, (and i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<and> i32:$src0, i32:$src1))]
 >;
 
 def S_AND_B64 : SOP2_64 <"s_and_b64",
-  [(set i64:$sdst, (and i64:$src0, i64:$src1))]
+  [(set i64:$sdst, (UniformBinFrag<and> i64:$src0, i64:$src1))]
 >;
 
 def S_OR_B32 : SOP2_32 <"s_or_b32",
-  [(set i32:$sdst, (or i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<or> i32:$src0, i32:$src1))]
 >;
 
 def S_OR_B64 : SOP2_64 <"s_or_b64",
-  [(set i64:$sdst, (or i64:$src0, i64:$src1))]
+  [(set i64:$sdst, (UniformBinFrag<or> i64:$src0, i64:$src1))]
 >;
 
 def S_XOR_B32 : SOP2_32 <"s_xor_b32",
-  [(set i32:$sdst, (xor i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<xor> i32:$src0, i32:$src1))]
 >;
 
 def S_XOR_B64 : SOP2_64 <"s_xor_b64",
-  [(set i64:$sdst, (xor i64:$src0, i64:$src1))]
+  [(set i64:$sdst, (UniformBinFrag<xor> i64:$src0, i64:$src1))]
 >;
 
 def S_XNOR_B32 : SOP2_32 <"s_xnor_b32",
   [(set i32:$sdst, (not (xor_oneuse i32:$src0, i32:$src1)))]
 >;
 
 def S_XNOR_B64 : SOP2_64 <"s_xnor_b64",
   [(set i64:$sdst, (not (xor_oneuse i64:$src0, i64:$src1)))]

 def S_NOR_B32 : SOP2_32 <"s_nor_b32">;
 def S_NOR_B64 : SOP2_64 <"s_nor_b64">;
 } // End Defs = [SCC]
 
 // Use added complexity so these patterns are preferred to the VALU patterns.
 let AddedComplexity = 1 in {
 
 let Defs = [SCC] in {
+// TODO: b64 versions require VOP3 change since v_lshlrev_b64 is VOP3
 def S_LSHL_B32 : SOP2_32 <"s_lshl_b32",
-  [(set i32:$sdst, (shl i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<shl> i32:$src0, i32:$src1))]
 >;
 def S_LSHL_B64 : SOP2_64_32 <"s_lshl_b64",
   [(set i64:$sdst, (shl i64:$src0, i32:$src1))]
 >;
 def S_LSHR_B32 : SOP2_32 <"s_lshr_b32",
-  [(set i32:$sdst, (srl i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<srl> i32:$src0, i32:$src1))]
 >;
 def S_LSHR_B64 : SOP2_64_32 <"s_lshr_b64",
   [(set i64:$sdst, (srl i64:$src0, i32:$src1))]
 >;
 def S_ASHR_I32 : SOP2_32 <"s_ashr_i32",
-  [(set i32:$sdst, (sra i32:$src0, i32:$src1))]
+  [(set i32:$sdst, (UniformBinFrag<sra> i32:$src0, i32:$src1))]
 >;
 def S_ASHR_I64 : SOP2_64_32 <"s_ashr_i64",
   [(set i64:$sdst, (sra i64:$src0, i32:$src1))]
 >;
 } // End Defs = [SCC]
 
 def S_BFM_B32 : SOP2_32 <"s_bfm_b32",
-  [(set i32:$sdst, (AMDGPUbfm i32:$src0, i32:$src1))]>;
+  [(set i32:$sdst, (UniformBinFrag<AMDGPUbfm> i32:$src0, i32:$src1))]>;
 def S_BFM_B64 : SOP2_64_32_32 <"s_bfm_b64">;
+
+// TODO: S_MUL_I32 require V_MUL_LO_I32 from VOP3 change
 def S_MUL_I32 : SOP2_32 <"s_mul_i32",
   [(set i32:$sdst, (mul i32:$src0, i32:$src1))]> {
   let isCommutable = 1;
 }
 
 } // End AddedComplexity = 1
 
 let Defs = [SCC] in {

lib/Target/AMDGPU/VOP2Instructions.td

 multiclass VOP2Inst <string opName,
                      VOPProfile P,
                      SDPatternOperator node = null_frag,
                      string revOp = opName,
                      bit GFX9Renamed = 0> {
 
   let renamedInGFX9 = GFX9Renamed in {
 
-    def _e32 : VOP2_Pseudo <opName, P>,
+    def _e32 : VOP2_Pseudo <opName, P, VOPPatOrNull<node,P>.ret>,
                Commutable_REV<revOp#"_e32", !eq(revOp, opName)>;
   
-    def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>,
+    def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<OpOrNull<node, P>.ret, P>.ret>,
                Commutable_REV<revOp#"_e64", !eq(revOp, opName)>;
 
     def _sdwa  : VOP2_SDWA_Pseudo <opName, P>;
 
   }
 }
 
 multiclass VOP2bInst <string opName,
                       VOPProfile P,
                       SDPatternOperator node = null_frag,
                       string revOp = opName,
                       bit GFX9Renamed = 0,
                       bit useSGPRInput = !eq(P.NumSrcArgs, 3)> {
   let renamedInGFX9 = GFX9Renamed in {
     let SchedRW = [Write32Bit, WriteSALU] in {
       let Uses = !if(useSGPRInput, [VCC, EXEC], [EXEC]), Defs = [VCC] in {
-        def _e32 : VOP2_Pseudo <opName, P>,
+        def _e32 : VOP2_Pseudo <opName, P, VOPPatOrNull<node,P>.ret>,
                    Commutable_REV<revOp#"_e32", !eq(revOp, opName)>;
 
         def _sdwa  : VOP2_SDWA_Pseudo <opName, P> {
           let AsmMatchConverter = "cvtSdwaVOP2b";
         }
       }
 
-      def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>,
+      def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<OpOrNull<node, P>.ret, P>.ret>,
                  Commutable_REV<revOp#"_e64", !eq(revOp, opName)>;
     }
   }
 }
 
 multiclass VOP2eInst <string opName,
                       VOPProfile P,
                       SDPatternOperator node = null_frag,

 }
 
 //===----------------------------------------------------------------------===//
 // VOP2 Instructions
 //===----------------------------------------------------------------------===//
 
 let SubtargetPredicate = isGCN in {
 
-defm V_CNDMASK_B32 : VOP2eInst <"v_cndmask_b32", VOP2e_I32_I32_I32_I1>;
 def V_MADMK_F32 : VOP2_Pseudo <"v_madmk_f32", VOP_MADMK_F32, [], "">;
+let isCommutable = 1, AddedComplexity = 13 in {
+  def V_MADAK_F32 : VOP2_Pseudo <"v_madak_f32", VOP_MADAK_F32, 
+    [(set f32:$vdst, (fadd (fmul f32:$src0, f32:$src1), fpimm:$imm))], "">;
+}
+// These are special and do not read the exec mask.
+let isConvergent = 1, Uses = []<Register> in {
+def V_READLANE_B32 : VOP2_Pseudo<"v_readlane_b32", VOP_READLANE,
+  [(set i32:$vdst, (int_amdgcn_readlane i32:$src0, i32:$src1))], "">;
+
+let Constraints = "$vdst = $vdst_in", DisableEncoding="$vdst_in" in {
+def V_WRITELANE_B32 : VOP2_Pseudo<"v_writelane_b32", VOP_WRITELANE,
+  [(set i32:$vdst, (int_amdgcn_writelane i32:$src0, i32:$src1, i32:$vdst_in))], "">;
+} // End $vdst = $vdst_in, DisableEncoding $vdst_in
+} // End isConvergent = 1
+}
+
+let SubtargetPredicate = isGCN, Predicates = [isGCN] in {
+
+defm V_CNDMASK_B32 : VOP2eInst <"v_cndmask_b32", VOP2e_I32_I32_I32_I1>;
+
 
 let isCommutable = 1 in {
 defm V_ADD_F32 : VOP2Inst <"v_add_f32", VOP_F32_F32_F32, fadd>;
 defm V_SUB_F32 : VOP2Inst <"v_sub_f32", VOP_F32_F32_F32, fsub>;
 defm V_SUBREV_F32 : VOP2Inst <"v_subrev_f32", VOP_F32_F32_F32, null_frag, "v_sub_f32">;
 defm V_MUL_LEGACY_F32 : VOP2Inst <"v_mul_legacy_f32", VOP_F32_F32_F32, AMDGPUfmul_legacy>;
 defm V_MUL_F32 : VOP2Inst <"v_mul_f32", VOP_F32_F32_F32, fmul>;
-defm V_MUL_I32_I24 : VOP2Inst <"v_mul_i32_i24", VOP_I32_I32_I32, AMDGPUmul_i24>;
-defm V_MUL_HI_I32_I24 : VOP2Inst <"v_mul_hi_i32_i24", VOP_I32_I32_I32, AMDGPUmulhi_i24>;
-defm V_MUL_U32_U24 : VOP2Inst <"v_mul_u32_u24", VOP_I32_I32_I32, AMDGPUmul_u24>;
-defm V_MUL_HI_U32_U24 : VOP2Inst <"v_mul_hi_u32_u24", VOP_I32_I32_I32, AMDGPUmulhi_u24>;
+defm V_MUL_I32_I24 : VOP2Inst <"v_mul_i32_i24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmul_i24>;
+defm V_MUL_HI_I32_I24 : VOP2Inst <"v_mul_hi_i32_i24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmulhi_i24>;
+defm V_MUL_U32_U24 : VOP2Inst <"v_mul_u32_u24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmul_u24>;
+defm V_MUL_HI_U32_U24 : VOP2Inst <"v_mul_hi_u32_u24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmulhi_u24>;
 defm V_MIN_F32 : VOP2Inst <"v_min_f32", VOP_F32_F32_F32, fminnum>;
 defm V_MAX_F32 : VOP2Inst <"v_max_f32", VOP_F32_F32_F32, fmaxnum>;
-defm V_MIN_I32 : VOP2Inst <"v_min_i32", VOP_I32_I32_I32>;
-defm V_MAX_I32 : VOP2Inst <"v_max_i32", VOP_I32_I32_I32>;
-defm V_MIN_U32 : VOP2Inst <"v_min_u32", VOP_I32_I32_I32>;
-defm V_MAX_U32 : VOP2Inst <"v_max_u32", VOP_I32_I32_I32>;
+defm V_MIN_I32 : VOP2Inst <"v_min_i32", VOP_PAT_GEN<VOP_I32_I32_I32>, smin>;
+defm V_MAX_I32 : VOP2Inst <"v_max_i32", VOP_PAT_GEN<VOP_I32_I32_I32>, smax>;
+defm V_MIN_U32 : VOP2Inst <"v_min_u32", VOP_PAT_GEN<VOP_I32_I32_I32>, umin>;
+defm V_MAX_U32 : VOP2Inst <"v_max_u32", VOP_PAT_GEN<VOP_I32_I32_I32>, umax>;
+
+let SubtargetPredicate = isGCN in {
 defm V_LSHRREV_B32 : VOP2Inst <"v_lshrrev_b32", VOP_I32_I32_I32, null_frag, "v_lshr_b32">;
 defm V_ASHRREV_I32 : VOP2Inst <"v_ashrrev_i32", VOP_I32_I32_I32, null_frag, "v_ashr_i32">;
 defm V_LSHLREV_B32 : VOP2Inst <"v_lshlrev_b32", VOP_I32_I32_I32, null_frag, "v_lshl_b32">;
-defm V_AND_B32 : VOP2Inst <"v_and_b32", VOP_I32_I32_I32>;
-defm V_OR_B32 : VOP2Inst <"v_or_b32", VOP_I32_I32_I32>;
-defm V_XOR_B32 : VOP2Inst <"v_xor_b32", VOP_I32_I32_I32>;
+}
+defm V_AND_B32 : VOP2Inst <"v_and_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, and>;
+defm V_OR_B32 : VOP2Inst <"v_or_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, or>;
+defm V_XOR_B32 : VOP2Inst <"v_xor_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, xor>;
 
 let Constraints = "$vdst = $src2", DisableEncoding="$src2",
     isConvertibleToThreeAddress = 1 in {
 defm V_MAC_F32 : VOP2Inst <"v_mac_f32", VOP_MAC_F32>;
 }
 
-def V_MADAK_F32 : VOP2_Pseudo <"v_madak_f32", VOP_MADAK_F32, [], "">;
+
 
 // No patterns so that the scalar instructions are always selected.
 // The scalar versions will be replaced with vector when needed later.
 
 // V_ADD_I32, V_SUB_I32, and V_SUBREV_I32 where renamed to *_U32 in VI,
 // but the VI instructions behave the same as the SI versions.
-defm V_ADD_I32 : VOP2bInst <"v_add_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_add_i32", 1>;
-defm V_SUB_I32 : VOP2bInst <"v_sub_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_sub_i32", 1>;
-defm V_SUBREV_I32 : VOP2bInst <"v_subrev_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_sub_i32", 1>;
+
+// TODO: VOPPatGen only supports generation the pattern from P.Ins32,
+// we need to be able to handle P.Ins64 as well
 defm V_ADDC_U32 : VOP2bInst <"v_addc_u32", VOP2b_I32_I1_I32_I32_I1, null_frag, "v_addc_u32", 1>;
 defm V_SUBB_U32 : VOP2bInst <"v_subb_u32", VOP2b_I32_I1_I32_I32_I1, null_frag, "v_subb_u32", 1>;
 defm V_SUBBREV_U32 : VOP2bInst <"v_subbrev_u32", VOP2b_I32_I1_I32_I32_I1, null_frag, "v_subb_u32", 1>;
 
+defm V_ADD_I32 : VOP2bInst <"v_add_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_add_i32", 1>;
+defm V_SUB_I32 : VOP2bInst <"v_sub_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_sub_i32", 1>;
+defm V_SUBREV_I32 : VOP2bInst <"v_subrev_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_sub_i32", 1>;
 
 let SubtargetPredicate = HasAddNoCarryInsts in {
  defm V_ADD_U32 : VOP2Inst <"v_add_u32", VOP_I32_I32_I32, null_frag, "v_add_u32", 1>;
  defm V_SUB_U32 : VOP2Inst <"v_sub_u32", VOP_I32_I32_I32, null_frag, "v_sub_u32", 1>;
  defm V_SUBREV_U32 : VOP2Inst <"v_subrev_u32", VOP_I32_I32_I32, null_frag, "v_sub_u32", 1>;
 }
 
 } // End isCommutable = 1
 
-// These are special and do not read the exec mask.
-let isConvergent = 1, Uses = []<Register> in {
-def V_READLANE_B32 : VOP2_Pseudo<"v_readlane_b32", VOP_READLANE,
-  [(set i32:$vdst, (int_amdgcn_readlane i32:$src0, i32:$src1))], "">;
-
-let Constraints = "$vdst = $vdst_in", DisableEncoding="$vdst_in" in {
-def V_WRITELANE_B32 : VOP2_Pseudo<"v_writelane_b32", VOP_WRITELANE,
-  [(set i32:$vdst, (int_amdgcn_writelane i32:$src0, i32:$src1, i32:$vdst_in))], "">;
-} // End $vdst = $vdst_in, DisableEncoding $vdst_in
-} // End isConvergent = 1
 
 defm V_BFM_B32 : VOP2Inst <"v_bfm_b32", VOP_NO_EXT<VOP_I32_I32_I32>>;
 defm V_BCNT_U32_B32 : VOP2Inst <"v_bcnt_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>>;
 defm V_MBCNT_LO_U32_B32 : VOP2Inst <"v_mbcnt_lo_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, int_amdgcn_mbcnt_lo>;
 defm V_MBCNT_HI_U32_B32 : VOP2Inst <"v_mbcnt_hi_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, int_amdgcn_mbcnt_hi>;
 defm V_LDEXP_F32 : VOP2Inst <"v_ldexp_f32", VOP_NO_EXT<VOP_F32_F32_I32>, AMDGPUldexp>;
 defm V_CVT_PKACCUM_U8_F32 : VOP2Inst <"v_cvt_pkaccum_u8_f32", VOP_NO_EXT<VOP_I32_F32_I32>>; // TODO: set "Uses = dst"
 defm V_CVT_PKNORM_I16_F32 : VOP2Inst <"v_cvt_pknorm_i16_f32", VOP_NO_EXT<VOP_V2I16_F32_F32>, AMDGPUpknorm_i16_f32>;
 defm V_CVT_PKNORM_U16_F32 : VOP2Inst <"v_cvt_pknorm_u16_f32", VOP_NO_EXT<VOP_V2I16_F32_F32>, AMDGPUpknorm_u16_f32>;
 defm V_CVT_PKRTZ_F16_F32 : VOP2Inst <"v_cvt_pkrtz_f16_f32", VOP_NO_EXT<VOP_V2F16_F32_F32>, AMDGPUpkrtz_f16_f32>;
 defm V_CVT_PK_U16_U32 : VOP2Inst <"v_cvt_pk_u16_u32", VOP_NO_EXT<VOP_V2I16_I32_I32>, AMDGPUpk_u16_u32>;
 defm V_CVT_PK_I16_I32 : VOP2Inst <"v_cvt_pk_i16_i32", VOP_NO_EXT<VOP_V2I16_I32_I32>, AMDGPUpk_i16_i32>;
 
-} // End SubtargetPredicate = isGCN
+} // End SubtargetPredicate = isGCN, Predicates = [isGCN]
 
-def : GCNPat<
-    (AMDGPUadde i32:$src0, i32:$src1, i1:$src2),
-    (V_ADDC_U32_e64 $src0, $src1, $src2)
->;
-
-def : GCNPat<
-    (AMDGPUsube i32:$src0, i32:$src1, i1:$src2),
-    (V_SUBB_U32_e64 $src0, $src1, $src2)
->;
 
 // These instructions only exist on SI and CI
-let SubtargetPredicate = isSICI in {
+let SubtargetPredicate = isSICI, Predicates = [isSICI] in {
 
 defm V_MIN_LEGACY_F32 : VOP2Inst <"v_min_legacy_f32", VOP_F32_F32_F32, AMDGPUfmin_legacy>;
 defm V_MAX_LEGACY_F32 : VOP2Inst <"v_max_legacy_f32", VOP_F32_F32_F32, AMDGPUfmax_legacy>;
 
 let isCommutable = 1 in {
 defm V_MAC_LEGACY_F32 : VOP2Inst <"v_mac_legacy_f32", VOP_F32_F32_F32>;
-defm V_LSHR_B32 : VOP2Inst <"v_lshr_b32", VOP_I32_I32_I32>;
-defm V_ASHR_I32 : VOP2Inst <"v_ashr_i32", VOP_I32_I32_I32>;
-defm V_LSHL_B32 : VOP2Inst <"v_lshl_b32", VOP_I32_I32_I32>;
+defm V_LSHR_B32 : VOP2Inst <"v_lshr_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, srl>;
+defm V_ASHR_I32 : VOP2Inst <"v_ashr_i32", VOP_PAT_GEN<VOP_I32_I32_I32>, sra>;
+defm V_LSHL_B32 : VOP2Inst <"v_lshl_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, shl>;
 } // End isCommutable = 1
 
-} // End let SubtargetPredicate = SICI
+} // End let SubtargetPredicate = SICI, Predicates = [isSICI]
+
+class DivergentBinOp<SDPatternOperator Op, VOP_Pseudo Inst> :
+  GCNPat<
+      (getDivergentFrag<Op>.ret Inst.Pfl.Src0VT:$src0, Inst.Pfl.Src1VT:$src1),
+      !if(!cast<Commutable_REV>(Inst).IsOrig, 
+        (Inst $src0, $src1),
+        (Inst $src1, $src0)
+      )
+  >;
+
+let AddedComplexity = 1 in {
+  def : DivergentBinOp<srl, V_LSHRREV_B32_e64>;
+  def : DivergentBinOp<sra, V_ASHRREV_I32_e64>;
+  def : DivergentBinOp<shl, V_LSHLREV_B32_e64>;
+}
+
+let SubtargetPredicate = HasAddNoCarryInsts in {
+  def : DivergentBinOp<add, V_ADD_U32_e32>;
+  def : DivergentBinOp<sub, V_SUB_U32_e32>;
+  def : DivergentBinOp<sub, V_SUBREV_U32_e32>;
+}
+
+
+def : DivergentBinOp<add, V_ADD_I32_e32>;
+
+def : DivergentBinOp<add, V_ADD_I32_e64>;
+def : DivergentBinOp<sub, V_SUB_I32_e32>;
+
+def : DivergentBinOp<sub, V_SUBREV_I32_e32>;
+
+def : DivergentBinOp<srl, V_LSHRREV_B32_e32>;
+def : DivergentBinOp<sra, V_ASHRREV_I32_e32>;
+def : DivergentBinOp<shl, V_LSHLREV_B32_e32>;
+def : DivergentBinOp<adde, V_ADDC_U32_e32>;
+def : DivergentBinOp<sube, V_SUBB_U32_e32>;
+
+def : GCNPat<
+    (AMDGPUsube i32:$src0, i32:$src1, i1:$src2),
+    (V_SUBB_U32_e64 $src0, $src1, $src2)
+>;
+
+def : GCNPat<
+    (AMDGPUadde i32:$src0, i32:$src1, i1:$src2),
+    (V_ADDC_U32_e64 $src0, $src1, $src2)
+>;
+
+class divergent_i64_BinOp <SDPatternOperator Op, Instruction Inst> :
+  GCNPat<
+      (getDivergentFrag<Op>.ret i64:$src0, i64:$src1),
+      (REG_SEQUENCE VReg_64,
+        (Inst  
+          (i32 (EXTRACT_SUBREG $src0, sub0)),
+          (i32 (EXTRACT_SUBREG $src1, sub0))
+        ), sub0,
+        (Inst  
+          (i32 (EXTRACT_SUBREG $src0, sub1)),
+          (i32 (EXTRACT_SUBREG $src1, sub1))
+        ), sub1
+      )
+  >;  
+
+def :  divergent_i64_BinOp <and, V_AND_B32_e32>;
+def :  divergent_i64_BinOp <or,  V_OR_B32_e32>;
+def :  divergent_i64_BinOp <xor, V_XOR_B32_e32>;
 
 let SubtargetPredicate = Has16BitInsts in {
 
 def V_MADMK_F16 : VOP2_Pseudo <"v_madmk_f16", VOP_MADMK_F16, [], "">;
 defm V_LSHLREV_B16 : VOP2Inst <"v_lshlrev_b16", VOP_I16_I16_I16>;
 defm V_LSHRREV_B16 : VOP2Inst <"v_lshrrev_b16", VOP_I16_I16_I16>;
 defm V_ASHRREV_I16 : VOP2Inst <"v_ashrrev_i16", VOP_I16_I16_I16>;
 defm V_LDEXP_F16 : VOP2Inst <"v_ldexp_f16", VOP_F16_F16_I32, AMDGPUldexp>;

lib/Target/AMDGPU/VOPInstructions.td

class VOP_DPP <string OpName, VOPProfile P> :
   let AssemblerPredicate = !if(P.HasExt, HasDPP, DisableInst);
   let AsmVariantName = !if(P.HasExt, AMDGPUAsmVariants.DPP,
                                      AMDGPUAsmVariants.Disable);
   let Constraints = !if(P.NumSrcArgs, "$old = $vdst", "");
   let DisableEncoding = !if(P.NumSrcArgs, "$old", "");
   let DecoderNamespace = "DPP";
 }
 
+class getNumNodeArgs<SDPatternOperator Op> {
+  SDNode N = !cast<SDNode>(Op);
+  SDTypeProfile TP = N.TypeProfile;
+  int ret = TP.NumOperands;
+}
+
+
+class getDivergentFrag<SDPatternOperator Op> {
+
+  int NumSrcArgs = getNumNodeArgs<Op>.ret;
+  PatFrag ret = PatFrag <
+    !if(!eq(NumSrcArgs, 1),
+             (ops node:$src0),
+             !if(!eq(NumSrcArgs, 2),
+               (ops node:$src0, node:$src1),
+               (ops node:$src0, node:$src1, node:$src2))),
+    !if(!eq(NumSrcArgs, 1),
+             (Op $src0),
+             !if(!eq(NumSrcArgs, 2),
+               (Op $src0, $src1),
+               (Op $src0, $src1, $src2))),
+    [{ return N->isDivergent(); }]
+  >;
+}
+
+class VOPPatGen<SDPatternOperator Op, VOPProfile P> {
+
+  PatFrag Operator = getDivergentFrag < Op >.ret;
+  
+  dag Ins = !foreach(tmp, P.Ins32, !subst(ins, Operator, 
+                                         !subst(P.Src0RC32, P.Src0VT,
+                                               !subst(P.Src1RC32, P.Src1VT, tmp))));
+
+
+  dag Outs = !foreach(tmp, P.Outs32, !subst(outs, set, 
+                                           !subst(P.DstRC, P.DstVT, tmp)));
+
+  list<dag> ret =  [!con(Outs, (set Ins))];
+}
+
+class VOPPatOrNull<SDPatternOperator Op, VOPProfile P> {
+  list<dag> ret = !if(!ne(P.NeedPatGen,0), VOPPatGen<Op, P>.ret, []);
+}
+
+class OpOrNull<SDPatternOperator Op, VOPProfile P> {
+  SDPatternOperator ret = !if(!eq(P.NeedPatGen, 1), null_frag, Op);
+}
+
+
 include "VOPCInstructions.td"
 include "VOP1Instructions.td"
 include "VOP2Instructions.td"
 include "VOP3Instructions.td"
 include "VOP3PInstructions.td"

test/CodeGen/AMDGPU/add.ll

entry:
   %0 = add <16 x i32> %a, %b
   store <16 x i32> %0, <16 x i32> addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_add_i32:
 ; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
 ; GCN: {{buffer|flat|global}}_load_dword [[B:v[0-9]+]]
-; SIVI: v_add_{{i|u}}32_e32 v{{[0-9]+}}, vcc, [[B]], [[A]]
+; SIVI: v_add_{{i|u}}32_e32 v{{[0-9]+}}, vcc, [[A]], [[B]]
 ; GFX9: v_add_u32_e32 v{{[0-9]+}}, [[A]], [[B]]
 define amdgpu_kernel void @v_add_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %tid
   %b_ptr = getelementptr i32, i32 addrspace(1)* %gep, i32 1
   %a = load volatile i32, i32 addrspace(1)* %gep
   %b = load volatile i32, i32 addrspace(1)* %b_ptr
   %result = add i32 %a, %b

test/CodeGen/AMDGPU/amdgcn.private-memory.ll

 ; RUN: llc -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-PROMOTE %s
 ; RUN: llc -mattr=+promote-alloca,-flat-for-global -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-PROMOTE -check-prefix=HSA %s
 ; RUN: llc -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-ALLOCA %s
 ; RUN: llc -mattr=-promote-alloca,-flat-for-global -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri < %s | FileCheck  -check-prefix=GCN -check-prefix=GCN-ALLOCA -check-prefix=HSA %s
 ; RUN: llc -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=GCN-PROMOTE %s
 ; RUN: llc -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN  -check-prefix=GCN-ALLOCA %s
 
 
 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 
 
 ; Make sure we don't overwrite workitem information with private memory
 
 ; GCN-LABEL: {{^}}work_item_info:
 ; GCN-NOT: v0
-; GCN: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, v0, v{{[0-9]+}}
+; GCN: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, v{{[0-9]+}}, v0
 ; GCN: buffer_store_dword [[RESULT]]
 define amdgpu_kernel void @work_item_info(i32 addrspace(1)* %out, i32 %in) {
 entry:
   %0 = alloca [2 x i32], addrspace(5)
   %1 = getelementptr [2 x i32], [2 x i32] addrspace(5)* %0, i32 0, i32 0
   %2 = getelementptr [2 x i32], [2 x i32] addrspace(5)* %0, i32 0, i32 1
   store i32 0, i32 addrspace(5)* %1
   store i32 1, i32 addrspace(5)* %2
   %3 = getelementptr [2 x i32], [2 x i32] addrspace(5)* %0, i32 0, i32 %in
   %4 = load i32, i32 addrspace(5)* %3
   %5 = call i32 @llvm.amdgcn.workitem.id.x()
   %6 = add i32 %4, %5
   store i32 %6, i32 addrspace(1)* %out
   ret void
 }

test/CodeGen/AMDGPU/bfe-patterns.ll

define amdgpu_kernel void @v_ubfe_sub_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #1 {
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_ubfe_sub_multi_use_shl_i32:
 ; GCN: {{buffer|flat}}_load_dword [[SRC:v[0-9]+]]
 ; GCN: {{buffer|flat}}_load_dword [[WIDTH:v[0-9]+]]
 ; GCN: v_sub_{{[iu]}}32_e32 [[SUB:v[0-9]+]], vcc, 32, [[WIDTH]]
 
-; SI-NEXT: v_lshl_b32_e32 [[SHL:v[0-9]+]], [[SRC]], [[SUB]]
-; SI-NEXT: v_lshr_b32_e32 [[BFE:v[0-9]+]], [[SHL]], [[SUB]]
-
-; VI-NEXT: v_lshlrev_b32_e32 [[SHL:v[0-9]+]], [[SUB]], [[SRC]]
-; VI-NEXT: v_lshrrev_b32_e32 [[BFE:v[0-9]+]], [[SUB]], [[SHL]]
+; GCN-NEXT: v_lshlrev_b32_e32 [[SHL:v[0-9]+]], [[SUB]], [[SRC]]
+; GCN-NEXT: v_lshrrev_b32_e32 [[BFE:v[0-9]+]], [[SUB]], [[SHL]]
 
 ; GCN: [[BFE]]
 ; GCN: [[SHL]]
 define amdgpu_kernel void @v_ubfe_sub_multi_use_shl_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #1 {
   %id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
   %in0.gep = getelementptr i32, i32 addrspace(1)* %in0, i32 %id.x
   %in1.gep = getelementptr i32, i32 addrspace(1)* %in1, i32 %id.x
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %id.x
define amdgpu_kernel void @v_sbfe_sub_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #1 {
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_sbfe_sub_multi_use_shl_i32:
 ; GCN: {{buffer|flat}}_load_dword [[SRC:v[0-9]+]]
 ; GCN: {{buffer|flat}}_load_dword [[WIDTH:v[0-9]+]]
 ; GCN: v_sub_{{[iu]}}32_e32 [[SUB:v[0-9]+]], vcc, 32, [[WIDTH]]
 
-; SI-NEXT: v_lshl_b32_e32 [[SHL:v[0-9]+]], [[SRC]], [[SUB]]
-; SI-NEXT: v_ashr_i32_e32 [[BFE:v[0-9]+]], [[SHL]], [[SUB]]
-
-; VI-NEXT: v_lshlrev_b32_e32 [[SHL:v[0-9]+]], [[SUB]], [[SRC]]
-; VI-NEXT: v_ashrrev_i32_e32 [[BFE:v[0-9]+]], [[SUB]], [[SHL]]
+; GCN-NEXT: v_lshlrev_b32_e32 [[SHL:v[0-9]+]], [[SUB]], [[SRC]]
+; GCN-NEXT: v_ashrrev_i32_e32 [[BFE:v[0-9]+]], [[SUB]], [[SHL]]
 
 ; GCN: [[BFE]]
 ; GCN: [[SHL]]
 define amdgpu_kernel void @v_sbfe_sub_multi_use_shl_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in0, i32 addrspace(1)* %in1) #1 {
   %id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
   %in0.gep = getelementptr i32, i32 addrspace(1)* %in0, i32 %id.x
   %in1.gep = getelementptr i32, i32 addrspace(1)* %in1, i32 %id.x
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %id.x

test/CodeGen/AMDGPU/ctpop64.ll

 ; VI: flat_load_dwordx4   v{{\[}}[[VAL0:[0-9]+]]:[[VAL3:[0-9]+]]{{\]}}, v{{\[[0-9]+:[0-9]+\]}}
 
 ; GCN-DAG: v_bcnt_u32_b32{{(_e64)*}} [[MIDRESULT0:v[0-9]+]], v{{[0-9]+}}, 0
 ; GCN-DAG: v_bcnt_u32_b32{{(_e32)*(_e64)*}} [[MIDRESULT1:v[0-9]+]], v[[VAL3]], [[MIDRESULT0]]
 
 ; GCN-DAG: v_bcnt_u32_b32{{(_e64)*}} [[MIDRESULT2:v[0-9]+]], v[[VAL0]], 0
 ; GCN-DAG: v_bcnt_u32_b32{{(_e32)*(_e64)*}} [[MIDRESULT3:v[0-9]+]], v{{[0-9]+}}, [[MIDRESULT2]]
 
-; GCN: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, [[MIDRESULT1]], [[MIDRESULT2]]
+; GCN: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, [[MIDRESULT2]], [[MIDRESULT1]]
 
 ; GCN: buffer_store_dword [[RESULT]],
 ; GCN: s_endpgm
 define amdgpu_kernel void @v_ctpop_i128(i32 addrspace(1)* noalias %out, i128 addrspace(1)* noalias %in) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %in.gep = getelementptr i128, i128 addrspace(1)* %in, i32 %tid
   %val = load i128, i128 addrspace(1)* %in.gep, align 8
   %ctpop = call i128 @llvm.ctpop.i128(i128 %val) nounwind readnone
   %truncctpop = trunc i128 %ctpop to i32
   store i32 %truncctpop, i32 addrspace(1)* %out, align 4
   ret void
 }

test/CodeGen/AMDGPU/extract-lowbits.ll

 }
 
 define i32 @bzhi32_d1_indexzext(i32 %val, i8 %numlowbits) nounwind {
 ; SI-LABEL: bzhi32_d1_indexzext:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; SI-NEXT:    v_sub_i32_e32 v1, vcc, 32, v1
 ; SI-NEXT:    v_and_b32_e32 v1, 0xff, v1
-; SI-NEXT:    v_lshl_b32_e32 v0, v0, v1
-; SI-NEXT:    v_lshr_b32_e32 v0, v0, v1
+; SI-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; SI-NEXT:    v_lshrrev_b32_e32 v0, v1, v0
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: bzhi32_d1_indexzext:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; VI-NEXT:    v_sub_u16_e32 v1, 32, v1
 ; VI-NEXT:    v_and_b32_e32 v1, 0xff, v1
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
 ; VI-NEXT:    v_lshrrev_b32_e32 v0, v1, v0
 ; VI-NEXT:    s_setpc_b64 s[30:31]
   %numhighbits = sub i8 32, %numlowbits
   %sh_prom = zext i8 %numhighbits to i32
   %highbitscleared = shl i32 %val, %sh_prom
   %masked = lshr i32 %highbitscleared, %sh_prom
   ret i32 %masked
 }

test/CodeGen/AMDGPU/fabs.f16.ll

 }
 
 ; FIXME: Should do fabs after conversion to avoid converting multiple
 ; times in this particular case.
 
 ; GCN-LABEL: {{^}}v_fabs_fold_self_v2f16:
 ; GCN: {{flat|global}}_load_dword [[VAL:v[0-9]+]]
 
-; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
-; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
-; CI: v_cvt_f32_f16_e32
-; CI: v_cvt_f32_f16_e32
-; CI: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+; CI: v_lshrrev_b32_e32 [[VREG:v[0-9]+]], 16, v{{[0-9]+}}
+; CI: v_cvt_f32_f16_e32 [[NORM:v[0-9]+]], [[VREG]]
+; CI: v_cvt_f32_f16_e64 [[ABS:v[0-9]+]], {{\|}}[[VREG]]{{\|}}
+; CI: v_mul_f32_e32 v{{[0-9]+}}, [[ABS]], [[NORM]]
 ; CI: v_cvt_f16_f32
 ; CI: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 ; CI: v_cvt_f16_f32
 
 ; VI: v_mul_f16_sdwa v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; VI: v_mul_f16_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
 
 ; GFX9: v_and_b32_e32 [[FABS:v[0-9]+]], 0x7fff7fff, [[VAL]]

test/CodeGen/AMDGPU/insert_vector_elt.v2i16.ll

 ; GCN-DAG: {{flat|global}}_load_dword [[VEC:v[0-9]+]]
 
 ; GFX9-DAG: s_movk_i32 [[K:s[0-9]+]], 0x3e7
 ; GFX9-DAG: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VEC]]
 ; GFX9: v_lshl_or_b32 [[RES:v[0-9]+]], [[K]], 16, [[ELT0]]
 
 ; CI: v_and_b32_e32 [[AND:v[0-9]+]], 0xffff, [[VEC]]
 ; CI: v_or_b32_e32 [[RES:v[0-9]+]], 0x3e70000, [[AND]]
-; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[VEC]], [[K]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[K]], [[VEC]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
 
 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RES]]
 define amdgpu_kernel void @v_insertelement_v2i16_1(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 999, i32 1
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_insertelement_v2i16_1_inlineimm:
 ; VI: v_mov_b32_e32 [[K:v[0-9]+]], 0xfff10000
 ; GCN: {{flat|global}}_load_dword [[VEC:v[0-9]+]]
 ; CI:   v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VEC]]
 ; GFX9: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VEC]]
 ; CI: v_or_b32_e32 [[RES:v[0-9]+]], 0xfff10000, [[ELT0]]
-; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[VEC]], [[K]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[K]], [[VEC]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
 ; GFX9: v_lshl_or_b32 [[RES:v[0-9]+]], -15, 16, [[ELT0]]
 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RES]]
 define amdgpu_kernel void @v_insertelement_v2i16_1_inlineimm(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep

 
 ; GFX9-DAG: s_movk_i32 [[K:s[0-9]+]], 0x4500
 ; GFX9-DAG: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VEC]]
 ; GFX9: v_lshl_or_b32 [[RES:v[0-9]+]], [[K]], 16, [[ELT0]]
 
 ; CI: v_and_b32_e32 [[AND:v[0-9]+]], 0xffff, [[VEC]]
 ; CI: v_or_b32_e32 [[RES:v[0-9]+]], 0x45000000, [[AND]]
 
-; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[VEC]], [[K]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[K]], [[VEC]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
 
 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RES]]
 define amdgpu_kernel void @v_insertelement_v2f16_1(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x half>, <2 x half> addrspace(1)* %in.gep
   %vecins = insertelement <2 x half> %vec, half 5.000000e+00, i32 1
   store <2 x half> %vecins, <2 x half> addrspace(1)* %out.gep
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_insertelement_v2f16_1_inlineimm:
 ; VI: v_mov_b32_e32 [[K:v[0-9]+]], 0x230000
 ; GCN: {{flat|global}}_load_dword [[VEC:v[0-9]+]]
 ; CI: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VEC]]
 ; GFX9: v_and_b32_e32 [[ELT0:v[0-9]+]], 0xffff, [[VEC]]
 ; CI: v_or_b32_e32 [[RES:v[0-9]+]], 0x230000, [[ELT0]]
-; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[VEC]], [[K]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI: v_or_b32_sdwa [[RES:v[0-9]+]], [[K]], [[VEC]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
 ; GFX9: v_lshl_or_b32 [[RES:v[0-9]+]], 35, 16, [[ELT0]]
 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RES]]
 define amdgpu_kernel void @v_insertelement_v2f16_1_inlineimm(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %out, i64 %tid.ext
   %vec = load <2 x half>, <2 x half> addrspace(1)* %in.gep
define amdgpu_kernel void @v_insertelement_v2i16_dynamic_sgpr(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, i32 %idx) #0 {
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 999, i32 %idx
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_insertelement_v2i16_dynamic_vgpr:
 ; GFX89-DAG: s_mov_b32 [[MASKK:s[0-9]+]], 0xffff{{$}}
-; GCN-DAG: s_movk_i32 [[K:s[0-9]+]], 0x3e7
+; GFX89-DAG: s_movk_i32 [[K:s[0-9]+]], 0x3e7
+; CI-DAG: s_movk_i32 [[K:s[0-9]+]], 0x3e7
 
 ; GCN: {{flat|global}}_load_dword [[IDX:v[0-9]+]]
 ; GCN: {{flat|global}}_load_dword [[VEC:v[0-9]+]]
 
 ; GFX89-DAG: v_lshlrev_b32_e32 [[SCALED_IDX:v[0-9]+]], 4, [[IDX]]
 ; GFX89-DAG: v_lshlrev_b32_e64 [[MASK:v[0-9]+]], [[SCALED_IDX]], [[MASKK]]
 
 ; CI-DAG: v_lshlrev_b32_e32 [[SCALED_IDX:v[0-9]+]], 4, [[IDX]]
define amdgpu_kernel void @v_insertelement_v2i16_dynamic_vgpr(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, i32 addrspace(1)* %idx.ptr) #0 {
   %vec = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
   %vecins = insertelement <2 x i16> %vec, i16 999, i32 %idx
   store <2 x i16> %vecins, <2 x i16> addrspace(1)* %out.gep
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_insertelement_v2f16_dynamic_vgpr:
 ; GFX89-DAG: s_mov_b32 [[MASKK:s[0-9]+]], 0xffff{{$}}
-; GCN-DAG: s_movk_i32 [[K:s[0-9]+]], 0x1234
+; GFX89-DAG: s_movk_i32 [[K:s[0-9]+]], 0x1234
+; CI-DAG: s_movk_i32  [[K:s[0-9]+]], 0x1234
 
 ; GCN: {{flat|global}}_load_dword [[IDX:v[0-9]+]]
 ; GCN: {{flat|global}}_load_dword [[VEC:v[0-9]+]]
 
 ; GFX89-DAG: v_lshlrev_b32_e32 [[SCALED_IDX:v[0-9]+]], 4, [[IDX]]
 ; GFX89-DAG: v_lshlrev_b32_e64 [[MASK:v[0-9]+]], [[SCALED_IDX]], [[MASKK]]
 
 ; CI-DAG: v_lshlrev_b32_e32 [[SCALED_IDX:v[0-9]+]], 4, [[IDX]]

 ; GCN-DAG: s_load_dword [[VAL:s[0-9]+]]
 ; GCN-DAG: {{flat|global}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
 
 ; GFX9: v_and_b32_e32 [[AND:v[0-9]+]], 0xffff, v[[LO]]
 ; GFX9: v_lshl_or_b32 v[[INS_HALF:[0-9]+]], [[VAL]], 16, [[AND]]
 
 ; VI: s_lshl_b32 [[VAL_HI:s[0-9]+]], [[VAL]], 16
 ; VI-DAG: v_mov_b32_e32 [[COPY_VAL:v[0-9]+]], [[VAL_HI]]
-; VI: v_or_b32_sdwa v[[INS_HALF:[0-9]+]], v[[LO]], [[COPY_VAL]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI: v_or_b32_sdwa v[[INS_HALF:[0-9]+]], [[COPY_VAL]], v[[LO]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
 
 ; CI: s_lshl_b32 [[VAL_HI:s[0-9]+]], [[VAL]], 16
 ; CI: v_and_b32_e32 [[AND:v[0-9]+]], 0xffff, v[[LO]]
 ; CI: v_or_b32_e32 v[[INS_HALF:[0-9]+]], [[VAL_HI]], [[AND]]
 
 ; GCN: {{flat|global}}_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[INS_HALF]]:[[HI]]{{\]}}
 define amdgpu_kernel void @v_insertelement_v4f16_1(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %in, i32 %val) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1

 ; GCN-DAG: s_load_dword [[VAL:s[0-9]+]]
 ; GCN-DAG: {{flat|global}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
 
 ; GFX9: v_and_b32_e32 [[AND:v[0-9]+]], 0xffff, v[[HI]]
 ; GFX9: v_lshl_or_b32 v[[INS_HI:[0-9]+]], [[VAL]], 16, [[AND]]
 
 ; VI: s_lshl_b32 [[VAL_HI:s[0-9]+]], [[VAL]], 16
 ; VI-DAG: v_mov_b32_e32 [[COPY_VAL:v[0-9]+]], [[VAL_HI]]
-; VI: v_or_b32_sdwa v[[INS_HI:[0-9]+]], v[[HI]], [[COPY_VAL]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI: v_or_b32_sdwa v[[INS_HI:[0-9]+]], [[COPY_VAL]], v[[HI]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
 
 ; CI: s_lshl_b32 [[VAL_HI:s[0-9]+]], [[VAL]], 16
 ; CI: v_and_b32_e32 [[AND:v[0-9]+]], 0xffff, v[[HI]]
 ; CI: v_or_b32_e32 v[[INS_HI:[0-9]+]], [[VAL_HI]], [[AND]]
 
 ; GCN: {{flat|global}}_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO]]:[[INS_HI]]{{\]}}
 define amdgpu_kernel void @v_insertelement_v4f16_3(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %in, i32 %val) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1

test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll

 ; VI: v_mov_b32_dpp v0, v1 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1 bound_ctrl:0 ; encoding: [0xfa,0x02,0x00,0x7e,0x01,0x01,0x08,0x11]
 define amdgpu_kernel void @dpp_test(i32 addrspace(1)* %out, i32 %in1, i32 %in2) {
   %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 1) #0
   store i32 %tmp0, i32 addrspace(1)* %out
   ret void
 }
 
 ; VI-LABEL: {{^}}dpp_test1:
-; VI-OPT: v_add_u32_e32 [[REG:v[0-9]+]], vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; VI: v_add_u32_e32 [[REG:v[0-9]+]], vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; VI-NOOPT: v_mov_b32_e32 v{{[0-9]+}}, 0
-; VI-NOOPT: v_mov_b32_e32 [[REG:v[0-9]+]], v{{[0-9]+}}
+; VI-NOOPT0: v_mov_b32_e32 [[REG:v[0-9]+]], v{{[0-9]+}}
 ; VI-NEXT: s_nop 0
 ; VI-NEXT: s_nop 0
 ; VI-NEXT: v_mov_b32_dpp v2, [[REG]] quad_perm:[1,0,3,2] row_mask:0xf bank_mask:0xf
 @0 = internal unnamed_addr addrspace(3) global [448 x i32] undef, align 4
 define weak_odr amdgpu_kernel void @dpp_test1(i32* %arg) local_unnamed_addr {
 bb:
   %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
   %tmp1 = zext i32 %tmp to i64

test/CodeGen/AMDGPU/shift-i128.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
 
 define i128 @v_shl_i128_vv(i128 %lhs, i128 %rhs) {
 ; GCN-LABEL: v_shl_i128_vv:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    v_lshl_b64 v[5:6], v[0:1], v4
-; GCN-NEXT:    v_lshl_b64 v[7:8], v[2:3], v4
-; GCN-NEXT:    v_sub_i32_e32 v9, vcc, 64, v4
-; GCN-NEXT:    v_subrev_i32_e32 v11, vcc, 64, v4
-; GCN-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v4
-; GCN-NEXT:    v_cndmask_b32_e32 v6, 0, v6, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v5, 0, v5, vcc
-; GCN-NEXT:    v_lshr_b64 v[9:10], v[0:1], v9
-; GCN-NEXT:    v_lshl_b64 v[0:1], v[0:1], v11
-; GCN-NEXT:    v_or_b32_e32 v7, v7, v9
-; GCN-NEXT:    v_or_b32_e32 v8, v8, v10
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v1, v3, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, v5
-; GCN-NEXT:    v_mov_b32_e32 v1, v6
-; GCN-NEXT:    s_setpc_b64 s[30:31]
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_lshl_b64 v[5:6], v[2:3], v4
+; GCN-NEXT: v_sub_i32_e32 v9, vcc, 64, v4
+; GCN-NEXT: v_subrev_i32_e32 v11, vcc, 64, v4
+; GCN-NEXT: v_lshl_b64 v[7:8], v[0:1], v4
+; GCN-NEXT: v_lshr_b64 v[9:10], v[0:1], v9
+; GCN-NEXT: v_lshl_b64 v[0:1], v[0:1], v11
+; GCN-NEXT: v_or_b32_e32 v6, v6, v10
+; GCN-NEXT: v_or_b32_e32 v5, v5, v9
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc, 64, v4
+; GCN-NEXT: v_cndmask_b32_e32 v6, v1, v6, vcc
+; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v8, vcc
+; GCN-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v3, v6, v3, s[6:7]
+; GCN-NEXT: v_cndmask_b32_e64 v2, v0, v2, s[6:7]
+; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v7, vcc
+; GCN-NEXT: s_setpc_b64 s[30:31]
   %shl = shl i128 %lhs, %rhs
   ret i128 %shl
 }
 
 define i128 @v_lshr_i128_vv(i128 %lhs, i128 %rhs) {
 ; GCN-LABEL: v_lshr_i128_vv:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    v_lshr_b64 v[5:6], v[2:3], v4
-; GCN-NEXT:    v_lshr_b64 v[7:8], v[0:1], v4
-; GCN-NEXT:    v_sub_i32_e32 v9, vcc, 64, v4
-; GCN-NEXT:    v_subrev_i32_e32 v11, vcc, 64, v4
-; GCN-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v4
-; GCN-NEXT:    v_cndmask_b32_e32 v6, 0, v6, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v5, 0, v5, vcc
-; GCN-NEXT:    v_lshl_b64 v[9:10], v[2:3], v9
-; GCN-NEXT:    v_lshr_b64 v[2:3], v[2:3], v11
-; GCN-NEXT:    v_or_b32_e32 v7, v7, v9
-; GCN-NEXT:    v_or_b32_e32 v8, v8, v10
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v7, vcc
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v2, v5
-; GCN-NEXT:    v_mov_b32_e32 v3, v6
-; GCN-NEXT:    s_setpc_b64 s[30:31]
+
+; GCN-NEXT: v_lshr_b64 v[5:6], v[0:1], v4
+; GCN-NEXT: v_sub_i32_e32 v9, vcc, 64, v4
+; GCN-NEXT: v_subrev_i32_e32 v11, vcc, 64, v4
+; GCN-NEXT: v_lshr_b64 v[7:8], v[2:3], v4
+; GCN-NEXT: v_lshl_b64 v[9:10], v[2:3], v9
+; GCN-NEXT: v_lshr_b64 v[2:3], v[2:3], v11
+; GCN-NEXT: v_or_b32_e32 v6, v6, v10
+; GCN-NEXT: v_or_b32_e32 v5, v5, v9
+; GCN-NEXT: v_cmp_gt_u32_e32 vcc, 64, v4
+; GCN-NEXT: v_cndmask_b32_e32 v6, v3, v6, vcc
+; GCN-NEXT: v_cndmask_b32_e32 v2, v2, v5, vcc
+; GCN-NEXT: v_cndmask_b32_e32 v3, 0, v8, vcc
+; GCN-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v4
+; GCN-NEXT: v_cndmask_b32_e64 v1, v6, v1, s[6:7]
+; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[6:7]
+; GCN-NEXT: v_cndmask_b32_e32 v2, 0, v7, vcc
+; GCN-NEXT: s_setpc_b64 s[30:31]
   %shl = lshr i128 %lhs, %rhs
   ret i128 %shl
 }
 
 define i128 @v_ashr_i128_vv(i128 %lhs, i128 %rhs) {
 ; GCN-LABEL: v_ashr_i128_vv:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_ashrrev_i32_e32 v9, 31, v3
 ; GCN-NEXT:    v_ashr_i64 v[5:6], v[2:3], v4
 ; GCN-NEXT:    v_lshr_b64 v[7:8], v[0:1], v4
 ; GCN-NEXT:    v_sub_i32_e32 v10, vcc, 64, v4
 ; GCN-NEXT:    v_subrev_i32_e32 v11, vcc, 64, v4
 ; GCN-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v4
 ; GCN-NEXT:    v_cndmask_b32_e32 v6, v9, v6, vcc
 ; GCN-NEXT:    v_cndmask_b32_e32 v5, v9, v5, vcc
 ; GCN-NEXT:    v_lshl_b64 v[9:10], v[2:3], v10
 ; GCN-NEXT:    v_ashr_i64 v[2:3], v[2:3], v11
-; GCN-NEXT:    v_or_b32_e32 v7, v7, v9
 ; GCN-NEXT:    v_or_b32_e32 v8, v8, v10
+; GCN-NEXT:    v_or_b32_e32 v7, v7, v9
 ; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v7, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
 ; GCN-NEXT:    v_mov_b32_e32 v2, v5
 ; GCN-NEXT:    v_mov_b32_e32 v3, v6
 ; GCN-NEXT:    s_setpc_b64 s[30:31]

 
 define <2 x i128> @v_shl_v2i128_vv(<2 x i128> %lhs, <2 x i128> %rhs) {
 ; GCN-LABEL: v_shl_v2i128_vv:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_lshl_b64 v[16:17], v[2:3], v8
 ; GCN-NEXT:    v_sub_i32_e32 v18, vcc, 64, v8
 ; GCN-NEXT:    v_lshr_b64 v[18:19], v[0:1], v18
-; GCN-NEXT:    v_or_b32_e32 v20, v16, v18
-; GCN-NEXT:    v_or_b32_e32 v21, v17, v19
+; GCN-NEXT:    v_or_b32_e32 v20, v17, v19
+; GCN-NEXT:    v_or_b32_e32 v21, v16, v18
 ; GCN-NEXT:    v_sub_i32_e32 v16, vcc, 64, v12
 ; GCN-NEXT:    v_lshr_b64 v[16:17], v[4:5], v16
 ; GCN-NEXT:    v_lshl_b64 v[18:19], v[6:7], v12
-; GCN-NEXT:    v_or_b32_e32 v16, v18, v16
 ; GCN-NEXT:    v_or_b32_e32 v17, v19, v17
+; GCN-NEXT:    v_or_b32_e32 v16, v18, v16
 ; GCN-NEXT:    v_cmp_eq_u64_e64 s[6:7], 0, v[10:11]
-; GCN-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GCN-NEXT:    v_or_b32_e32 v11, v9, v11
+; GCN-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GCN-NEXT:    v_cmp_eq_u64_e64 s[8:9], 0, v[14:15]
-; GCN-NEXT:    v_or_b32_e32 v14, v12, v14
 ; GCN-NEXT:    v_or_b32_e32 v15, v13, v15
+; GCN-NEXT:    v_or_b32_e32 v14, v12, v14
 ; GCN-NEXT:    v_cmp_gt_u64_e64 s[10:11], 64, v[8:9]
 ; GCN-NEXT:    v_subrev_i32_e32 v18, vcc, 64, v8
 ; GCN-NEXT:    v_lshl_b64 v[8:9], v[0:1], v8
 ; GCN-NEXT:    v_lshl_b64 v[0:1], v[0:1], v18
 ; GCN-NEXT:    v_cmp_gt_u64_e64 s[12:13], 64, v[12:13]
 ; GCN-NEXT:    v_subrev_i32_e32 v18, vcc, 64, v12
 ; GCN-NEXT:    v_lshl_b64 v[12:13], v[4:5], v12
 ; GCN-NEXT:    v_lshl_b64 v[4:5], v[4:5], v18
 ; GCN-NEXT:    s_and_b64 vcc, s[6:7], s[10:11]
-; GCN-NEXT:    v_cndmask_b32_e32 v18, v1, v21, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v19, v0, v20, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v18, v1, v20, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v19, v0, v21, vcc
 ; GCN-NEXT:    s_and_b64 s[6:7], s[8:9], s[12:13]
 ; GCN-NEXT:    v_cndmask_b32_e64 v17, v5, v17, s[6:7]
 ; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v16, s[6:7]
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v9, vcc
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v8, vcc
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, v13, s[6:7]
 ; GCN-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[10:11]
 ; GCN-NEXT:    v_cndmask_b32_e32 v3, v18, v3, vcc

 
 define <2 x i128> @v_lshr_v2i128_vv(<2 x i128> %lhs, <2 x i128> %rhs) {
 ; GCN-LABEL: v_lshr_v2i128_vv:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_lshr_b64 v[16:17], v[0:1], v8
 ; GCN-NEXT:    v_sub_i32_e32 v18, vcc, 64, v8
 ; GCN-NEXT:    v_lshl_b64 v[18:19], v[2:3], v18
-; GCN-NEXT:    v_or_b32_e32 v20, v16, v18
-; GCN-NEXT:    v_or_b32_e32 v21, v17, v19
+; GCN-NEXT:    v_or_b32_e32 v20, v17, v19
+; GCN-NEXT:    v_or_b32_e32 v21, v16, v18
 ; GCN-NEXT:    v_sub_i32_e32 v16, vcc, 64, v12
 ; GCN-NEXT:    v_lshl_b64 v[16:17], v[6:7], v16
 ; GCN-NEXT:    v_lshr_b64 v[18:19], v[4:5], v12
-; GCN-NEXT:    v_or_b32_e32 v16, v18, v16
 ; GCN-NEXT:    v_or_b32_e32 v17, v19, v17
+; GCN-NEXT:    v_or_b32_e32 v16, v18, v16
 ; GCN-NEXT:    v_cmp_eq_u64_e64 s[6:7], 0, v[10:11]
-; GCN-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GCN-NEXT:    v_or_b32_e32 v11, v9, v11
+; GCN-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GCN-NEXT:    v_cmp_eq_u64_e64 s[8:9], 0, v[14:15]
-; GCN-NEXT:    v_or_b32_e32 v14, v12, v14
 ; GCN-NEXT:    v_or_b32_e32 v15, v13, v15
+; GCN-NEXT:    v_or_b32_e32 v14, v12, v14
 ; GCN-NEXT:    v_cmp_gt_u64_e64 s[10:11], 64, v[8:9]
 ; GCN-NEXT:    v_subrev_i32_e32 v18, vcc, 64, v8
 ; GCN-NEXT:    v_lshr_b64 v[8:9], v[2:3], v8
 ; GCN-NEXT:    v_lshr_b64 v[2:3], v[2:3], v18
 ; GCN-NEXT:    v_cmp_gt_u64_e64 s[12:13], 64, v[12:13]
 ; GCN-NEXT:    v_subrev_i32_e32 v18, vcc, 64, v12
 ; GCN-NEXT:    v_lshr_b64 v[12:13], v[6:7], v12
 ; GCN-NEXT:    v_lshr_b64 v[6:7], v[6:7], v18
 ; GCN-NEXT:    s_and_b64 vcc, s[6:7], s[10:11]
-; GCN-NEXT:    v_cndmask_b32_e32 v18, v3, v21, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v19, v2, v20, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v18, v3, v20, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v19, v2, v21, vcc
 ; GCN-NEXT:    s_and_b64 s[6:7], s[8:9], s[12:13]
 ; GCN-NEXT:    v_cndmask_b32_e64 v17, v7, v17, s[6:7]
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, v6, v16, s[6:7]
 ; GCN-NEXT:    v_cndmask_b32_e32 v3, 0, v9, vcc
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, 0, v8, vcc
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, v13, s[6:7]
 ; GCN-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[10:11]
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v18, v1, vcc

 
 define <2 x i128> @v_ashr_v2i128_vv(<2 x i128> %lhs, <2 x i128> %rhs) {
 ; GCN-LABEL: v_ashr_v2i128_vv:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_lshr_b64 v[16:17], v[0:1], v8
 ; GCN-NEXT:    v_sub_i32_e32 v18, vcc, 64, v8
 ; GCN-NEXT:    v_lshl_b64 v[18:19], v[2:3], v18
-; GCN-NEXT:    v_or_b32_e32 v20, v16, v18
-; GCN-NEXT:    v_or_b32_e32 v21, v17, v19
+; GCN-NEXT:    v_or_b32_e32 v20, v17, v19
+; GCN-NEXT:    v_or_b32_e32 v21, v16, v18
 ; GCN-NEXT:    v_sub_i32_e32 v16, vcc, 64, v12
 ; GCN-NEXT:    v_lshl_b64 v[16:17], v[6:7], v16
 ; GCN-NEXT:    v_lshr_b64 v[18:19], v[4:5], v12
-; GCN-NEXT:    v_or_b32_e32 v18, v18, v16
 ; GCN-NEXT:    v_or_b32_e32 v19, v19, v17
+; GCN-NEXT:    v_or_b32_e32 v18, v18, v16
 ; GCN-NEXT:    v_cmp_eq_u64_e64 s[6:7], 0, v[10:11]
-; GCN-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GCN-NEXT:    v_or_b32_e32 v11, v9, v11
+; GCN-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GCN-NEXT:    v_cmp_eq_u64_e64 s[8:9], 0, v[14:15]
-; GCN-NEXT:    v_or_b32_e32 v14, v12, v14
 ; GCN-NEXT:    v_or_b32_e32 v15, v13, v15
+; GCN-NEXT:    v_or_b32_e32 v14, v12, v14
 ; GCN-NEXT:    v_cmp_gt_u64_e64 s[10:11], 64, v[8:9]
 ; GCN-NEXT:    v_subrev_i32_e32 v9, vcc, 64, v8
 ; GCN-NEXT:    v_ashr_i64 v[16:17], v[2:3], v9
 ; GCN-NEXT:    s_and_b64 s[6:7], s[6:7], s[10:11]
-; GCN-NEXT:    v_cndmask_b32_e64 v17, v17, v21, s[6:7]
-; GCN-NEXT:    v_cndmask_b32_e64 v16, v16, v20, s[6:7]
+; GCN-NEXT:    v_cndmask_b32_e64 v17, v17, v20, s[6:7]
+; GCN-NEXT:    v_cndmask_b32_e64 v16, v16, v21, s[6:7]
 ; GCN-NEXT:    v_cmp_gt_u64_e64 s[10:11], 64, v[12:13]
 ; GCN-NEXT:    v_ashr_i64 v[8:9], v[2:3], v8
 ; GCN-NEXT:    v_ashrrev_i32_e32 v20, 31, v3
 ; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, 64, v12
 ; GCN-NEXT:    v_ashr_i64 v[12:13], v[6:7], v12
 ; GCN-NEXT:    v_ashrrev_i32_e32 v21, 31, v7
 ; GCN-NEXT:    v_ashr_i64 v[2:3], v[6:7], v2
 ; GCN-NEXT:    s_and_b64 vcc, s[8:9], s[10:11]

test/CodeGen/AMDGPU/shl.v2i16.ll

 ; VI: v_lshlrev_b16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 ; VI: v_lshlrev_b16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
 ; VI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 
 ; CI: s_mov_b32 [[MASK:s[0-9]+]], 0xffff{{$}}
 ; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[LHS]]
 ; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
 ; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
-; CI: v_lshl_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 ; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
 ; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
 ; CI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 define amdgpu_kernel void @v_shl_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %tid.ext = sext i32 %tid to i64
   %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
   %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext

test/CodeGen/AMDGPU/sibling-call.ll

 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -mattr=-flat-for-global -enable-ipra=0 -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,CIVI,MESA %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -enable-ipra=0 -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI,CIVI,MESA %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=-flat-for-global -enable-ipra=0 -amdgpu-sroa=0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,MESA %s
 target datalayout = "A5"
 
 ; FIXME: Why is this commuted only sometimes?
 ; GCN-LABEL: {{^}}i32_fastcc_i32_i32:
 ; GCN: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v1, v0
+; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v0, v1
 ; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
 ; GCN-NEXT: s_setpc_b64
 define fastcc i32 @i32_fastcc_i32_i32(i32 %arg0, i32 %arg1) #1 {
   %add0 = add i32 %arg0, %arg1
   ret i32 %add0
 }
 
 ; GCN-LABEL: {{^}}i32_fastcc_i32_i32_stack_object:
 ; GCN: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v1, v0
+; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v0, v1
 ; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
 ; GCN: s_mov_b32 s5, s32
 ; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s5 offset:24
 ; GCN: s_waitcnt vmcnt(0)
 ; GCN: s_setpc_b64
 ; GCN: ; ScratchSize: 68
 define fastcc i32 @i32_fastcc_i32_i32_stack_object(i32 %arg0, i32 %arg1) #1 {
   %alloca = alloca [16 x i32], align 4, addrspace(5)

 }
 
 ; GCN-LABEL: {{^}}i32_fastcc_i32_byval_i32:
 ; GCN: s_waitcnt
 ; GCN-NEXT: s_mov_b32 s5, s32
 ; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s5 offset:4
 ; GCN-NEXT: s_waitcnt vmcnt(0)
 
-; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v1, v0
+; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v0, v1
 ; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
 
 ; GCN-NEXT: s_setpc_b64 s[30:31]
 define fastcc i32 @i32_fastcc_i32_byval_i32(i32 %arg0, i32 addrspace(5)* byval align 4 %arg1) #1 {
   %arg1.load = load i32, i32 addrspace(5)* %arg1, align 4
   %add0 = add i32 %arg0, %arg1.load
   ret i32 %add0
 }
entry:
   ret i32 %ret
 }
 
 ; GCN-LABEL: {{^}}i32_fastcc_i32_i32_a32i32:
 ; GCN: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-DAG: buffer_load_dword [[LOAD_0:v[0-9]+]], off, s[0:3], s5 offset:4
 ; GCN-DAG: buffer_load_dword [[LOAD_1:v[0-9]+]], off, s[0:3], s5 offset:8
 
-; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v1, v0
-; CIVI: v_add_{{i|u}}32_e32 v0, vcc, [[LOAD_0]], v0
-; CIVI: v_add_{{i|u}}32_e32 v0, vcc, [[LOAD_1]], v0
+; CIVI-NEXT: v_add_{{i|u}}32_e32 v0, vcc, v0, v1
+; CIVI: v_add_{{i|u}}32_e32 v0, vcc, v0, [[LOAD_0]]
+; CIVI: v_add_{{i|u}}32_e32 v0, vcc, v0, [[LOAD_1]]
 
 
 ; GFX9-NEXT: v_add_u32_e32 v0, v0, v1
 ; GFX9: v_add_u32_e32 v0, v0, [[LOAD_0]]
 ; GFX9: v_add_u32_e32 v0, v0, [[LOAD_1]]
 
 ; GCN-NEXT: s_setpc_b64
 define fastcc i32 @i32_fastcc_i32_i32_a32i32(i32 %arg0, i32 %arg1, [32 x i32] %large) #1 {

test/CodeGen/AMDGPU/sub.ll

define amdgpu_kernel void @test_sub_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
   %a = load <4 x i32>, <4 x i32> addrspace(1) * %in
   %b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr
   %result = sub <4 x i32> %a, %b
   store <4 x i32> %result, <4 x i32> addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}test_sub_i16:
-; SI: v_subrev_i32_e32 v{{[0-9]+}}, vcc,
+; SI: v_sub_i32_e32 v{{[0-9]+}}, vcc,
 ; GFX89: v_sub_u16_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
 define amdgpu_kernel void @test_sub_i16(i16 addrspace(1)* %out, i16 addrspace(1)* %in) {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %gep = getelementptr i16, i16 addrspace(1)* %in, i32 %tid
   %b_ptr = getelementptr i16, i16 addrspace(1)* %gep, i32 1
   %a = load volatile i16, i16 addrspace(1)* %gep
   %b = load volatile i16, i16 addrspace(1)* %b_ptr
   %result = sub i16 %a, %b