GlobalISel: Implement lower for G_SHUFFLE_VECTOR

[aditya_nandakumar]

LGTM - conditional to the parent patch.

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

3774

Would it be possible to buildInstr in a way that it can be CSE'd (collect the regs in a SmallVector and build in the end)?
While I think the chances of CSEing a G_BUILD_VECTOR is quite low, it probably doesn't hurt to enable it.

[arsenm]

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

3774

It's possible, but I find the intermediate vectors for these sorts of things kind of annoying to use and it seemed more straightforward to add operands directly

[aditya_nandakumar]

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

3774

Even if it's annoying, I'd prefer we try to build instructions in a way that's CSEable and in parallel maybe look at ways to make this less annoying.
Personally I like the intermediate vector + buildVector at once vs buildInstrNoInsert + insertInstr (maybe because I'm used to it).

[arsenm]

Use buildBiuildVector

[arsenm]

r368709

include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

public:
 
   LegalizeResult lowerU64ToF32BitOps(MachineInstr &MI);
   LegalizeResult lowerUITOFP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerSITOFP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerMinMax(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFCopySign(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFMinNumMaxNum(MachineInstr &MI);
   LegalizeResult lowerUnmergeValues(MachineInstr &MI);
+  LegalizeResult lowerShuffleVector(MachineInstr &MI);
 
 private:
   MachineRegisterInfo &MRI;
   const LegalizerInfo &LI;
   /// To keep track of changes made by the LegalizerHelper.
   GISelChangeObserver &Observer;
 };
 

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

case TargetOpcode::G_SEXT_INREG: {
     Register TmpRes = MRI.createGenericVirtualRegister(DstTy);
 
     auto MIBSz = MIRBuilder.buildConstant(DstTy, DstTy.getScalarSizeInBits() - SizeInBits);
     MIRBuilder.buildInstr(TargetOpcode::G_SHL, {TmpRes}, {SrcReg, MIBSz->getOperand(0).getReg()});
     MIRBuilder.buildInstr(TargetOpcode::G_ASHR, {DstReg}, {TmpRes, MIBSz->getOperand(0).getReg()});
     MI.eraseFromParent();
     return Legalized;
   }
+  case G_SHUFFLE_VECTOR:
+    return lowerShuffleVector(MI);
   }
 }
 
 LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorImplicitDef(
     MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy) {
   SmallVector<Register, 2> DstRegs;
 
   unsigned NarrowSize = NarrowTy.getSizeInBits();
if (!DstTy.isVector() && SrcTy.isVector() &&
     }
 
     MI.eraseFromParent();
     return Legalized;
   }
 
   return UnableToLegalize;
 }
+
+LegalizerHelper::LegalizeResult
+LegalizerHelper::lowerShuffleVector(MachineInstr &MI) {
+  Register DstReg = MI.getOperand(0).getReg();
+  Register Src0Reg = MI.getOperand(1).getReg();
+  Register Src1Reg = MI.getOperand(2).getReg();
+  LLT DstTy = MRI.getType(DstReg);
+  LLT EltTy = DstTy.getElementType();
+  int NumElts = DstTy.getNumElements();
+  LLT IdxTy = LLT::scalar(32);
+
+  const Constant *ShufMask = MI.getOperand(3).getShuffleMask();
+
+  SmallVector<int, 32> Mask;
+  ShuffleVectorInst::getShuffleMask(ShufMask, Mask);
+
+  Register Undef;
+  auto MIB = MIRBuilder.buildInstrNoInsert(TargetOpcode::G_BUILD_VECTOR)
+    .addDef(DstReg);
+
+  for (int Idx : Mask) {
+    if (Idx < 0) {
+      if (!Undef.isValid())
+        Undef = MIRBuilder.buildUndef(EltTy).getReg(0);
+      MIB.addUse(Undef);
+      continue;
+    }
+
+    Register SrcVec = Idx < NumElts ? Src0Reg : Src1Reg;
+    int ExtractIdx = Idx < NumElts ? Idx : Idx - NumElts;
+    auto IdxK = MIRBuilder.buildConstant(IdxTy, ExtractIdx);
+    auto Extract = MIRBuilder.buildExtractVectorElement(EltTy, SrcVec, IdxK);
+    MIB.addUse(Extract.getReg(0));
+  }
+
+  MIRBuilder.insertInstr(MIB);
+  MI.eraseFromParent();
+  return Legalized;
+}

lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

getActionDefinitionsBuilder(G_BUILD_VECTOR)
       .clampNumElements(0, V2S64, V8S64)
       .minScalarSameAs(1, 0)
       .legalIf(isRegisterType(0))
       .minScalarOrElt(0, S32);
 
   getActionDefinitionsBuilder(G_CONCAT_VECTORS)
     .legalIf(isRegisterType(0));
 
+  // TODO: Don't fully scalarize v2s16 pieces
+  getActionDefinitionsBuilder(G_SHUFFLE_VECTOR).lower();
+
   // Merge/Unmerge
   for (unsigned Op : {G_MERGE_VALUES, G_UNMERGE_VALUES}) {
     unsigned BigTyIdx = Op == G_MERGE_VALUES ? 0 : 1;
     unsigned LitTyIdx = Op == G_MERGE_VALUES ? 1 : 0;
 
     auto notValidElt = [=](const LegalityQuery &Query, unsigned TypeIdx) {
       const LLT &Ty = Query.Types[TypeIdx];
       if (Ty.isVector()) {

test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.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
+
+---
+name: shufflevector_v2s32_0_1
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v2s32_0_1
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 0
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 32
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[EXTRACT]](s32), [[EXTRACT1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0, 1)
+    $vgpr0_vgpr1 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s32_1_0
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v2s32_1_0
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 32
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 0
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[EXTRACT]](s32), [[EXTRACT1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(1, 0)
+    $vgpr0_vgpr1 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s32_0_0
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v2s32_0_0
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 0
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 0
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[EXTRACT]](s32), [[EXTRACT1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0, 0)
+    $vgpr0_vgpr1 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s32_undef_undef
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v2s32_undef_undef
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[DEF]](s32), [[DEF]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(undef, undef)
+    $vgpr0_vgpr1 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s32_undef_0
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v2s32_undef_0
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 0
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[DEF]](s32), [[EXTRACT]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(undef, 0)
+    $vgpr0_vgpr1 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s32_0_undef
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v2s32_0_undef
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<2 x s32>), 0
+    ; CHECK: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[EXTRACT]](s32), [[DEF]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0, undef)
+    $vgpr0_vgpr1 = COPY %2
+
+...
+
+---
+name: shufflevector_v3s32_3_2_1
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5
+
+    ; CHECK-LABEL: name: shufflevector_v3s32_3_2_1
+    ; CHECK: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5
+    ; CHECK: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY1]](<3 x s32>), 0
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<3 x s32>), 64
+    ; CHECK: [[EXTRACT2:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<3 x s32>), 32
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[EXTRACT]](s32), [[EXTRACT1]](s32), [[EXTRACT2]](s32)
+    ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
+    %2:_(<3 x s32>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(3, 2, 1)
+    $vgpr0_vgpr1_vgpr2 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s16_0_1
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; CHECK-LABEL: name: shufflevector_v2s16_0_1
+    ; CHECK: liveins: $vgpr0, $vgpr1
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s16) = G_EXTRACT [[COPY]](<2 x s16>), 0
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(s16) = G_EXTRACT [[COPY]](<2 x s16>), 16
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[EXTRACT]](s16), [[EXTRACT1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr1
+    %2:_(<2 x s16>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(0, 1)
+    $vgpr0 = COPY %2
+
+...
+
+---
+name: shufflevector_v2s16_1_0
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; CHECK-LABEL: name: shufflevector_v2s16_1_0
+    ; CHECK: liveins: $vgpr0, $vgpr1
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(s16) = G_EXTRACT [[COPY]](<2 x s16>), 16
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(s16) = G_EXTRACT [[COPY]](<2 x s16>), 0
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[EXTRACT]](s16), [[EXTRACT1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr1
+    %2:_(<2 x s16>) = G_SHUFFLE_VECTOR %0, %1, shufflemask(1, 0)
+    $vgpr0 = COPY %2
+
+...
+
+---
+name: shufflevector_v3s16_2_0
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: shufflevector_v3s16_2_0
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0
+    ; CHECK: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>)
+    ; CHECK: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16)
+    ; CHECK: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16)
+    ; CHECK: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT]](s32), [[SEXT1]](s32), [[SEXT2]](s32)
+    ; CHECK: [[EXTRACT2:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR]](<3 x s32>), 32
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT2]](s32)
+    ; CHECK: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>)
+    ; CHECK: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16)
+    ; CHECK: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16)
+    ; CHECK: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV5]](s16)
+    ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT3]](s32), [[SEXT4]](s32), [[SEXT5]](s32)
+    ; CHECK: [[EXTRACT3:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR1]](<3 x s32>), 32
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT3]](s32)
+    ; CHECK: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[DEF]](s32)
+    ; CHECK: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>)
+    ; CHECK: [[SEXT6:%[0-9]+]]:_(s32) = G_SEXT [[UV6]](s16)
+    ; CHECK: [[SEXT7:%[0-9]+]]:_(s32) = G_SEXT [[UV7]](s16)
+    ; CHECK: [[SEXT8:%[0-9]+]]:_(s32) = G_SEXT [[UV8]](s16)
+    ; CHECK: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT6]](s32), [[SEXT7]](s32), [[SEXT8]](s32)
+    ; CHECK: [[EXTRACT4:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR2]](<3 x s32>), 0
+    ; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT4]](s32)
+    ; CHECK: [[BUILD_VECTOR3:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR3]](<4 x s16>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<3 x s16>) = G_EXTRACT %0, 0
+    %3:_(<3 x s16>) = G_EXTRACT %1, 0
+    %4:_(<4 x s16>) = G_SHUFFLE_VECTOR %2, %3, shufflemask(5, 1, 3, 0)
+    $vgpr0_vgpr1 = COPY %4
+
+...