[GlobalISel][AMDGPU] Lower G_UMULO/G_SMULO

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

602

Would you get better results if you:

1. scalarized vectors first
2. Promoted small scalar types first?

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umulo.mir

4

Do you need the abort=0s?

[pdhaliwal]

Removed global-isel-abort=0

[pdhaliwal]

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

602

In both cases, the MULO is somehow considered legal by compiler. For e.g. even for s32, expansion does not occur, compiler is directly using S/UMULO instructions. I am investigating this.

[pdhaliwal]

Moved ops close to ADDO

[pdhaliwal]

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

602

It was due to missing corresponding definition of widenScalar for this operation. I tried implementing one for UMULO as it was easier. Didn't see any better result.

[foad]

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umulo.mir

61

Why do we get ANYEXT followed by AND with 1? In the scalar s32 case we just get ZEXT which is nicer.

116–119

Is there a good reason why this is using UADDO plus ZEXT plus a second ADD, instead of using UADDE?

219

This expansion does an s32 multiply and an s16 multiply. It would be better to just do one s32 multiply -- you can extract all the information you need from the result of that.

[pdhaliwal]

Hi, apologies for late reply as I got sidetracked to some other work.

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umulo.mir

61

This is coming from widening of BUILD_VECTOR resulting from legalization of ICMP instruction. So, I guess ZEXT is presumed dead once BUILD_VECTOR gets legalized.

116–119

This is because of legalization of UMULH for s64. I am thinking of having different patch for this as it impacts narrowing of UMULH.

219

I was able to get unsigned operation use single s32 multiply. Signed is getting bit tricky.

[pdhaliwal]

• Scalarize the vectors first
• Using widened operation for smaller types

[pdhaliwal]

Ping!

[arsenm]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1829–1830

Why is this using SExtInReg in the signed case, but ZExt in the other? SExtInReg doesn't widen the type

[foad]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1826

Maybe assert that WideTy is at least twice as wide as SrcTy, otherwise the trick we use for calculating overflow below does not work.

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smulo.mir

3

Can you use -check-prefixes=GCN,GFX8 and GCN,GFX9 so that update_mir_test_checks will common up the identical ones?

113

This looks wrong as Matt noted above. Doesn't G_SEXT_INREG require identical source and result types? Would this fail MIR verification?

[pdhaliwal]

Addressed review comments.

[pdhaliwal]

Can you use -check-prefixes=GCN,GFX8 and GCN,GFX9 so that update_mir_test_checks will common up the identical ones?

It does not work. Script warns as WARNING: Ignoring common prefixes: {'GCN'}: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umulo.mir

[arsenm]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1836

Can you copy some of the comments from the DAG version? Stuff like
// Unsigned overflow occurred if the high part is non-zero.
and

// Signed overflow occurred if the high part does not sign extend the low.

3652–3655

Should just directly extract the reg here, there's no reason to refer to the MachineOperand. This is also using value copies of MachineOperand, which are generally not a good idea

3677–3678

You can hide the createGenericVirtual register calls with build Instr like buildInstr(Opc, {ResultTy, OverFlowTy}...

3686–3687

You should preserve the boolean type of the incoming, not hardcode to s1. We also have LLT.changeElementType for this

[pdhaliwal]

Addressed review comments.

[foad]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1826

Did you test with assertions enabled? I think this needs to be >=.

[pdhaliwal]

Fixed the assert.

[pdhaliwal]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1826

Thanks for pointing out. I was not testing with assertions enabled so I missed it.

[arsenm]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1832

Hmm. This is actually going beyond what I expect widenScalar to do. In general widenScalar should try to produce the original opcode. In this case, the DAG does the same overflow opcode in the wider type, and then ors the flag at the end.

1836

The DAG version still has a few more comments ( To determine if the result overflowed in a larger type...)

1842–1845

This is a common enough pattern. The DAG provides a getZExtInReg to help produce masks like this, maybe move this to MIRBuilder?

[pdhaliwal]

Rebase & Review comments

[pdhaliwal]

Ping!

[arsenm]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1854–1855

I don't think the unsigned case is right. The DAG version inserts a shift here, not a mask

[pdhaliwal]

Use SHRL for unsigned case.

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1854–1855

I have changed it use shift instead of masking. Just curious, why was previous logic wrong? I thought zero'ing the upper bits of multiplication result and then comparing it with latter should provide the correct result.

[foad]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1828

If the wide type is at least twice as wide as the original type, then the widened multiply provably will not overflow, so you don't need the final "or".

If we want to support widening to a type that is less than twice as wide as the original type, then remove this assert and only do the final "or" if WideTy.getScalarSizeInBits() >= 2 * SrcBitWidth. But I'm not sure how you would write a test for that code path.

[pdhaliwal]

Removed assert. Now WideTy overflow check is only done when WideTy is not sufficient enough.
Added test case for s24 to verify the logic.

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1828

That "or" exists to match the existing DAG logic. But you are right that the assert makes "or" redundant. I have removed the assert and added tests for s24 to verify the logic.

[foad]

I think this looks good, just some nits inline. If there are any further improvements they can be done as follow ups.

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1854–1855

I'm pretty sure the previous logic was fine too, it's just a different way of checking the upper part is zero.

1867

Needs a comment that the multiply can't possible overflow if the wide type is >= 2 * original width.

It's a shame that you have to duplicate this check from line 1845. Maybe @arsenm knows a cleaner way to write this.

3650

Personally I would be tempted to generalize LegalizerHelper::fewerElementsVectorMultiEltType into a generic function that can handle any operation that works on vector elements independently. But that does not have to be part of this patch.

[arsenm]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1854–1855

Oh yes, I just can't read. Arguably avoiding the shift is better since shifts can be more expensive

[pdhaliwal]

This version of the patch has bit more changes. Please let me know if it still looks good.

• Revert the unsigned case to use Masks.
• Simplified the logic for widenScalar

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

1854–1855

I have reverted to using masks.

[pdhaliwal]

Fix comments.

[foad]

Still looks good to me.

llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

private:
   LegalizeResult
   widenScalarExtract(MachineInstr &MI, unsigned TypeIdx, LLT WideTy);
   LegalizeResult
   widenScalarInsert(MachineInstr &MI, unsigned TypeIdx, LLT WideTy);
   LegalizeResult widenScalarAddSubOverflow(MachineInstr &MI, unsigned TypeIdx,
                                            LLT WideTy);
   LegalizeResult widenScalarAddSubShlSat(MachineInstr &MI, unsigned TypeIdx,
                                          LLT WideTy);
+  LegalizeResult widenScalarMulo(MachineInstr &MI, unsigned TypeIdx,
+                                 LLT WideTy);
 
   /// Helper function to split a wide generic register into bitwise blocks with
   /// the given Type (which implies the number of blocks needed). The generic
   /// registers created are appended to Ops, starting at bit 0 of Reg.
   void extractParts(Register Reg, LLT Ty, int NumParts,
                     SmallVectorImpl<Register> &VRegs);
 
   /// Version which handles irregular splits.
LegalizeResult fewerElementsVectorUnmergeValues(MachineInstr &MI,
                                                   unsigned TypeIdx,
                                                   LLT NarrowTy);
   LegalizeResult fewerElementsVectorMerge(MachineInstr &MI, unsigned TypeIdx,
                                           LLT NarrowTy);
   LegalizeResult fewerElementsVectorExtractInsertVectorElt(MachineInstr &MI,
                                                            unsigned TypeIdx,
                                                            LLT NarrowTy);
 
+  LegalizeResult fewerElementsVectorMulo(MachineInstr &MI, unsigned TypeIdx,
+                                         LLT NarrowTy);
+
   LegalizeResult
   reduceLoadStoreWidth(MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy);
 
   /// Legalize an instruction by reducing the operation width, either by
   /// narrowing the type of the operation or by reducing the number of elements
   /// of a vector.
   /// The used strategy (narrow vs. fewerElements) is decided by \p NarrowTy.
   /// Narrow is used if the scalar type of \p NarrowTy and \p DstTy differ,

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

auto Result = IsSigned ? MIRBuilder.buildAShr(WideTy, WideInst, ShiftK)
                          : MIRBuilder.buildLShr(WideTy, WideInst, ShiftK);
 
   MIRBuilder.buildTrunc(DstReg, Result);
   MI.eraseFromParent();
   return Legalized;
 }
 
 LegalizerHelper::LegalizeResult
+LegalizerHelper::widenScalarMulo(MachineInstr &MI, unsigned TypeIdx,
+                                 LLT WideTy) {
+  if (TypeIdx == 1)
+    return UnableToLegalize;
+
+  bool IsSigned = MI.getOpcode() == TargetOpcode::G_SMULO;
+  Register Result = MI.getOperand(0).getReg();
+  Register Overflow = MI.getOperand(1).getReg();
+  Register LHS = MI.getOperand(2).getReg();
+  Register RHS = MI.getOperand(3).getReg();
+  LLT SrcTy = MRI.getType(LHS);
+  unsigned SrcBitWidth = SrcTy.getScalarSizeInBits();
+  assert(WideTy.getScalarSizeInBits() == 2 * SrcBitWidth);
+
+  unsigned ExtOp = IsSigned ? TargetOpcode::G_SEXT : TargetOpcode::G_ZEXT;
+  auto LeftOperand = MIRBuilder.buildInstr(ExtOp, {WideTy}, {LHS});
+  auto RightOperand = MIRBuilder.buildInstr(ExtOp, {WideTy}, {RHS});
+
+  auto Mul = MIRBuilder.buildMul(WideTy, LeftOperand, RightOperand);
+  MIRBuilder.buildTrunc(Result, Mul);
+
+  if (IsSigned) {
+    // For signed, overflow occurred when the high part does not sign-extend
+    // the low part.
+    auto SExtResult = MIRBuilder.buildSExtInReg(WideTy, Mul, SrcBitWidth);
+    MIRBuilder.buildICmp(CmpInst::ICMP_NE, Overflow, Mul, SExtResult);
+  } else {
+    // Unsigned overflow occurred if the high part is non-zero
+    auto Mask = MIRBuilder.buildConstant(
+        WideTy,
+        APInt::getLowBitsSet(WideTy.getScalarSizeInBits(), SrcBitWidth));
+    auto And = MIRBuilder.buildAnd(WideTy, Mul, Mask);
+    MIRBuilder.buildICmp(CmpInst::ICMP_NE, Overflow, Mul, And);
+  }
+
+  MI.eraseFromParent();
+  return Legalized;
+}
+
+LegalizerHelper::LegalizeResult
 LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
   switch (MI.getOpcode()) {
   default:
     return UnableToLegalize;
   case TargetOpcode::G_EXTRACT:
     return widenScalarExtract(MI, TypeIdx, WideTy);
   case TargetOpcode::G_INSERT:
     return widenScalarInsert(MI, TypeIdx, WideTy);
   case TargetOpcode::G_MERGE_VALUES:
     return widenScalarMergeValues(MI, TypeIdx, WideTy);
   case TargetOpcode::G_UNMERGE_VALUES:
     return widenScalarUnmergeValues(MI, TypeIdx, WideTy);
   case TargetOpcode::G_SADDO:
   case TargetOpcode::G_SSUBO:
   case TargetOpcode::G_UADDO:
   case TargetOpcode::G_USUBO:
   case TargetOpcode::G_SADDE:
   case TargetOpcode::G_SSUBE:
   case TargetOpcode::G_UADDE:
   case TargetOpcode::G_USUBE:
     return widenScalarAddSubOverflow(MI, TypeIdx, WideTy);
+  case TargetOpcode::G_UMULO:
+  case TargetOpcode::G_SMULO:
+    return widenScalarMulo(MI, TypeIdx, WideTy);
   case TargetOpcode::G_SADDSAT:
   case TargetOpcode::G_SSUBSAT:
   case TargetOpcode::G_SSHLSAT:
   case TargetOpcode::G_UADDSAT:
   case TargetOpcode::G_USUBSAT:
   case TargetOpcode::G_USHLSAT:
     return widenScalarAddSubShlSat(MI, TypeIdx, WideTy);
   case TargetOpcode::G_CTTZ:
for (int J = 0; J != PartsPerUnmerge; ++J)
       MIB.addDef(MI.getOperand(I * PartsPerUnmerge + J).getReg());
     MIB.addUse(Unmerge.getReg(I));
   }
 
   MI.eraseFromParent();
   return Legalized;
 }
 
+LegalizerHelper::LegalizeResult
+LegalizerHelper::fewerElementsVectorMulo(MachineInstr &MI, unsigned TypeIdx,
+                                         LLT NarrowTy) {
+  Register Result = MI.getOperand(0).getReg();
+  Register Overflow = MI.getOperand(1).getReg();
+  Register LHS = MI.getOperand(2).getReg();
+  Register RHS = MI.getOperand(3).getReg();
+
+  LLT SrcTy = MRI.getType(LHS);
+  if (!SrcTy.isVector())
+    return UnableToLegalize;
+
+  LLT ElementType = SrcTy.getElementType();
+  LLT OverflowElementTy = MRI.getType(Overflow).getElementType();
+  const int NumResult = SrcTy.getNumElements();
+  LLT GCDTy = getGCDType(SrcTy, NarrowTy);
+
+  // Unmerge the operands to smaller parts of GCD type.
+  auto UnmergeLHS = MIRBuilder.buildUnmerge(GCDTy, LHS);
+  auto UnmergeRHS = MIRBuilder.buildUnmerge(GCDTy, RHS);
+
+  const int NumOps = UnmergeLHS->getNumOperands() - 1;
+  const int PartsPerUnmerge = NumResult / NumOps;
+  LLT OverflowTy = LLT::scalarOrVector(PartsPerUnmerge, OverflowElementTy);
+  LLT ResultTy = LLT::scalarOrVector(PartsPerUnmerge, ElementType);
+
+  // Perform the operation over unmerged parts.
+  SmallVector<Register, 8> ResultParts;
+  SmallVector<Register, 8> OverflowParts;
+  for (int I = 0; I != NumOps; ++I) {
+    Register Operand1 = UnmergeLHS->getOperand(I).getReg();
+    Register Operand2 = UnmergeRHS->getOperand(I).getReg();
+    auto PartMul = MIRBuilder.buildInstr(MI.getOpcode(), {ResultTy, OverflowTy},
+                                         {Operand1, Operand2});
+    ResultParts.push_back(PartMul->getOperand(0).getReg());
+    OverflowParts.push_back(PartMul->getOperand(1).getReg());
+  }
+
+  LLT ResultLCMTy = buildLCMMergePieces(SrcTy, NarrowTy, GCDTy, ResultParts);
+  LLT OverflowLCMTy =
+      LLT::scalarOrVector(ResultLCMTy.getNumElements(), OverflowElementTy);
+
+  // Recombine the pieces to the original result and overflow registers.
+  buildWidenedRemergeToDst(Result, ResultLCMTy, ResultParts);
+  buildWidenedRemergeToDst(Overflow, OverflowLCMTy, OverflowParts);
+  MI.eraseFromParent();
+  return Legalized;
+}
+
 // Handle FewerElementsVector a G_BUILD_VECTOR or G_CONCAT_VECTORS that produces
 // a vector
 //
 // Create a G_BUILD_VECTOR or G_CONCAT_VECTORS of NarrowTy pieces, padding with
 // undef as necessary.
 //
 // %3:_(<3 x s16>) = G_BUILD_VECTOR %0, %1, %2
 //   -> <2 x s16>
LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
   case G_FSHL:
   case G_FSHR:
   case G_FREEZE:
   case G_SADDSAT:
   case G_SSUBSAT:
   case G_UADDSAT:
   case G_USUBSAT:
     return reduceOperationWidth(MI, TypeIdx, NarrowTy);
+  case G_UMULO:
+  case G_SMULO:
+    return fewerElementsVectorMulo(MI, TypeIdx, NarrowTy);
   case G_SHL:
   case G_LSHR:
   case G_ASHR:
   case G_SSHLSAT:
   case G_USHLSAT:
   case G_CTLZ:
   case G_CTLZ_ZERO_UNDEF:
   case G_CTTZ:

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

Mulh
       .clampMaxNumElements(0, S8, 2)
       .lowerFor({V2S8});
   }
 
   Mulh
     .scalarize(0)
     .lower();
 
   // Report legal for any types we can handle anywhere. For the cases only legal
   // on the SALU, RegBankSelect will be able to re-legalize.
   getActionDefinitionsBuilder({G_AND, G_OR, G_XOR})
     .legalFor({S32, S1, S64, V2S32, S16, V2S16, V4S16})
     .clampScalar(0, S32, S64)
     .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
     .fewerElementsIf(vectorWiderThan(0, 64), fewerEltsToSize64Vector(0))
     .widenScalarToNextPow2(0)
     .scalarize(0);
getActionDefinitionsBuilder(G_FSHR)
     .lower();
 
   getActionDefinitionsBuilder(G_READCYCLECOUNTER)
     .legalFor({S64});
 
   getActionDefinitionsBuilder(G_FENCE)
     .alwaysLegal();
 
+  getActionDefinitionsBuilder({G_SMULO, G_UMULO})
+      .scalarize(0)
+      .minScalar(0, S32)
+      .lower();
+
   getActionDefinitionsBuilder({
       // TODO: Verify V_BFI_B32 is generated from expanded bit ops
       G_FCOPYSIGN,
 
       G_ATOMIC_CMPXCHG_WITH_SUCCESS,
       G_ATOMICRMW_NAND,
       G_ATOMICRMW_FSUB,
       G_READ_REGISTER,

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smulo.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=GFX8
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx906 -O0 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=GFX9
+
+---
+name: test_smulo_s32
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GFX8-LABEL: name: test_smulo_s32
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[COPY]], [[COPY1]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; GFX8: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[SMULH]](s32), [[ASHR]]
+    ; GFX8: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[MUL]](s32)
+    ; GFX8: $vgpr1 = COPY [[SEXT]](s32)
+    ; GFX9-LABEL: name: test_smulo_s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[COPY]], [[COPY1]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; GFX9: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[SMULH]](s32), [[ASHR]]
+    ; GFX9: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[MUL]](s32)
+    ; GFX9: $vgpr1 = COPY [[SEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32), %3:_(s1) = G_SMULO %0, %1
+    %4:_(s32) = G_SEXT %3
+    $vgpr0 = COPY %2
+    $vgpr1 = COPY %4
+...
+
+---
+name: test_smulo_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; GFX8-LABEL: name: test_smulo_v2s32
+    ; GFX8: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX8: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX8: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX8: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[UV]], [[UV2]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; GFX8: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[SMULH]](s32), [[ASHR]]
+    ; GFX8: [[SMULH1:%[0-9]+]]:_(s32) = G_SMULH [[UV1]], [[UV3]]
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV3]]
+    ; GFX8: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[MUL1]], [[C]](s32)
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[SMULH1]](s32), [[ASHR1]]
+    ; GFX8: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[MUL]](s32), [[MUL1]](s32)
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ANYEXT]](s32)
+    ; GFX8: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 1
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ANYEXT1]](s32)
+    ; GFX8: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 1
+    ; GFX8: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG]](s32), [[SEXT_INREG1]](s32)
+    ; GFX8: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX8: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    ; GFX9-LABEL: name: test_smulo_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[UV]], [[UV2]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
+    ; GFX9: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[SMULH]](s32), [[ASHR]]
+    ; GFX9: [[SMULH1:%[0-9]+]]:_(s32) = G_SMULH [[UV1]], [[UV3]]
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV3]]
+    ; GFX9: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[MUL1]], [[C]](s32)
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[SMULH1]](s32), [[ASHR1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[MUL]](s32), [[MUL1]](s32)
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ANYEXT]], 1
+    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ANYEXT1]], 1
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG]](s32), [[SEXT_INREG1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>), %3:_(<2 x s1>) = G_SMULO %0, %1
+    %4:_(<2 x s32>) = G_SEXT %3
+    $vgpr0_vgpr1 = COPY %2
+    $vgpr2_vgpr3 = COPY %4
+...
+
+---
+name: test_smulo_s16
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GFX8-LABEL: name: test_smulo_s16
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX8: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 16
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
+    ; GFX8: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[SEXT_INREG3]](s32)
+    ; GFX8: $vgpr1 = COPY [[SEXT]](s32)
+    ; GFX9-LABEL: name: test_smulo_s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX9: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 16
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
+    ; GFX9: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[SEXT_INREG3]](s32)
+    ; GFX9: $vgpr1 = COPY [[SEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s16) = G_TRUNC %0
+    %3:_(s16) = G_TRUNC %1
+    %4:_(s16), %6:_(s1) = G_SMULO %2, %3
+    %5:_(s32) = G_SEXT %4
+    %7:_(s32) = G_SEXT %6
+    $vgpr0 = COPY %5
+    $vgpr1 = COPY %7
+...
+
+---
+name: test_smulo_s8
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GFX8-LABEL: name: test_smulo_s8
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX8: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 8
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 8
+    ; GFX8: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[SEXT_INREG3]](s32)
+    ; GFX8: $vgpr1 = COPY [[SEXT]](s32)
+    ; GFX9-LABEL: name: test_smulo_s8
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX9: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 8
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 8
+    ; GFX9: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[SEXT_INREG3]](s32)
+    ; GFX9: $vgpr1 = COPY [[SEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s8) = G_TRUNC %0
+    %3:_(s8) = G_TRUNC %1
+    %4:_(s8), %6:_(s1) = G_SMULO %2, %3
+    %5:_(s32) = G_SEXT %4
+    %7:_(s32) = G_SEXT %6
+    $vgpr0 = COPY %5
+    $vgpr1 = COPY %7
+...
+
+---
+name: test_smulo_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; GFX8-LABEL: name: test_smulo_v2s16
+    ; GFX8: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX8: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX8: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; GFX8: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; GFX8: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX8: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 16
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; GFX8: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
+    ; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; GFX8: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 16
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG3]], [[SEXT_INREG4]]
+    ; GFX8: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL1]], 16
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[SEXT_INREG5]]
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; GFX8: [[COPY6:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C]]
+    ; GFX8: [[COPY7:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C]]
+    ; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
+    ; GFX8: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX8: [[COPY8:%[0-9]+]]:_(s32) = COPY [[ANYEXT]](s32)
+    ; GFX8: [[SEXT_INREG6:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY8]], 1
+    ; GFX8: [[COPY9:%[0-9]+]]:_(s32) = COPY [[ANYEXT1]](s32)
+    ; GFX8: [[SEXT_INREG7:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY9]], 1
+    ; GFX8: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG6]](s32), [[SEXT_INREG7]](s32)
+    ; GFX8: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[BITCAST]](<2 x s16>)
+    ; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
+    ; GFX8: [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
+    ; GFX8: [[SEXT_INREG8:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY10]], 16
+    ; GFX8: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX8: [[SEXT_INREG9:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY11]], 16
+    ; GFX8: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG8]](s32), [[SEXT_INREG9]](s32)
+    ; GFX8: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    ; GFX8: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_smulo_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX9: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 16
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; GFX9: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; GFX9: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 16
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG3]], [[SEXT_INREG4]]
+    ; GFX9: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL1]], 16
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[SEXT_INREG5]]
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: [[SEXT_INREG6:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ANYEXT]], 1
+    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX9: [[SEXT_INREG7:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ANYEXT1]], 1
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG6]](s32), [[SEXT_INREG7]](s32)
+    ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
+    ; GFX9: [[SEXT_INREG8:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY8]], 16
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX9: [[SEXT_INREG9:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY9]], 16
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG8]](s32), [[SEXT_INREG9]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    ; GFX9: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s16>) = G_TRUNC %0
+    %3:_(<2 x s16>) = G_TRUNC %1
+    %4:_(<2 x s16>), %6:_(<2 x s1>) = G_SMULO %2, %3
+    %7:_(<2 x s32>) = G_SEXT %6
+    %5:_(<2 x s32>) = G_SEXT %4
+    $vgpr0_vgpr1 = COPY %5
+    $vgpr2_vgpr3 = COPY %7
+...
+
+
+---
+name: test_smulo_v2s8
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+    ; GFX8-LABEL: name: test_smulo_v2s8
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 8
+    ; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
+    ; GFX8: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 8
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX8: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 8
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX8: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY6]], 8
+    ; GFX8: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
+    ; GFX8: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY7]], 8
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG3]], [[SEXT_INREG4]]
+    ; GFX8: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL1]], 8
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[SEXT_INREG5]]
+    ; GFX8: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[MUL]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[MUL1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; GFX8: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX8: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; GFX8: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; GFX8: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX8: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX8: $vgpr1 = COPY [[ANYEXT1]](s32)
+    ; GFX8: $vgpr2 = COPY [[ANYEXT2]](s32)
+    ; GFX9-LABEL: name: test_smulo_v2s8
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 8
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
+    ; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 8
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX9: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 8
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY6]], 8
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
+    ; GFX9: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY7]], 8
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG3]], [[SEXT_INREG4]]
+    ; GFX9: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL1]], 8
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[SEXT_INREG5]]
+    ; GFX9: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[MUL]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[MUL1]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; GFX9: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9: $vgpr1 = COPY [[ANYEXT1]](s32)
+    ; GFX9: $vgpr2 = COPY [[ANYEXT2]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = COPY $vgpr2
+    %3:_(s32) = COPY $vgpr3
+    %5:_(s8) = G_TRUNC %0
+    %6:_(s8) = G_TRUNC %1
+    %7:_(s8) = G_TRUNC %2
+    %8:_(s8) = G_TRUNC %3
+    %11:_(<2 x s8>) = G_BUILD_VECTOR %5, %6
+    %12:_(<2 x s8>) = G_BUILD_VECTOR %7, %8
+    %13:_(<2 x s8>), %19:_(<2 x s1>) = G_SMULO %11, %12
+    %20:_(<2 x s32>) = G_SEXT %19
+    %14:_(s8), %15:_(s8) = G_UNMERGE_VALUES %13
+    %21:_(s1), %22:_(s1) = G_UNMERGE_VALUES %19
+    %17:_(s16) = G_MERGE_VALUES %14, %15
+    %18:_(s32) = G_ANYEXT %17
+    %23:_(s32) = G_ANYEXT %21
+    %24:_(s32) = G_ANYEXT %22
+    $vgpr0 = COPY %18
+    $vgpr1 = COPY %23
+    $vgpr2 = COPY %24
+...
+
+---
+name: test_smulo_v4s8
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; GFX8-LABEL: name: test_smulo_v4s8
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
+    ; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX8: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
+    ; GFX8: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX8: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
+    ; GFX8: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C]](s32)
+    ; GFX8: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C1]](s32)
+    ; GFX8: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C2]](s32)
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX8: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 8
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX8: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 8
+    ; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
+    ; GFX8: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 8
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG3]], [[SEXT_INREG4]]
+    ; GFX8: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL1]], 8
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[SEXT_INREG5]]
+    ; GFX8: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
+    ; GFX8: [[SEXT_INREG6:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY6]], 8
+    ; GFX8: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
+    ; GFX8: [[SEXT_INREG7:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY7]], 8
+    ; GFX8: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG6]], [[SEXT_INREG7]]
+    ; GFX8: [[SEXT_INREG8:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL2]], 8
+    ; GFX8: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL2]](s32), [[SEXT_INREG8]]
+    ; GFX8: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
+    ; GFX8: [[SEXT_INREG9:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY8]], 8
+    ; GFX8: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
+    ; GFX8: [[SEXT_INREG10:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY9]], 8
+    ; GFX8: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG9]], [[SEXT_INREG10]]
+    ; GFX8: [[SEXT_INREG11:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL3]], 8
+    ; GFX8: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL3]](s32), [[SEXT_INREG11]]
+    ; GFX8: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX8: [[COPY10:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
+    ; GFX8: [[COPY11:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
+    ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
+    ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
+    ; GFX8: [[COPY12:%[0-9]+]]:_(s32) = COPY [[MUL2]](s32)
+    ; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
+    ; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C1]](s32)
+    ; GFX8: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
+    ; GFX8: [[COPY13:%[0-9]+]]:_(s32) = COPY [[MUL3]](s32)
+    ; GFX8: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
+    ; GFX8: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
+    ; GFX8: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[OR2]](s32)
+    ; GFX8: $vgpr1 = COPY [[ANYEXT]](s32)
+    ; GFX9-LABEL: name: test_smulo_v4s8
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
+    ; GFX9: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C]](s32)
+    ; GFX9: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C1]](s32)
+    ; GFX9: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C2]](s32)
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; GFX9: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL]], 8
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[SEXT_INREG2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX9: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 8
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
+    ; GFX9: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 8
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG3]], [[SEXT_INREG4]]
+    ; GFX9: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL1]], 8
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[SEXT_INREG5]]
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
+    ; GFX9: [[SEXT_INREG6:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY6]], 8
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
+    ; GFX9: [[SEXT_INREG7:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY7]], 8
+    ; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG6]], [[SEXT_INREG7]]
+    ; GFX9: [[SEXT_INREG8:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL2]], 8
+    ; GFX9: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL2]](s32), [[SEXT_INREG8]]
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
+    ; GFX9: [[SEXT_INREG9:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY8]], 8
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
+    ; GFX9: [[SEXT_INREG10:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY9]], 8
+    ; GFX9: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG9]], [[SEXT_INREG10]]
+    ; GFX9: [[SEXT_INREG11:%[0-9]+]]:_(s32) = G_SEXT_INREG [[MUL3]], 8
+    ; GFX9: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL3]](s32), [[SEXT_INREG11]]
+    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
+    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
+    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[MUL2]](s32)
+    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
+    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C1]](s32)
+    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
+    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[MUL3]](s32)
+    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
+    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
+    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[OR2]](s32)
+    ; GFX9: $vgpr1 = COPY [[ANYEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8) = G_UNMERGE_VALUES %0
+    %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8) = G_UNMERGE_VALUES %1
+    %10:_(<4 x s8>) = G_BUILD_VECTOR %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8)
+    %11:_(<4 x s8>) = G_BUILD_VECTOR %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8)
+    %12:_(<4 x s8>), %18:_(<4 x s1>) = G_SMULO %10:_, %11:_
+    %13:_(s8), %14:_(s8), %15:_(s8), %16:_(s8) = G_UNMERGE_VALUES %12:_(<4 x s8>)
+    %19:_(s1), %20:_(s1), %21:_(s1), %22:_(s1) = G_UNMERGE_VALUES %18:_(<4 x s1>)
+    %17:_(s32) = G_MERGE_VALUES %13, %14, %15, %16
+    %23:_(s32) = G_ANYEXT %19
+    $vgpr0 = COPY %17
+    $vgpr1 = COPY %23
+...

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umulo.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=GFX8
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx906 -O0 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=GFX9
+
+---
+name: test_umulo_s32
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GFX8-LABEL: name: test_umulo_s32
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[COPY]], [[COPY1]]
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
+    ; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[MUL]](s32)
+    ; GFX8: $vgpr1 = COPY [[ZEXT]](s32)
+    ; GFX9-LABEL: name: test_umulo_s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[COPY]], [[COPY1]]
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY]], [[COPY1]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
+    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[MUL]](s32)
+    ; GFX9: $vgpr1 = COPY [[ZEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32), %3:_(s1) = G_UMULO %0, %1
+    %4:_(s32) = G_ZEXT %3
+    $vgpr0 = COPY %2
+    $vgpr1 = COPY %4
+...
+
+---
+name: test_umulo_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; GFX8-LABEL: name: test_umulo_v2s32
+    ; GFX8: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX8: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX8: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX8: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
+    ; GFX8: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV1]], [[UV3]]
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV3]]
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UMULH1]](s32), [[C]]
+    ; GFX8: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[MUL]](s32), [[MUL1]](s32)
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ANYEXT]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ANYEXT1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
+    ; GFX8: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[AND]](s32), [[AND1]](s32)
+    ; GFX8: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX8: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    ; GFX9-LABEL: name: test_umulo_v2s32
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UMULH]](s32), [[C]]
+    ; GFX9: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV1]], [[UV3]]
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV3]]
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UMULH1]](s32), [[C]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[MUL]](s32), [[MUL1]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[ANYEXT]], [[C1]]
+    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[ANYEXT1]], [[C1]]
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[AND]](s32), [[AND1]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>), %3:_(<2 x s1>) = G_UMULO %0, %1
+    %4:_(<2 x s32>) = G_ZEXT %3
+    $vgpr0_vgpr1 = COPY %2
+    $vgpr2_vgpr3 = COPY %4
+...
+
+---
+name: test_umulo_s64
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; GFX8-LABEL: name: test_umulo_s64
+    ; GFX8: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
+    ; GFX8: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
+    ; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
+    ; GFX8: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
+    ; GFX8: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[MUL]], [[MUL1]]
+    ; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO1]](s1)
+    ; GFX8: [[UADDO2:%[0-9]+]]:_(s32), [[UADDO3:%[0-9]+]]:_(s1) = G_UADDO [[UADDO]], [[UMULH]]
+    ; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO3]](s1)
+    ; GFX8: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[ZEXT]], [[ZEXT1]]
+    ; GFX8: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV3]]
+    ; GFX8: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV1]], [[UV2]]
+    ; GFX8: [[UMULH2:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV3]]
+    ; GFX8: [[UADDO4:%[0-9]+]]:_(s32), [[UADDO5:%[0-9]+]]:_(s1) = G_UADDO [[MUL2]], [[UMULH1]]
+    ; GFX8: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO5]](s1)
+    ; GFX8: [[UADDO6:%[0-9]+]]:_(s32), [[UADDO7:%[0-9]+]]:_(s1) = G_UADDO [[UADDO4]], [[UMULH2]]
+    ; GFX8: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO7]](s1)
+    ; GFX8: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ZEXT2]], [[ZEXT3]]
+    ; GFX8: [[UADDO8:%[0-9]+]]:_(s32), [[UADDO9:%[0-9]+]]:_(s1) = G_UADDO [[UADDO6]], [[ADD]]
+    ; GFX8: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO9]](s1)
+    ; GFX8: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[ADD1]], [[ZEXT4]]
+    ; GFX8: [[UMULH3:%[0-9]+]]:_(s32) = G_UMULH [[UV1]], [[UV3]]
+    ; GFX8: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[UMULH3]], [[ADD2]]
+    ; GFX8: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO8]](s32), [[ADD3]](s32)
+    ; GFX8: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
+    ; GFX8: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
+    ; GFX8: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
+    ; GFX8: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV6]]
+    ; GFX8: [[MUL4:%[0-9]+]]:_(s32) = G_MUL [[UV5]], [[UV6]]
+    ; GFX8: [[MUL5:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV7]]
+    ; GFX8: [[UMULH4:%[0-9]+]]:_(s32) = G_UMULH [[UV4]], [[UV6]]
+    ; GFX8: [[ADD4:%[0-9]+]]:_(s32) = G_ADD [[MUL4]], [[MUL5]]
+    ; GFX8: [[ADD5:%[0-9]+]]:_(s32) = G_ADD [[ADD4]], [[UMULH4]]
+    ; GFX8: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL3]](s32), [[ADD5]](s32)
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MV]](s64), [[C]]
+    ; GFX8: [[ZEXT5:%[0-9]+]]:_(s64) = G_ZEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0_vgpr1 = COPY [[MV1]](s64)
+    ; GFX8: $vgpr2_vgpr3 = COPY [[ZEXT5]](s64)
+    ; GFX9-LABEL: name: test_umulo_s64
+    ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
+    ; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
+    ; GFX9: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[MUL]], [[MUL1]]
+    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO1]](s1)
+    ; GFX9: [[UADDO2:%[0-9]+]]:_(s32), [[UADDO3:%[0-9]+]]:_(s1) = G_UADDO [[UADDO]], [[UMULH]]
+    ; GFX9: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO3]](s1)
+    ; GFX9: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[ZEXT]], [[ZEXT1]]
+    ; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV3]]
+    ; GFX9: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV1]], [[UV2]]
+    ; GFX9: [[UMULH2:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV3]]
+    ; GFX9: [[UADDO4:%[0-9]+]]:_(s32), [[UADDO5:%[0-9]+]]:_(s1) = G_UADDO [[MUL2]], [[UMULH1]]
+    ; GFX9: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO5]](s1)
+    ; GFX9: [[UADDO6:%[0-9]+]]:_(s32), [[UADDO7:%[0-9]+]]:_(s1) = G_UADDO [[UADDO4]], [[UMULH2]]
+    ; GFX9: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO7]](s1)
+    ; GFX9: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ZEXT2]], [[ZEXT3]]
+    ; GFX9: [[UADDO8:%[0-9]+]]:_(s32), [[UADDO9:%[0-9]+]]:_(s1) = G_UADDO [[UADDO6]], [[ADD]]
+    ; GFX9: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UADDO9]](s1)
+    ; GFX9: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[ADD1]], [[ZEXT4]]
+    ; GFX9: [[UMULH3:%[0-9]+]]:_(s32) = G_UMULH [[UV1]], [[UV3]]
+    ; GFX9: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[UMULH3]], [[ADD2]]
+    ; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[UADDO8]](s32), [[ADD3]](s32)
+    ; GFX9: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
+    ; GFX9: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
+    ; GFX9: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
+    ; GFX9: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV6]]
+    ; GFX9: [[MUL4:%[0-9]+]]:_(s32) = G_MUL [[UV5]], [[UV6]]
+    ; GFX9: [[MUL5:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV7]]
+    ; GFX9: [[UMULH4:%[0-9]+]]:_(s32) = G_UMULH [[UV4]], [[UV6]]
+    ; GFX9: [[ADD4:%[0-9]+]]:_(s32) = G_ADD [[MUL4]], [[MUL5]]
+    ; GFX9: [[ADD5:%[0-9]+]]:_(s32) = G_ADD [[ADD4]], [[UMULH4]]
+    ; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL3]](s32), [[ADD5]](s32)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MV]](s64), [[C]]
+    ; GFX9: [[ZEXT5:%[0-9]+]]:_(s64) = G_ZEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[MV1]](s64)
+    ; GFX9: $vgpr2_vgpr3 = COPY [[ZEXT5]](s64)
+    %0:_(s64) = COPY $vgpr0_vgpr1
+    %1:_(s64) = COPY $vgpr2_vgpr3
+    %2:_(s64), %3:_(s1) = G_UMULO %0, %1
+    %4:_(s64) = G_ZEXT %3
+    $vgpr0_vgpr1 = COPY %2
+    $vgpr2_vgpr3 = COPY %4
+...
+
+---
+name: test_umulo_s16
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GFX8-LABEL: name: test_umulo_s16
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[AND3]](s32)
+    ; GFX8: $vgpr1 = COPY [[ZEXT]](s32)
+    ; GFX9-LABEL: name: test_umulo_s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[AND3]](s32)
+    ; GFX9: $vgpr1 = COPY [[ZEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s16) = G_TRUNC %0
+    %3:_(s16) = G_TRUNC %1
+    %4:_(s16), %6:_(s1) = G_UMULO %2, %3
+    %5:_(s32) = G_ZEXT %4
+    %7:_(s32) = G_ZEXT %6
+    $vgpr0 = COPY %5
+    $vgpr1 = COPY %7
+...
+
+---
+name: test_umulo_s8
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GFX8-LABEL: name: test_umulo_s8
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[AND3]](s32)
+    ; GFX8: $vgpr1 = COPY [[ZEXT]](s32)
+    ; GFX9-LABEL: name: test_umulo_s8
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX9: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[AND3]](s32)
+    ; GFX9: $vgpr1 = COPY [[ZEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s8) = G_TRUNC %0
+    %3:_(s8) = G_TRUNC %1
+    %4:_(s8), %6:_(s1) = G_UMULO %2, %3
+    %5:_(s32) = G_ZEXT %4
+    %7:_(s32) = G_ZEXT %6
+    $vgpr0 = COPY %5
+    $vgpr1 = COPY %7
+...
+
+---
+name: test_umulo_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; GFX8-LABEL: name: test_umulo_v2s16
+    ; GFX8: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX8: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX8: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX8: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; GFX8: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; GFX8: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C]]
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[AND3]], [[AND4]]
+    ; GFX8: [[AND5:%[0-9]+]]:_(s32) = G_AND [[MUL1]], [[C]]
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[AND5]]
+    ; GFX8: [[COPY6:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C]]
+    ; GFX8: [[COPY7:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX8: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C]]
+    ; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND7]], [[C1]](s32)
+    ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND6]], [[SHL]]
+    ; GFX8: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX8: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; GFX8: [[COPY8:%[0-9]+]]:_(s32) = COPY [[ANYEXT]](s32)
+    ; GFX8: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C2]]
+    ; GFX8: [[COPY9:%[0-9]+]]:_(s32) = COPY [[ANYEXT1]](s32)
+    ; GFX8: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C2]]
+    ; GFX8: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[AND8]](s32), [[AND9]](s32)
+    ; GFX8: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[BITCAST]](<2 x s16>)
+    ; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
+    ; GFX8: [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
+    ; GFX8: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
+    ; GFX8: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX8: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
+    ; GFX8: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[AND10]](s32), [[AND11]](s32)
+    ; GFX8: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    ; GFX8: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    ; GFX9-LABEL: name: test_umulo_v2s16
+    ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C]]
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[AND3]], [[AND4]]
+    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[MUL1]], [[C]]
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[AND5]]
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY6]](s32), [[COPY7]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[ANYEXT]], [[C1]]
+    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[ANYEXT1]], [[C1]]
+    ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[AND6]](s32), [[AND7]](s32)
+    ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C2]](s32)
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
+    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C]]
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C]]
+    ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[AND8]](s32), [[AND9]](s32)
+    ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR1]](<2 x s32>)
+    ; GFX9: $vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s16>) = G_TRUNC %0
+    %3:_(<2 x s16>) = G_TRUNC %1
+    %4:_(<2 x s16>), %6:_(<2 x s1>) = G_UMULO %2, %3
+    %7:_(<2 x s32>) = G_ZEXT %6
+    %5:_(<2 x s32>) = G_ZEXT %4
+    $vgpr0_vgpr1 = COPY %5
+    $vgpr2_vgpr3 = COPY %7
+...
+
+
+---
+name: test_umulo_v2s8
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+    ; GFX8-LABEL: name: test_umulo_v2s8
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX8: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C]]
+    ; GFX8: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
+    ; GFX8: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C]]
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[AND3]], [[AND4]]
+    ; GFX8: [[AND5:%[0-9]+]]:_(s32) = G_AND [[MUL1]], [[C]]
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[AND5]]
+    ; GFX8: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[MUL]](s32)
+    ; GFX8: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
+    ; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[MUL1]](s32)
+    ; GFX8: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
+    ; GFX8: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX8: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C2]](s16)
+    ; GFX8: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL]]
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; GFX8: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX8: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX8: $vgpr1 = COPY [[ANYEXT1]](s32)
+    ; GFX8: $vgpr2 = COPY [[ANYEXT2]](s32)
+    ; GFX9-LABEL: name: test_umulo_v2s8
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C]]
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
+    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C]]
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[AND3]], [[AND4]]
+    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[MUL1]], [[C]]
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[AND5]]
+    ; GFX9: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[MUL]](s32)
+    ; GFX9: [[AND6:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C1]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[MUL1]](s32)
+    ; GFX9: [[AND7:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C1]]
+    ; GFX9: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND7]], [[C2]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND6]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
+    ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
+    ; GFX9: $vgpr1 = COPY [[ANYEXT1]](s32)
+    ; GFX9: $vgpr2 = COPY [[ANYEXT2]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = COPY $vgpr2
+    %3:_(s32) = COPY $vgpr3
+    %5:_(s8) = G_TRUNC %0
+    %6:_(s8) = G_TRUNC %1
+    %7:_(s8) = G_TRUNC %2
+    %8:_(s8) = G_TRUNC %3
+    %11:_(<2 x s8>) = G_BUILD_VECTOR %5, %6
+    %12:_(<2 x s8>) = G_BUILD_VECTOR %7, %8
+    %13:_(<2 x s8>), %19:_(<2 x s1>) = G_UMULO %11, %12
+    %20:_(<2 x s32>) = G_ZEXT %19
+    %14:_(s8), %15:_(s8) = G_UNMERGE_VALUES %13
+    %21:_(s1), %22:_(s1) = G_UNMERGE_VALUES %19
+    %17:_(s16) = G_MERGE_VALUES %14, %15
+    %18:_(s32) = G_ANYEXT %17
+    %23:_(s32) = G_ANYEXT %21
+    %24:_(s32) = G_ANYEXT %22
+    $vgpr0 = COPY %18
+    $vgpr1 = COPY %23
+    $vgpr2 = COPY %24
+...
+
+---
+name: test_umulo_v4s8
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; GFX8-LABEL: name: test_umulo_v4s8
+    ; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
+    ; GFX8: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX8: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
+    ; GFX8: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX8: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
+    ; GFX8: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C]](s32)
+    ; GFX8: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C1]](s32)
+    ; GFX8: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C2]](s32)
+    ; GFX8: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX8: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
+    ; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX8: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
+    ; GFX8: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX8: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C3]]
+    ; GFX8: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX8: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
+    ; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
+    ; GFX8: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
+    ; GFX8: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[AND3]], [[AND4]]
+    ; GFX8: [[AND5:%[0-9]+]]:_(s32) = G_AND [[MUL1]], [[C3]]
+    ; GFX8: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[AND5]]
+    ; GFX8: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
+    ; GFX8: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
+    ; GFX8: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
+    ; GFX8: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
+    ; GFX8: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[AND6]], [[AND7]]
+    ; GFX8: [[AND8:%[0-9]+]]:_(s32) = G_AND [[MUL2]], [[C3]]
+    ; GFX8: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL2]](s32), [[AND8]]
+    ; GFX8: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
+    ; GFX8: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
+    ; GFX8: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
+    ; GFX8: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
+    ; GFX8: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[AND9]], [[AND10]]
+    ; GFX8: [[AND11:%[0-9]+]]:_(s32) = G_AND [[MUL3]], [[C3]]
+    ; GFX8: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL3]](s32), [[AND11]]
+    ; GFX8: [[COPY10:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX8: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
+    ; GFX8: [[COPY11:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX8: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
+    ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C]](s32)
+    ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL]]
+    ; GFX8: [[COPY12:%[0-9]+]]:_(s32) = COPY [[MUL2]](s32)
+    ; GFX8: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
+    ; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C1]](s32)
+    ; GFX8: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
+    ; GFX8: [[COPY13:%[0-9]+]]:_(s32) = COPY [[MUL3]](s32)
+    ; GFX8: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
+    ; GFX8: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C2]](s32)
+    ; GFX8: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
+    ; GFX8: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX8: $vgpr0 = COPY [[OR2]](s32)
+    ; GFX8: $vgpr1 = COPY [[ANYEXT]](s32)
+    ; GFX9-LABEL: name: test_umulo_v4s8
+    ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
+    ; GFX9: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C]](s32)
+    ; GFX9: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C1]](s32)
+    ; GFX9: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C3]]
+    ; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C3]]
+    ; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[AND1]]
+    ; GFX9: [[AND2:%[0-9]+]]:_(s32) = G_AND [[MUL]], [[C3]]
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL]](s32), [[AND2]]
+    ; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; GFX9: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C3]]
+    ; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
+    ; GFX9: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C3]]
+    ; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[AND3]], [[AND4]]
+    ; GFX9: [[AND5:%[0-9]+]]:_(s32) = G_AND [[MUL1]], [[C3]]
+    ; GFX9: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL1]](s32), [[AND5]]
+    ; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
+    ; GFX9: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C3]]
+    ; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
+    ; GFX9: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C3]]
+    ; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[AND6]], [[AND7]]
+    ; GFX9: [[AND8:%[0-9]+]]:_(s32) = G_AND [[MUL2]], [[C3]]
+    ; GFX9: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL2]](s32), [[AND8]]
+    ; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
+    ; GFX9: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C3]]
+    ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
+    ; GFX9: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C3]]
+    ; GFX9: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[AND9]], [[AND10]]
+    ; GFX9: [[AND11:%[0-9]+]]:_(s32) = G_AND [[MUL3]], [[C3]]
+    ; GFX9: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[MUL3]](s32), [[AND11]]
+    ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
+    ; GFX9: [[AND12:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C3]]
+    ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
+    ; GFX9: [[AND13:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C3]]
+    ; GFX9: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND13]], [[C]](s32)
+    ; GFX9: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND12]], [[SHL]]
+    ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[MUL2]](s32)
+    ; GFX9: [[AND14:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C3]]
+    ; GFX9: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND14]], [[C1]](s32)
+    ; GFX9: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
+    ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[MUL3]](s32)
+    ; GFX9: [[AND15:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C3]]
+    ; GFX9: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND15]], [[C2]](s32)
+    ; GFX9: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
+    ; GFX9: $vgpr0 = COPY [[OR2]](s32)
+    ; GFX9: $vgpr1 = COPY [[ANYEXT]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8) = G_UNMERGE_VALUES %0
+    %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8) = G_UNMERGE_VALUES %1
+    %10:_(<4 x s8>) = G_BUILD_VECTOR %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8)
+    %11:_(<4 x s8>) = G_BUILD_VECTOR %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8)
+    %12:_(<4 x s8>), %18:_(<4 x s1>) = G_UMULO %10:_, %11:_
+    %13:_(s8), %14:_(s8), %15:_(s8), %16:_(s8) = G_UNMERGE_VALUES %12:_(<4 x s8>)
+    %19:_(s1), %20:_(s1), %21:_(s1), %22:_(s1) = G_UNMERGE_VALUES %18:_(<4 x s1>)
+    %17:_(s32) = G_MERGE_VALUES %13, %14, %15, %16
+    %23:_(s32) = G_ANYEXT %19
+    $vgpr0 = COPY %17
+    $vgpr1 = COPY %23
+...