GlobalISel: Implement lower for G_EXTRACT_VECTOR_ELT

[arsenm]

Summary

Use the basic store to stack and reload.

[arsenm]

Fix test source error that should have been caught by the verifier

[arsenm]

Fix MachinePointerInfo address space in variable case

[paquette]

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

5193

Would be nice to have a simple example of what the lowering looks like in the comment here; the testcase is pretty long.

5199

maybe some debug output here?

[arsenm]

Comments

[paquette]

LGTM

[arsenm]

0b7de7966ff0015d4ffb522e994b3c67307f284a

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

private:
   /// Perform generic multiplication of values held in multiple registers.
   /// Generated instructions use only types NarrowTy and i1.
   /// Destination can be same or two times size of the source.
   void multiplyRegisters(SmallVectorImpl<Register> &DstRegs,
                          ArrayRef<Register> Src1Regs,
                          ArrayRef<Register> Src2Regs, LLT NarrowTy);
 
 public:
+  /// Return the alignment to use for a stack temporary object with the given
+  /// type.
+  Align getStackTemporaryAlignment(LLT Type, Align MinAlign = Align()) const;
+
+  /// Create a stack temporary based on the size in bytes and the alignment
+  MachineInstrBuilder createStackTemporary(TypeSize Bytes, Align Alignment,
+                                           MachinePointerInfo &PtrInfo);
+
+  /// Get a pointer to vector element \p Index located in memory for a vector of
+  /// type \p VecTy starting at a base address of \p VecPtr. If \p Index is out
+  /// of bounds the returned pointer is unspecified, but will be within the
+  /// vector bounds.
+  Register getVectorElementPointer(Register VecPtr, LLT VecTy, Register Index);
+
   LegalizeResult fewerElementsVectorImplicitDef(MachineInstr &MI,
                                                 unsigned TypeIdx, LLT NarrowTy);
 
   /// Legalize a instruction with a vector type where each operand may have a
   /// different element type. All type indexes must have the same number of
   /// elements.
   LegalizeResult fewerElementsVectorMultiEltType(MachineInstr &MI,
                                                  unsigned TypeIdx, LLT NarrowTy);
public:
   LegalizeResult lowerMinMax(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFCopySign(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFMinNumMaxNum(MachineInstr &MI);
   LegalizeResult lowerFMad(MachineInstr &MI);
   LegalizeResult lowerIntrinsicRound(MachineInstr &MI);
   LegalizeResult lowerFFloor(MachineInstr &MI);
   LegalizeResult lowerMergeValues(MachineInstr &MI);
   LegalizeResult lowerUnmergeValues(MachineInstr &MI);
+  LegalizeResult lowerExtractVectorElt(MachineInstr &MI);
   LegalizeResult lowerShuffleVector(MachineInstr &MI);
   LegalizeResult lowerDynStackAlloc(MachineInstr &MI);
   LegalizeResult lowerExtract(MachineInstr &MI);
   LegalizeResult lowerInsert(MachineInstr &MI);
   LegalizeResult lowerSADDO_SSUBO(MachineInstr &MI);
   LegalizeResult lowerAddSubSatToMinMax(MachineInstr &MI);
   LegalizeResult lowerAddSubSatToAddoSubo(MachineInstr &MI);
   LegalizeResult lowerBswap(MachineInstr &MI);

llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h

public:
   /// Loads the value stored at \p Addr. Puts the result in \p Res.
   ///
   /// \pre setBasicBlock or setMI must have been called.
   /// \pre \p Res must be a generic virtual register.
   /// \pre \p Addr must be a generic virtual register with pointer type.
   ///
   /// \return a MachineInstrBuilder for the newly created instruction.
   MachineInstrBuilder buildLoad(const DstOp &Res, const SrcOp &Addr,
-                                MachineMemOperand &MMO);
+                                MachineMemOperand &MMO) {
+    return buildLoadInstr(TargetOpcode::G_LOAD, Res, Addr, MMO);
+  }
+
+  /// Build and insert a G_LOAD instruction, while constructing the
+  /// MachineMemOperand.
+  MachineInstrBuilder
+  buildLoad(const DstOp &Res, const SrcOp &Addr, MachinePointerInfo PtrInfo,
+            Align Alignment,
+            MachineMemOperand::Flags MMOFlags = MachineMemOperand::MONone,
+            const AAMDNodes &AAInfo = AAMDNodes());
 
   /// Build and insert `Res = <opcode> Addr, MMO`.
   ///
   /// Loads the value stored at \p Addr. Puts the result in \p Res.
   ///
   /// \pre setBasicBlock or setMI must have been called.
   /// \pre \p Res must be a generic virtual register.
   /// \pre \p Addr must be a generic virtual register with pointer type.
public:
   /// \pre setBasicBlock or setMI must have been called.
   /// \pre \p Val must be a generic virtual register.
   /// \pre \p Addr must be a generic virtual register with pointer type.
   ///
   /// \return a MachineInstrBuilder for the newly created instruction.
   MachineInstrBuilder buildStore(const SrcOp &Val, const SrcOp &Addr,
                                  MachineMemOperand &MMO);
 
+  /// Build and insert a G_STORE instruction, while constructing the
+  /// MachineMemOperand.
+  MachineInstrBuilder
+  buildStore(const SrcOp &Val, const SrcOp &Addr, MachinePointerInfo PtrInfo,
+             Align Alignment,
+             MachineMemOperand::Flags MMOFlags = MachineMemOperand::MONone,
+             const AAMDNodes &AAInfo = AAMDNodes());
+
   /// Build and insert `Res0, ... = G_EXTRACT Src, Idx0`.
   ///
   /// \pre setBasicBlock or setMI must have been called.
   /// \pre \p Res and \p Src must be generic virtual registers.
   ///
   /// \return a MachineInstrBuilder for the newly created instruction.
   MachineInstrBuilder buildExtract(const DstOp &Res, const SrcOp &Src, uint64_t Index);
 

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

 /// primary legalization.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/CodeGen/GlobalISel/LegalizerHelper.h"
 #include "llvm/CodeGen/GlobalISel/CallLowering.h"
 #include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
 #include "llvm/CodeGen/GlobalISel/LegalizerInfo.h"
+#include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/TargetFrameLowering.h"
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 
 #define DEBUG_TYPE "legalizer"
 
 using namespace llvm;
 using namespace LegalizeActions;
+using namespace MIPatternMatch;
 
 /// Try to break down \p OrigTy into \p NarrowTy sized pieces.
 ///
 /// Returns the number of \p NarrowTy elements needed to reconstruct \p OrigTy,
 /// with any leftover piece as type \p LeftoverTy
 ///
 /// Returns -1 in the first element of the pair if the breakdown is not
 /// satisfiable.
case TargetOpcode::G_SEXT_INREG: {
     Register TmpRes = MRI.createGenericVirtualRegister(DstTy);
 
     auto MIBSz = MIRBuilder.buildConstant(DstTy, DstTy.getScalarSizeInBits() - SizeInBits);
     MIRBuilder.buildShl(TmpRes, SrcReg, MIBSz->getOperand(0));
     MIRBuilder.buildAShr(DstReg, TmpRes, MIBSz->getOperand(0));
     MI.eraseFromParent();
     return Legalized;
   }
+  case G_EXTRACT_VECTOR_ELT:
+    return lowerExtractVectorElt(MI);
   case G_SHUFFLE_VECTOR:
     return lowerShuffleVector(MI);
   case G_DYN_STACKALLOC:
     return lowerDynStackAlloc(MI);
   case G_EXTRACT:
     return lowerExtract(MI);
   case G_INSERT:
     return lowerInsert(MI);
case G_SSUBSAT: {
     // preferred boolean type for the carry out for the query.
     if (LI.isLegalOrCustom({G_SMIN, Ty}) && LI.isLegalOrCustom({G_SMAX, Ty}))
       return lowerAddSubSatToMinMax(MI);
     return lowerAddSubSatToAddoSubo(MI);
   }
   }
 }
 
+Align LegalizerHelper::getStackTemporaryAlignment(LLT Ty,
+                                                  Align MinAlign) const {
+  // FIXME: We're missing a way to go back from LLT to llvm::Type to query the
+  // datalayout for the preferred alignment. Also there should be a target hook
+  // for this to allow targets to reduce the alignment and ignore the
+  // datalayout. e.g. AMDGPU should always use a 4-byte alignment, regardless of
+  // the type.
+  return std::max(Align(PowerOf2Ceil(Ty.getSizeInBytes())), MinAlign);
+}
+
+MachineInstrBuilder
+LegalizerHelper::createStackTemporary(TypeSize Bytes, Align Alignment,
+                                      MachinePointerInfo &PtrInfo) {
+  MachineFunction &MF = MIRBuilder.getMF();
+  const DataLayout &DL = MIRBuilder.getDataLayout();
+  int FrameIdx = MF.getFrameInfo().CreateStackObject(Bytes, Alignment, false);
+
+  unsigned AddrSpace = DL.getAllocaAddrSpace();
+  LLT FramePtrTy = LLT::pointer(AddrSpace, DL.getPointerSizeInBits(AddrSpace));
+
+  PtrInfo = MachinePointerInfo::getFixedStack(MF, FrameIdx);
+  return MIRBuilder.buildFrameIndex(FramePtrTy, FrameIdx);
+}
+
+static Register clampDynamicVectorIndex(MachineIRBuilder &B, Register IdxReg,
+                                        LLT VecTy) {
+  int64_t IdxVal;
+  if (mi_match(IdxReg, *B.getMRI(), m_ICst(IdxVal)))
+    return IdxReg;
+
+  LLT IdxTy = B.getMRI()->getType(IdxReg);
+  unsigned NElts = VecTy.getNumElements();
+  if (isPowerOf2_32(NElts)) {
+    APInt Imm = APInt::getLowBitsSet(IdxTy.getSizeInBits(), Log2_32(NElts));
+    return B.buildAnd(IdxTy, IdxReg, B.buildConstant(IdxTy, Imm)).getReg(0);
+  }
+
+  return B.buildUMin(IdxTy, IdxReg, B.buildConstant(IdxTy, NElts - 1))
+      .getReg(0);
+}
+
+Register LegalizerHelper::getVectorElementPointer(Register VecPtr, LLT VecTy,
+                                                  Register Index) {
+  LLT EltTy = VecTy.getElementType();
+
+  // Calculate the element offset and add it to the pointer.
+  unsigned EltSize = EltTy.getSizeInBits() / 8; // FIXME: should be ABI size.
+  assert(EltSize * 8 == EltTy.getSizeInBits() &&
+         "Converting bits to bytes lost precision");
+
+  Index = clampDynamicVectorIndex(MIRBuilder, Index, VecTy);
+
+  LLT IdxTy = MRI.getType(Index);
+  auto Mul = MIRBuilder.buildMul(IdxTy, Index,
+                                 MIRBuilder.buildConstant(IdxTy, EltSize));
+
+  LLT PtrTy = MRI.getType(VecPtr);
+  return MIRBuilder.buildPtrAdd(PtrTy, VecPtr, Mul).getReg(0);
+}
+
 LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorImplicitDef(
     MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy) {
   SmallVector<Register, 2> DstRegs;
 
   unsigned NarrowSize = NarrowTy.getSizeInBits();
   Register DstReg = MI.getOperand(0).getReg();
   unsigned Size = MRI.getType(DstReg).getSizeInBits();
   int NumParts = Size / NarrowSize;
for (unsigned I = 1; I != NumDst; ++I, Offset += DstSize) {
     auto Shift = MIRBuilder.buildLShr(IntTy, SrcReg, ShiftAmt);
     MIRBuilder.buildTrunc(MI.getOperand(I), Shift);
   }
 
   MI.eraseFromParent();
   return Legalized;
 }
 
+/// Lower a vector extract by writing the vector to a stack temporary and
+/// reloading the element.
+///
+/// %dst = G_EXTRACT_VECTOR_ELT %vec, %idx
+///  =>
+///  %stack_temp = G_FRAME_INDEX
+///  G_STORE %vec, %stack_temp
+///  %idx = clamp(%idx, %vec.getNumElements())
+///  %element_ptr = G_PTR_ADD %stack_temp, %idx
+///  %dst = G_LOAD %element_ptr
+LegalizerHelper::LegalizeResult
+LegalizerHelper::lowerExtractVectorElt(MachineInstr &MI) {
+  Register DstReg = MI.getOperand(0).getReg();
+  Register SrcVec = MI.getOperand(1).getReg();
+  Register Idx = MI.getOperand(2).getReg();
+  LLT VecTy = MRI.getType(SrcVec);
+  LLT EltTy = VecTy.getElementType();
+  if (!EltTy.isByteSized()) { // Not implemented.
+    LLVM_DEBUG(dbgs() << "Can't handle non-byte element vectors yet\n");
+    return UnableToLegalize;
+  }
+
+  unsigned EltBytes = EltTy.getSizeInBytes();
+  Align StoreAlign = getStackTemporaryAlignment(VecTy);
+  Align LoadAlign;
+
+  MachinePointerInfo PtrInfo;
+  auto StackTemp = createStackTemporary(TypeSize::Fixed(VecTy.getSizeInBytes()),
+                                        StoreAlign, PtrInfo);
+  MIRBuilder.buildStore(SrcVec, StackTemp, PtrInfo, StoreAlign);
+
+  // Get the pointer to the element, and be sure not to hit undefined behavior
+  // if the index is out of bounds.
+  Register LoadPtr = getVectorElementPointer(StackTemp.getReg(0), VecTy, Idx);
+
+  int64_t IdxVal;
+  if (mi_match(Idx, MRI, m_ICst(IdxVal))) {
+    int64_t Offset = IdxVal * EltBytes;
+    PtrInfo = PtrInfo.getWithOffset(Offset);
+    LoadAlign = commonAlignment(StoreAlign, Offset);
+  } else {
+    // We lose information with a variable offset.
+    LoadAlign = getStackTemporaryAlignment(EltTy);
+    PtrInfo = MachinePointerInfo(MRI.getType(LoadPtr).getAddressSpace());
+  }
+
+  MIRBuilder.buildLoad(DstReg, LoadPtr, PtrInfo, LoadAlign);
+  MI.eraseFromParent();
+  return Legalized;
+}
+
 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 Src0Ty = MRI.getType(Src0Reg);
   LLT DstTy = MRI.getType(DstReg);
   LLT IdxTy = LLT::scalar(32);

llvm/lib/CodeGen/GlobalISel/MachineIRBuilder.cpp

 //===-- llvm/CodeGen/GlobalISel/MachineIRBuilder.cpp - MIBuilder--*- C++ -*-==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 /// \file
 /// This file implements the MachineIRBuidler class.
 //===----------------------------------------------------------------------===//
 #include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
+#include "llvm/Analysis/MemoryLocation.h"
 #include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
-
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetOpcodes.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"

 
 MachineInstrBuilder MachineIRBuilder::buildBrCond(Register Tst,
                                                   MachineBasicBlock &Dest) {
   assert(getMRI()->getType(Tst).isScalar() && "invalid operand type");
 
   return buildInstr(TargetOpcode::G_BRCOND).addUse(Tst).addMBB(&Dest);
 }
 
-MachineInstrBuilder MachineIRBuilder::buildLoad(const DstOp &Res,
-                                                const SrcOp &Addr,
-                                                MachineMemOperand &MMO) {
-  return buildLoadInstr(TargetOpcode::G_LOAD, Res, Addr, MMO);
+MachineInstrBuilder
+MachineIRBuilder::buildLoad(const DstOp &Dst, const SrcOp &Addr,
+                            MachinePointerInfo PtrInfo, Align Alignment,
+                            MachineMemOperand::Flags MMOFlags,
+                            const AAMDNodes &AAInfo) {
+  MMOFlags |= MachineMemOperand::MOLoad;
+  assert((MMOFlags & MachineMemOperand::MOStore) == 0);
+
+  uint64_t Size = MemoryLocation::getSizeOrUnknown(
+      TypeSize::Fixed(Dst.getLLTTy(*getMRI()).getSizeInBytes()));
+  MachineMemOperand *MMO =
+      getMF().getMachineMemOperand(PtrInfo, MMOFlags, Size, Alignment, AAInfo);
+  return buildLoad(Dst, Addr, *MMO);
 }
 
 MachineInstrBuilder MachineIRBuilder::buildLoadInstr(unsigned Opcode,
                                                      const DstOp &Res,
                                                      const SrcOp &Addr,
                                                      MachineMemOperand &MMO) {
   assert(Res.getLLTTy(*getMRI()).isValid() && "invalid operand type");
   assert(Addr.getLLTTy(*getMRI()).isPointer() && "invalid operand type");
MachineInstrBuilder MachineIRBuilder::buildStore(const SrcOp &Val,
 
   auto MIB = buildInstr(TargetOpcode::G_STORE);
   Val.addSrcToMIB(MIB);
   Addr.addSrcToMIB(MIB);
   MIB.addMemOperand(&MMO);
   return MIB;
 }
 
+MachineInstrBuilder
+MachineIRBuilder::buildStore(const SrcOp &Val, const SrcOp &Addr,
+                             MachinePointerInfo PtrInfo, Align Alignment,
+                             MachineMemOperand::Flags MMOFlags,
+                             const AAMDNodes &AAInfo) {
+  MMOFlags |= MachineMemOperand::MOStore;
+  assert((MMOFlags & MachineMemOperand::MOLoad) == 0);
+
+  uint64_t Size = MemoryLocation::getSizeOrUnknown(
+      TypeSize::Fixed(Val.getLLTTy(*getMRI()).getSizeInBytes()));
+  MachineMemOperand *MMO =
+      getMF().getMachineMemOperand(PtrInfo, MMOFlags, Size, Alignment, AAInfo);
+  return buildStore(Val, Addr, *MMO);
+}
+
 MachineInstrBuilder MachineIRBuilder::buildAnyExt(const DstOp &Res,
                                                   const SrcOp &Op) {
   return buildInstr(TargetOpcode::G_ANYEXT, Res, Op);
 }
 
 MachineInstrBuilder MachineIRBuilder::buildSExt(const DstOp &Res,
                                                 const SrcOp &Op) {
   return buildInstr(TargetOpcode::G_SEXT, Res, Op);

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

getActionDefinitionsBuilder(Op)
           return (EltTy.getSizeInBits() == 16 ||
                   EltTy.getSizeInBits() % 32 == 0) &&
                  VecTy.getSizeInBits() % 32 == 0 &&
                  VecTy.getSizeInBits() <= MaxRegisterSize &&
                  IdxTy.getSizeInBits() == 32;
         })
       .clampScalar(EltTypeIdx, S32, S64)
       .clampScalar(VecTypeIdx, S32, S64)
-      .clampScalar(IdxTypeIdx, S32, S32);
+      .clampScalar(IdxTypeIdx, S32, S32)
+      // TODO: Clamp the number of elements before resorting to stack lowering.
+      // It should only be necessary with variable indexes.
+      // As a last resort, lower to the stack
+      .lower();
   }
 
   getActionDefinitionsBuilder(G_EXTRACT_VECTOR_ELT)
     .unsupportedIf([=](const LegalityQuery &Query) {
         const LLT &EltTy = Query.Types[1].getElementType();
         return Query.Types[0] != EltTy;
       });
 

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-extract-vector-elt.mir

bb.0:
     ; CHECK: [[EXTRACT:%[0-9]+]]:_(s32) = G_EXTRACT [[COPY]](<4 x s32>), 0
     ; CHECK: $vgpr0 = COPY [[EXTRACT]](s32)
     %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
     %1:_(s64) = G_CONSTANT i64 0
     %2:_(s32) = G_TRUNC %1
     %3:_(s32) = G_EXTRACT_VECTOR_ELT %0, %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: extract_vector_elt_7_v64s32
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: extract_vector_elt_7_v64s32
+    ; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $sgpr0_sgpr1
+    ; CHECK: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p1) :: (load 64, align 4, addrspace 4)
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 64
+    ; CHECK: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C]](s64)
+    ; CHECK: [[LOAD1:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD]](p1) :: (load 64 + 64, align 4, addrspace 4)
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 128
+    ; CHECK: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C1]](s64)
+    ; CHECK: [[LOAD2:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p1) :: (load 64 + 128, align 4, addrspace 4)
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 192
+    ; CHECK: [[PTR_ADD2:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; CHECK: [[LOAD3:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD2]](p1) :: (load 64 + 192, align 4, addrspace 4)
+    ; CHECK: [[FRAME_INDEX:%[0-9]+]]:_(p5) = G_FRAME_INDEX %stack.0
+    ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32), [[UV12:%[0-9]+]]:_(s32), [[UV13:%[0-9]+]]:_(s32), [[UV14:%[0-9]+]]:_(s32), [[UV15:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<16 x s32>)
+    ; CHECK: [[UV16:%[0-9]+]]:_(s32), [[UV17:%[0-9]+]]:_(s32), [[UV18:%[0-9]+]]:_(s32), [[UV19:%[0-9]+]]:_(s32), [[UV20:%[0-9]+]]:_(s32), [[UV21:%[0-9]+]]:_(s32), [[UV22:%[0-9]+]]:_(s32), [[UV23:%[0-9]+]]:_(s32), [[UV24:%[0-9]+]]:_(s32), [[UV25:%[0-9]+]]:_(s32), [[UV26:%[0-9]+]]:_(s32), [[UV27:%[0-9]+]]:_(s32), [[UV28:%[0-9]+]]:_(s32), [[UV29:%[0-9]+]]:_(s32), [[UV30:%[0-9]+]]:_(s32), [[UV31:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD1]](<16 x s32>)
+    ; CHECK: [[UV32:%[0-9]+]]:_(s32), [[UV33:%[0-9]+]]:_(s32), [[UV34:%[0-9]+]]:_(s32), [[UV35:%[0-9]+]]:_(s32), [[UV36:%[0-9]+]]:_(s32), [[UV37:%[0-9]+]]:_(s32), [[UV38:%[0-9]+]]:_(s32), [[UV39:%[0-9]+]]:_(s32), [[UV40:%[0-9]+]]:_(s32), [[UV41:%[0-9]+]]:_(s32), [[UV42:%[0-9]+]]:_(s32), [[UV43:%[0-9]+]]:_(s32), [[UV44:%[0-9]+]]:_(s32), [[UV45:%[0-9]+]]:_(s32), [[UV46:%[0-9]+]]:_(s32), [[UV47:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD2]](<16 x s32>)
+    ; CHECK: [[UV48:%[0-9]+]]:_(s32), [[UV49:%[0-9]+]]:_(s32), [[UV50:%[0-9]+]]:_(s32), [[UV51:%[0-9]+]]:_(s32), [[UV52:%[0-9]+]]:_(s32), [[UV53:%[0-9]+]]:_(s32), [[UV54:%[0-9]+]]:_(s32), [[UV55:%[0-9]+]]:_(s32), [[UV56:%[0-9]+]]:_(s32), [[UV57:%[0-9]+]]:_(s32), [[UV58:%[0-9]+]]:_(s32), [[UV59:%[0-9]+]]:_(s32), [[UV60:%[0-9]+]]:_(s32), [[UV61:%[0-9]+]]:_(s32), [[UV62:%[0-9]+]]:_(s32), [[UV63:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD3]](<16 x s32>)
+    ; CHECK: G_STORE [[UV]](s32), [[FRAME_INDEX]](p5) :: (store 4 into %stack.0, align 256, addrspace 5)
+    ; CHECK: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
+    ; CHECK: [[PTR_ADD3:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C3]](s32)
+    ; CHECK: G_STORE [[UV1]](s32), [[PTR_ADD3]](p5) :: (store 4 into %stack.0 + 4, align 256, addrspace 5)
+    ; CHECK: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CHECK: [[PTR_ADD4:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C4]](s32)
+    ; CHECK: G_STORE [[UV2]](s32), [[PTR_ADD4]](p5) :: (store 4 into %stack.0 + 8, align 256, addrspace 5)
+    ; CHECK: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
+    ; CHECK: [[PTR_ADD5:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C5]](s32)
+    ; CHECK: G_STORE [[UV3]](s32), [[PTR_ADD5]](p5) :: (store 4 into %stack.0 + 12, align 256, addrspace 5)
+    ; CHECK: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[PTR_ADD6:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C6]](s32)
+    ; CHECK: G_STORE [[UV4]](s32), [[PTR_ADD6]](p5) :: (store 4 into %stack.0 + 16, align 256, addrspace 5)
+    ; CHECK: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 20
+    ; CHECK: [[PTR_ADD7:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C7]](s32)
+    ; CHECK: G_STORE [[UV5]](s32), [[PTR_ADD7]](p5) :: (store 4 into %stack.0 + 20, align 256, addrspace 5)
+    ; CHECK: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CHECK: [[PTR_ADD8:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C8]](s32)
+    ; CHECK: G_STORE [[UV6]](s32), [[PTR_ADD8]](p5) :: (store 4 into %stack.0 + 24, align 256, addrspace 5)
+    ; CHECK: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 28
+    ; CHECK: [[PTR_ADD9:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C9]](s32)
+    ; CHECK: [[COPY1:%[0-9]+]]:_(p5) = COPY [[PTR_ADD9]](p5)
+    ; CHECK: G_STORE [[UV7]](s32), [[COPY1]](p5) :: (store 4 into %stack.0 + 28, align 256, addrspace 5)
+    ; CHECK: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 32
+    ; CHECK: [[PTR_ADD10:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C10]](s32)
+    ; CHECK: G_STORE [[UV8]](s32), [[PTR_ADD10]](p5) :: (store 4 into %stack.0 + 32, align 256, addrspace 5)
+    ; CHECK: [[C11:%[0-9]+]]:_(s32) = G_CONSTANT i32 36
+    ; CHECK: [[PTR_ADD11:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C11]](s32)
+    ; CHECK: G_STORE [[UV9]](s32), [[PTR_ADD11]](p5) :: (store 4 into %stack.0 + 36, align 256, addrspace 5)
+    ; CHECK: [[C12:%[0-9]+]]:_(s32) = G_CONSTANT i32 40
+    ; CHECK: [[PTR_ADD12:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C12]](s32)
+    ; CHECK: G_STORE [[UV10]](s32), [[PTR_ADD12]](p5) :: (store 4 into %stack.0 + 40, align 256, addrspace 5)
+    ; CHECK: [[C13:%[0-9]+]]:_(s32) = G_CONSTANT i32 44
+    ; CHECK: [[PTR_ADD13:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C13]](s32)
+    ; CHECK: G_STORE [[UV11]](s32), [[PTR_ADD13]](p5) :: (store 4 into %stack.0 + 44, align 256, addrspace 5)
+    ; CHECK: [[C14:%[0-9]+]]:_(s32) = G_CONSTANT i32 48
+    ; CHECK: [[PTR_ADD14:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C14]](s32)
+    ; CHECK: G_STORE [[UV12]](s32), [[PTR_ADD14]](p5) :: (store 4 into %stack.0 + 48, align 256, addrspace 5)
+    ; CHECK: [[C15:%[0-9]+]]:_(s32) = G_CONSTANT i32 52
+    ; CHECK: [[PTR_ADD15:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C15]](s32)
+    ; CHECK: G_STORE [[UV13]](s32), [[PTR_ADD15]](p5) :: (store 4 into %stack.0 + 52, align 256, addrspace 5)
+    ; CHECK: [[C16:%[0-9]+]]:_(s32) = G_CONSTANT i32 56
+    ; CHECK: [[PTR_ADD16:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C16]](s32)
+    ; CHECK: G_STORE [[UV14]](s32), [[PTR_ADD16]](p5) :: (store 4 into %stack.0 + 56, align 256, addrspace 5)
+    ; CHECK: [[C17:%[0-9]+]]:_(s32) = G_CONSTANT i32 60
+    ; CHECK: [[PTR_ADD17:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C17]](s32)
+    ; CHECK: G_STORE [[UV15]](s32), [[PTR_ADD17]](p5) :: (store 4 into %stack.0 + 60, align 256, addrspace 5)
+    ; CHECK: [[C18:%[0-9]+]]:_(s32) = G_CONSTANT i32 64
+    ; CHECK: [[PTR_ADD18:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C18]](s32)
+    ; CHECK: G_STORE [[UV16]](s32), [[PTR_ADD18]](p5) :: (store 4 into %stack.0 + 64, align 256, addrspace 5)
+    ; CHECK: [[C19:%[0-9]+]]:_(s32) = G_CONSTANT i32 68
+    ; CHECK: [[PTR_ADD19:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C19]](s32)
+    ; CHECK: G_STORE [[UV17]](s32), [[PTR_ADD19]](p5) :: (store 4 into %stack.0 + 68, align 256, addrspace 5)
+    ; CHECK: [[C20:%[0-9]+]]:_(s32) = G_CONSTANT i32 72
+    ; CHECK: [[PTR_ADD20:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C20]](s32)
+    ; CHECK: G_STORE [[UV18]](s32), [[PTR_ADD20]](p5) :: (store 4 into %stack.0 + 72, align 256, addrspace 5)
+    ; CHECK: [[C21:%[0-9]+]]:_(s32) = G_CONSTANT i32 76
+    ; CHECK: [[PTR_ADD21:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C21]](s32)
+    ; CHECK: G_STORE [[UV19]](s32), [[PTR_ADD21]](p5) :: (store 4 into %stack.0 + 76, align 256, addrspace 5)
+    ; CHECK: [[C22:%[0-9]+]]:_(s32) = G_CONSTANT i32 80
+    ; CHECK: [[PTR_ADD22:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C22]](s32)
+    ; CHECK: G_STORE [[UV20]](s32), [[PTR_ADD22]](p5) :: (store 4 into %stack.0 + 80, align 256, addrspace 5)
+    ; CHECK: [[C23:%[0-9]+]]:_(s32) = G_CONSTANT i32 84
+    ; CHECK: [[PTR_ADD23:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C23]](s32)
+    ; CHECK: G_STORE [[UV21]](s32), [[PTR_ADD23]](p5) :: (store 4 into %stack.0 + 84, align 256, addrspace 5)
+    ; CHECK: [[C24:%[0-9]+]]:_(s32) = G_CONSTANT i32 88
+    ; CHECK: [[PTR_ADD24:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C24]](s32)
+    ; CHECK: G_STORE [[UV22]](s32), [[PTR_ADD24]](p5) :: (store 4 into %stack.0 + 88, align 256, addrspace 5)
+    ; CHECK: [[C25:%[0-9]+]]:_(s32) = G_CONSTANT i32 92
+    ; CHECK: [[PTR_ADD25:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C25]](s32)
+    ; CHECK: G_STORE [[UV23]](s32), [[PTR_ADD25]](p5) :: (store 4 into %stack.0 + 92, align 256, addrspace 5)
+    ; CHECK: [[C26:%[0-9]+]]:_(s32) = G_CONSTANT i32 96
+    ; CHECK: [[PTR_ADD26:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C26]](s32)
+    ; CHECK: G_STORE [[UV24]](s32), [[PTR_ADD26]](p5) :: (store 4 into %stack.0 + 96, align 256, addrspace 5)
+    ; CHECK: [[C27:%[0-9]+]]:_(s32) = G_CONSTANT i32 100
+    ; CHECK: [[PTR_ADD27:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C27]](s32)
+    ; CHECK: G_STORE [[UV25]](s32), [[PTR_ADD27]](p5) :: (store 4 into %stack.0 + 100, align 256, addrspace 5)
+    ; CHECK: [[C28:%[0-9]+]]:_(s32) = G_CONSTANT i32 104
+    ; CHECK: [[PTR_ADD28:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C28]](s32)
+    ; CHECK: G_STORE [[UV26]](s32), [[PTR_ADD28]](p5) :: (store 4 into %stack.0 + 104, align 256, addrspace 5)
+    ; CHECK: [[C29:%[0-9]+]]:_(s32) = G_CONSTANT i32 108
+    ; CHECK: [[PTR_ADD29:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C29]](s32)
+    ; CHECK: G_STORE [[UV27]](s32), [[PTR_ADD29]](p5) :: (store 4 into %stack.0 + 108, align 256, addrspace 5)
+    ; CHECK: [[C30:%[0-9]+]]:_(s32) = G_CONSTANT i32 112
+    ; CHECK: [[PTR_ADD30:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C30]](s32)
+    ; CHECK: G_STORE [[UV28]](s32), [[PTR_ADD30]](p5) :: (store 4 into %stack.0 + 112, align 256, addrspace 5)
+    ; CHECK: [[C31:%[0-9]+]]:_(s32) = G_CONSTANT i32 116
+    ; CHECK: [[PTR_ADD31:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C31]](s32)
+    ; CHECK: G_STORE [[UV29]](s32), [[PTR_ADD31]](p5) :: (store 4 into %stack.0 + 116, align 256, addrspace 5)
+    ; CHECK: [[C32:%[0-9]+]]:_(s32) = G_CONSTANT i32 120
+    ; CHECK: [[PTR_ADD32:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C32]](s32)
+    ; CHECK: G_STORE [[UV30]](s32), [[PTR_ADD32]](p5) :: (store 4 into %stack.0 + 120, align 256, addrspace 5)
+    ; CHECK: [[C33:%[0-9]+]]:_(s32) = G_CONSTANT i32 124
+    ; CHECK: [[PTR_ADD33:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C33]](s32)
+    ; CHECK: G_STORE [[UV31]](s32), [[PTR_ADD33]](p5) :: (store 4 into %stack.0 + 124, align 256, addrspace 5)
+    ; CHECK: [[C34:%[0-9]+]]:_(s32) = G_CONSTANT i32 128
+    ; CHECK: [[PTR_ADD34:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C34]](s32)
+    ; CHECK: G_STORE [[UV32]](s32), [[PTR_ADD34]](p5) :: (store 4 into %stack.0 + 128, align 256, addrspace 5)
+    ; CHECK: [[C35:%[0-9]+]]:_(s32) = G_CONSTANT i32 132
+    ; CHECK: [[PTR_ADD35:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C35]](s32)
+    ; CHECK: G_STORE [[UV33]](s32), [[PTR_ADD35]](p5) :: (store 4 into %stack.0 + 132, align 256, addrspace 5)
+    ; CHECK: [[C36:%[0-9]+]]:_(s32) = G_CONSTANT i32 136
+    ; CHECK: [[PTR_ADD36:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C36]](s32)
+    ; CHECK: G_STORE [[UV34]](s32), [[PTR_ADD36]](p5) :: (store 4 into %stack.0 + 136, align 256, addrspace 5)
+    ; CHECK: [[C37:%[0-9]+]]:_(s32) = G_CONSTANT i32 140
+    ; CHECK: [[PTR_ADD37:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C37]](s32)
+    ; CHECK: G_STORE [[UV35]](s32), [[PTR_ADD37]](p5) :: (store 4 into %stack.0 + 140, align 256, addrspace 5)
+    ; CHECK: [[C38:%[0-9]+]]:_(s32) = G_CONSTANT i32 144
+    ; CHECK: [[PTR_ADD38:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C38]](s32)
+    ; CHECK: G_STORE [[UV36]](s32), [[PTR_ADD38]](p5) :: (store 4 into %stack.0 + 144, align 256, addrspace 5)
+    ; CHECK: [[C39:%[0-9]+]]:_(s32) = G_CONSTANT i32 148
+    ; CHECK: [[PTR_ADD39:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C39]](s32)
+    ; CHECK: G_STORE [[UV37]](s32), [[PTR_ADD39]](p5) :: (store 4 into %stack.0 + 148, align 256, addrspace 5)
+    ; CHECK: [[C40:%[0-9]+]]:_(s32) = G_CONSTANT i32 152
+    ; CHECK: [[PTR_ADD40:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C40]](s32)
+    ; CHECK: G_STORE [[UV38]](s32), [[PTR_ADD40]](p5) :: (store 4 into %stack.0 + 152, align 256, addrspace 5)
+    ; CHECK: [[C41:%[0-9]+]]:_(s32) = G_CONSTANT i32 156
+    ; CHECK: [[PTR_ADD41:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C41]](s32)
+    ; CHECK: G_STORE [[UV39]](s32), [[PTR_ADD41]](p5) :: (store 4 into %stack.0 + 156, align 256, addrspace 5)
+    ; CHECK: [[C42:%[0-9]+]]:_(s32) = G_CONSTANT i32 160
+    ; CHECK: [[PTR_ADD42:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C42]](s32)
+    ; CHECK: G_STORE [[UV40]](s32), [[PTR_ADD42]](p5) :: (store 4 into %stack.0 + 160, align 256, addrspace 5)
+    ; CHECK: [[C43:%[0-9]+]]:_(s32) = G_CONSTANT i32 164
+    ; CHECK: [[PTR_ADD43:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C43]](s32)
+    ; CHECK: G_STORE [[UV41]](s32), [[PTR_ADD43]](p5) :: (store 4 into %stack.0 + 164, align 256, addrspace 5)
+    ; CHECK: [[C44:%[0-9]+]]:_(s32) = G_CONSTANT i32 168
+    ; CHECK: [[PTR_ADD44:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C44]](s32)
+    ; CHECK: G_STORE [[UV42]](s32), [[PTR_ADD44]](p5) :: (store 4 into %stack.0 + 168, align 256, addrspace 5)
+    ; CHECK: [[C45:%[0-9]+]]:_(s32) = G_CONSTANT i32 172
+    ; CHECK: [[PTR_ADD45:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C45]](s32)
+    ; CHECK: G_STORE [[UV43]](s32), [[PTR_ADD45]](p5) :: (store 4 into %stack.0 + 172, align 256, addrspace 5)
+    ; CHECK: [[C46:%[0-9]+]]:_(s32) = G_CONSTANT i32 176
+    ; CHECK: [[PTR_ADD46:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C46]](s32)
+    ; CHECK: G_STORE [[UV44]](s32), [[PTR_ADD46]](p5) :: (store 4 into %stack.0 + 176, align 256, addrspace 5)
+    ; CHECK: [[C47:%[0-9]+]]:_(s32) = G_CONSTANT i32 180
+    ; CHECK: [[PTR_ADD47:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C47]](s32)
+    ; CHECK: G_STORE [[UV45]](s32), [[PTR_ADD47]](p5) :: (store 4 into %stack.0 + 180, align 256, addrspace 5)
+    ; CHECK: [[C48:%[0-9]+]]:_(s32) = G_CONSTANT i32 184
+    ; CHECK: [[PTR_ADD48:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C48]](s32)
+    ; CHECK: G_STORE [[UV46]](s32), [[PTR_ADD48]](p5) :: (store 4 into %stack.0 + 184, align 256, addrspace 5)
+    ; CHECK: [[C49:%[0-9]+]]:_(s32) = G_CONSTANT i32 188
+    ; CHECK: [[PTR_ADD49:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C49]](s32)
+    ; CHECK: G_STORE [[UV47]](s32), [[PTR_ADD49]](p5) :: (store 4 into %stack.0 + 188, align 256, addrspace 5)
+    ; CHECK: [[C50:%[0-9]+]]:_(s32) = G_CONSTANT i32 192
+    ; CHECK: [[PTR_ADD50:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C50]](s32)
+    ; CHECK: G_STORE [[UV48]](s32), [[PTR_ADD50]](p5) :: (store 4 into %stack.0 + 192, align 256, addrspace 5)
+    ; CHECK: [[C51:%[0-9]+]]:_(s32) = G_CONSTANT i32 196
+    ; CHECK: [[PTR_ADD51:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C51]](s32)
+    ; CHECK: G_STORE [[UV49]](s32), [[PTR_ADD51]](p5) :: (store 4 into %stack.0 + 196, align 256, addrspace 5)
+    ; CHECK: [[C52:%[0-9]+]]:_(s32) = G_CONSTANT i32 200
+    ; CHECK: [[PTR_ADD52:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C52]](s32)
+    ; CHECK: G_STORE [[UV50]](s32), [[PTR_ADD52]](p5) :: (store 4 into %stack.0 + 200, align 256, addrspace 5)
+    ; CHECK: [[C53:%[0-9]+]]:_(s32) = G_CONSTANT i32 204
+    ; CHECK: [[PTR_ADD53:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C53]](s32)
+    ; CHECK: G_STORE [[UV51]](s32), [[PTR_ADD53]](p5) :: (store 4 into %stack.0 + 204, align 256, addrspace 5)
+    ; CHECK: [[C54:%[0-9]+]]:_(s32) = G_CONSTANT i32 208
+    ; CHECK: [[PTR_ADD54:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C54]](s32)
+    ; CHECK: G_STORE [[UV52]](s32), [[PTR_ADD54]](p5) :: (store 4 into %stack.0 + 208, align 256, addrspace 5)
+    ; CHECK: [[C55:%[0-9]+]]:_(s32) = G_CONSTANT i32 212
+    ; CHECK: [[PTR_ADD55:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C55]](s32)
+    ; CHECK: G_STORE [[UV53]](s32), [[PTR_ADD55]](p5) :: (store 4 into %stack.0 + 212, align 256, addrspace 5)
+    ; CHECK: [[C56:%[0-9]+]]:_(s32) = G_CONSTANT i32 216
+    ; CHECK: [[PTR_ADD56:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C56]](s32)
+    ; CHECK: G_STORE [[UV54]](s32), [[PTR_ADD56]](p5) :: (store 4 into %stack.0 + 216, align 256, addrspace 5)
+    ; CHECK: [[C57:%[0-9]+]]:_(s32) = G_CONSTANT i32 220
+    ; CHECK: [[PTR_ADD57:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C57]](s32)
+    ; CHECK: G_STORE [[UV55]](s32), [[PTR_ADD57]](p5) :: (store 4 into %stack.0 + 220, align 256, addrspace 5)
+    ; CHECK: [[C58:%[0-9]+]]:_(s32) = G_CONSTANT i32 224
+    ; CHECK: [[PTR_ADD58:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C58]](s32)
+    ; CHECK: G_STORE [[UV56]](s32), [[PTR_ADD58]](p5) :: (store 4 into %stack.0 + 224, align 256, addrspace 5)
+    ; CHECK: [[C59:%[0-9]+]]:_(s32) = G_CONSTANT i32 228
+    ; CHECK: [[PTR_ADD59:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C59]](s32)
+    ; CHECK: G_STORE [[UV57]](s32), [[PTR_ADD59]](p5) :: (store 4 into %stack.0 + 228, align 256, addrspace 5)
+    ; CHECK: [[C60:%[0-9]+]]:_(s32) = G_CONSTANT i32 232
+    ; CHECK: [[PTR_ADD60:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C60]](s32)
+    ; CHECK: G_STORE [[UV58]](s32), [[PTR_ADD60]](p5) :: (store 4 into %stack.0 + 232, align 256, addrspace 5)
+    ; CHECK: [[C61:%[0-9]+]]:_(s32) = G_CONSTANT i32 236
+    ; CHECK: [[PTR_ADD61:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C61]](s32)
+    ; CHECK: G_STORE [[UV59]](s32), [[PTR_ADD61]](p5) :: (store 4 into %stack.0 + 236, align 256, addrspace 5)
+    ; CHECK: [[C62:%[0-9]+]]:_(s32) = G_CONSTANT i32 240
+    ; CHECK: [[PTR_ADD62:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C62]](s32)
+    ; CHECK: G_STORE [[UV60]](s32), [[PTR_ADD62]](p5) :: (store 4 into %stack.0 + 240, align 256, addrspace 5)
+    ; CHECK: [[C63:%[0-9]+]]:_(s32) = G_CONSTANT i32 244
+    ; CHECK: [[PTR_ADD63:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C63]](s32)
+    ; CHECK: G_STORE [[UV61]](s32), [[PTR_ADD63]](p5) :: (store 4 into %stack.0 + 244, align 256, addrspace 5)
+    ; CHECK: [[C64:%[0-9]+]]:_(s32) = G_CONSTANT i32 248
+    ; CHECK: [[PTR_ADD64:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C64]](s32)
+    ; CHECK: G_STORE [[UV62]](s32), [[PTR_ADD64]](p5) :: (store 4 into %stack.0 + 248, align 256, addrspace 5)
+    ; CHECK: [[C65:%[0-9]+]]:_(s32) = G_CONSTANT i32 252
+    ; CHECK: [[PTR_ADD65:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C65]](s32)
+    ; CHECK: G_STORE [[UV63]](s32), [[PTR_ADD65]](p5) :: (store 4 into %stack.0 + 252, align 256, addrspace 5)
+    ; CHECK: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD9]](p5) :: (load 4 from %stack.0 + 28, addrspace 5)
+    ; CHECK: S_ENDPGM 0, implicit [[LOAD4]](s32)
+    %0:_(p1) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_CONSTANT i32 7
+    %2:_(<64 x s32>) = G_LOAD %0 :: (load 256, align 4, addrspace 4)
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %2, %1
+    S_ENDPGM 0, implicit %3
+...
+
+---
+name: extract_vector_elt_varidx_v64s32
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1, $sgpr2
+
+    ; CHECK-LABEL: name: extract_vector_elt_varidx_v64s32
+    ; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr2
+    ; CHECK: [[LOAD:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[COPY]](p1) :: (load 64, align 4, addrspace 4)
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 64
+    ; CHECK: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C]](s64)
+    ; CHECK: [[LOAD1:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD]](p1) :: (load 64 + 64, align 4, addrspace 4)
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 128
+    ; CHECK: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C1]](s64)
+    ; CHECK: [[LOAD2:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD1]](p1) :: (load 64 + 128, align 4, addrspace 4)
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 192
+    ; CHECK: [[PTR_ADD2:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; CHECK: [[LOAD3:%[0-9]+]]:_(<16 x s32>) = G_LOAD [[PTR_ADD2]](p1) :: (load 64 + 192, align 4, addrspace 4)
+    ; CHECK: [[FRAME_INDEX:%[0-9]+]]:_(p5) = G_FRAME_INDEX %stack.0
+    ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32), [[UV12:%[0-9]+]]:_(s32), [[UV13:%[0-9]+]]:_(s32), [[UV14:%[0-9]+]]:_(s32), [[UV15:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD]](<16 x s32>)
+    ; CHECK: [[UV16:%[0-9]+]]:_(s32), [[UV17:%[0-9]+]]:_(s32), [[UV18:%[0-9]+]]:_(s32), [[UV19:%[0-9]+]]:_(s32), [[UV20:%[0-9]+]]:_(s32), [[UV21:%[0-9]+]]:_(s32), [[UV22:%[0-9]+]]:_(s32), [[UV23:%[0-9]+]]:_(s32), [[UV24:%[0-9]+]]:_(s32), [[UV25:%[0-9]+]]:_(s32), [[UV26:%[0-9]+]]:_(s32), [[UV27:%[0-9]+]]:_(s32), [[UV28:%[0-9]+]]:_(s32), [[UV29:%[0-9]+]]:_(s32), [[UV30:%[0-9]+]]:_(s32), [[UV31:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD1]](<16 x s32>)
+    ; CHECK: [[UV32:%[0-9]+]]:_(s32), [[UV33:%[0-9]+]]:_(s32), [[UV34:%[0-9]+]]:_(s32), [[UV35:%[0-9]+]]:_(s32), [[UV36:%[0-9]+]]:_(s32), [[UV37:%[0-9]+]]:_(s32), [[UV38:%[0-9]+]]:_(s32), [[UV39:%[0-9]+]]:_(s32), [[UV40:%[0-9]+]]:_(s32), [[UV41:%[0-9]+]]:_(s32), [[UV42:%[0-9]+]]:_(s32), [[UV43:%[0-9]+]]:_(s32), [[UV44:%[0-9]+]]:_(s32), [[UV45:%[0-9]+]]:_(s32), [[UV46:%[0-9]+]]:_(s32), [[UV47:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD2]](<16 x s32>)
+    ; CHECK: [[UV48:%[0-9]+]]:_(s32), [[UV49:%[0-9]+]]:_(s32), [[UV50:%[0-9]+]]:_(s32), [[UV51:%[0-9]+]]:_(s32), [[UV52:%[0-9]+]]:_(s32), [[UV53:%[0-9]+]]:_(s32), [[UV54:%[0-9]+]]:_(s32), [[UV55:%[0-9]+]]:_(s32), [[UV56:%[0-9]+]]:_(s32), [[UV57:%[0-9]+]]:_(s32), [[UV58:%[0-9]+]]:_(s32), [[UV59:%[0-9]+]]:_(s32), [[UV60:%[0-9]+]]:_(s32), [[UV61:%[0-9]+]]:_(s32), [[UV62:%[0-9]+]]:_(s32), [[UV63:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[LOAD3]](<16 x s32>)
+    ; CHECK: G_STORE [[UV]](s32), [[FRAME_INDEX]](p5) :: (store 4 into %stack.0, align 256, addrspace 5)
+    ; CHECK: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
+    ; CHECK: [[PTR_ADD3:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C3]](s32)
+    ; CHECK: G_STORE [[UV1]](s32), [[PTR_ADD3]](p5) :: (store 4 into %stack.0 + 4, align 256, addrspace 5)
+    ; CHECK: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CHECK: [[PTR_ADD4:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C4]](s32)
+    ; CHECK: G_STORE [[UV2]](s32), [[PTR_ADD4]](p5) :: (store 4 into %stack.0 + 8, align 256, addrspace 5)
+    ; CHECK: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
+    ; CHECK: [[PTR_ADD5:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C5]](s32)
+    ; CHECK: G_STORE [[UV3]](s32), [[PTR_ADD5]](p5) :: (store 4 into %stack.0 + 12, align 256, addrspace 5)
+    ; CHECK: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[PTR_ADD6:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C6]](s32)
+    ; CHECK: G_STORE [[UV4]](s32), [[PTR_ADD6]](p5) :: (store 4 into %stack.0 + 16, align 256, addrspace 5)
+    ; CHECK: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 20
+    ; CHECK: [[PTR_ADD7:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C7]](s32)
+    ; CHECK: G_STORE [[UV5]](s32), [[PTR_ADD7]](p5) :: (store 4 into %stack.0 + 20, align 256, addrspace 5)
+    ; CHECK: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CHECK: [[PTR_ADD8:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C8]](s32)
+    ; CHECK: G_STORE [[UV6]](s32), [[PTR_ADD8]](p5) :: (store 4 into %stack.0 + 24, align 256, addrspace 5)
+    ; CHECK: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 28
+    ; CHECK: [[PTR_ADD9:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C9]](s32)
+    ; CHECK: G_STORE [[UV7]](s32), [[PTR_ADD9]](p5) :: (store 4 into %stack.0 + 28, align 256, addrspace 5)
+    ; CHECK: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 32
+    ; CHECK: [[PTR_ADD10:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C10]](s32)
+    ; CHECK: G_STORE [[UV8]](s32), [[PTR_ADD10]](p5) :: (store 4 into %stack.0 + 32, align 256, addrspace 5)
+    ; CHECK: [[C11:%[0-9]+]]:_(s32) = G_CONSTANT i32 36
+    ; CHECK: [[PTR_ADD11:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C11]](s32)
+    ; CHECK: G_STORE [[UV9]](s32), [[PTR_ADD11]](p5) :: (store 4 into %stack.0 + 36, align 256, addrspace 5)
+    ; CHECK: [[C12:%[0-9]+]]:_(s32) = G_CONSTANT i32 40
+    ; CHECK: [[PTR_ADD12:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C12]](s32)
+    ; CHECK: G_STORE [[UV10]](s32), [[PTR_ADD12]](p5) :: (store 4 into %stack.0 + 40, align 256, addrspace 5)
+    ; CHECK: [[C13:%[0-9]+]]:_(s32) = G_CONSTANT i32 44
+    ; CHECK: [[PTR_ADD13:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C13]](s32)
+    ; CHECK: G_STORE [[UV11]](s32), [[PTR_ADD13]](p5) :: (store 4 into %stack.0 + 44, align 256, addrspace 5)
+    ; CHECK: [[C14:%[0-9]+]]:_(s32) = G_CONSTANT i32 48
+    ; CHECK: [[PTR_ADD14:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C14]](s32)
+    ; CHECK: G_STORE [[UV12]](s32), [[PTR_ADD14]](p5) :: (store 4 into %stack.0 + 48, align 256, addrspace 5)
+    ; CHECK: [[C15:%[0-9]+]]:_(s32) = G_CONSTANT i32 52
+    ; CHECK: [[PTR_ADD15:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C15]](s32)
+    ; CHECK: G_STORE [[UV13]](s32), [[PTR_ADD15]](p5) :: (store 4 into %stack.0 + 52, align 256, addrspace 5)
+    ; CHECK: [[C16:%[0-9]+]]:_(s32) = G_CONSTANT i32 56
+    ; CHECK: [[PTR_ADD16:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C16]](s32)
+    ; CHECK: G_STORE [[UV14]](s32), [[PTR_ADD16]](p5) :: (store 4 into %stack.0 + 56, align 256, addrspace 5)
+    ; CHECK: [[C17:%[0-9]+]]:_(s32) = G_CONSTANT i32 60
+    ; CHECK: [[PTR_ADD17:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C17]](s32)
+    ; CHECK: G_STORE [[UV15]](s32), [[PTR_ADD17]](p5) :: (store 4 into %stack.0 + 60, align 256, addrspace 5)
+    ; CHECK: [[C18:%[0-9]+]]:_(s32) = G_CONSTANT i32 64
+    ; CHECK: [[PTR_ADD18:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C18]](s32)
+    ; CHECK: G_STORE [[UV16]](s32), [[PTR_ADD18]](p5) :: (store 4 into %stack.0 + 64, align 256, addrspace 5)
+    ; CHECK: [[C19:%[0-9]+]]:_(s32) = G_CONSTANT i32 68
+    ; CHECK: [[PTR_ADD19:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C19]](s32)
+    ; CHECK: G_STORE [[UV17]](s32), [[PTR_ADD19]](p5) :: (store 4 into %stack.0 + 68, align 256, addrspace 5)
+    ; CHECK: [[C20:%[0-9]+]]:_(s32) = G_CONSTANT i32 72
+    ; CHECK: [[PTR_ADD20:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C20]](s32)
+    ; CHECK: G_STORE [[UV18]](s32), [[PTR_ADD20]](p5) :: (store 4 into %stack.0 + 72, align 256, addrspace 5)
+    ; CHECK: [[C21:%[0-9]+]]:_(s32) = G_CONSTANT i32 76
+    ; CHECK: [[PTR_ADD21:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C21]](s32)
+    ; CHECK: G_STORE [[UV19]](s32), [[PTR_ADD21]](p5) :: (store 4 into %stack.0 + 76, align 256, addrspace 5)
+    ; CHECK: [[C22:%[0-9]+]]:_(s32) = G_CONSTANT i32 80
+    ; CHECK: [[PTR_ADD22:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C22]](s32)
+    ; CHECK: G_STORE [[UV20]](s32), [[PTR_ADD22]](p5) :: (store 4 into %stack.0 + 80, align 256, addrspace 5)
+    ; CHECK: [[C23:%[0-9]+]]:_(s32) = G_CONSTANT i32 84
+    ; CHECK: [[PTR_ADD23:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C23]](s32)
+    ; CHECK: G_STORE [[UV21]](s32), [[PTR_ADD23]](p5) :: (store 4 into %stack.0 + 84, align 256, addrspace 5)
+    ; CHECK: [[C24:%[0-9]+]]:_(s32) = G_CONSTANT i32 88
+    ; CHECK: [[PTR_ADD24:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C24]](s32)
+    ; CHECK: G_STORE [[UV22]](s32), [[PTR_ADD24]](p5) :: (store 4 into %stack.0 + 88, align 256, addrspace 5)
+    ; CHECK: [[C25:%[0-9]+]]:_(s32) = G_CONSTANT i32 92
+    ; CHECK: [[PTR_ADD25:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C25]](s32)
+    ; CHECK: G_STORE [[UV23]](s32), [[PTR_ADD25]](p5) :: (store 4 into %stack.0 + 92, align 256, addrspace 5)
+    ; CHECK: [[C26:%[0-9]+]]:_(s32) = G_CONSTANT i32 96
+    ; CHECK: [[PTR_ADD26:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C26]](s32)
+    ; CHECK: G_STORE [[UV24]](s32), [[PTR_ADD26]](p5) :: (store 4 into %stack.0 + 96, align 256, addrspace 5)
+    ; CHECK: [[C27:%[0-9]+]]:_(s32) = G_CONSTANT i32 100
+    ; CHECK: [[PTR_ADD27:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C27]](s32)
+    ; CHECK: G_STORE [[UV25]](s32), [[PTR_ADD27]](p5) :: (store 4 into %stack.0 + 100, align 256, addrspace 5)
+    ; CHECK: [[C28:%[0-9]+]]:_(s32) = G_CONSTANT i32 104
+    ; CHECK: [[PTR_ADD28:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C28]](s32)
+    ; CHECK: G_STORE [[UV26]](s32), [[PTR_ADD28]](p5) :: (store 4 into %stack.0 + 104, align 256, addrspace 5)
+    ; CHECK: [[C29:%[0-9]+]]:_(s32) = G_CONSTANT i32 108
+    ; CHECK: [[PTR_ADD29:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C29]](s32)
+    ; CHECK: G_STORE [[UV27]](s32), [[PTR_ADD29]](p5) :: (store 4 into %stack.0 + 108, align 256, addrspace 5)
+    ; CHECK: [[C30:%[0-9]+]]:_(s32) = G_CONSTANT i32 112
+    ; CHECK: [[PTR_ADD30:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C30]](s32)
+    ; CHECK: G_STORE [[UV28]](s32), [[PTR_ADD30]](p5) :: (store 4 into %stack.0 + 112, align 256, addrspace 5)
+    ; CHECK: [[C31:%[0-9]+]]:_(s32) = G_CONSTANT i32 116
+    ; CHECK: [[PTR_ADD31:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C31]](s32)
+    ; CHECK: G_STORE [[UV29]](s32), [[PTR_ADD31]](p5) :: (store 4 into %stack.0 + 116, align 256, addrspace 5)
+    ; CHECK: [[C32:%[0-9]+]]:_(s32) = G_CONSTANT i32 120
+    ; CHECK: [[PTR_ADD32:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C32]](s32)
+    ; CHECK: G_STORE [[UV30]](s32), [[PTR_ADD32]](p5) :: (store 4 into %stack.0 + 120, align 256, addrspace 5)
+    ; CHECK: [[C33:%[0-9]+]]:_(s32) = G_CONSTANT i32 124
+    ; CHECK: [[PTR_ADD33:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C33]](s32)
+    ; CHECK: G_STORE [[UV31]](s32), [[PTR_ADD33]](p5) :: (store 4 into %stack.0 + 124, align 256, addrspace 5)
+    ; CHECK: [[C34:%[0-9]+]]:_(s32) = G_CONSTANT i32 128
+    ; CHECK: [[PTR_ADD34:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C34]](s32)
+    ; CHECK: G_STORE [[UV32]](s32), [[PTR_ADD34]](p5) :: (store 4 into %stack.0 + 128, align 256, addrspace 5)
+    ; CHECK: [[C35:%[0-9]+]]:_(s32) = G_CONSTANT i32 132
+    ; CHECK: [[PTR_ADD35:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C35]](s32)
+    ; CHECK: G_STORE [[UV33]](s32), [[PTR_ADD35]](p5) :: (store 4 into %stack.0 + 132, align 256, addrspace 5)
+    ; CHECK: [[C36:%[0-9]+]]:_(s32) = G_CONSTANT i32 136
+    ; CHECK: [[PTR_ADD36:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C36]](s32)
+    ; CHECK: G_STORE [[UV34]](s32), [[PTR_ADD36]](p5) :: (store 4 into %stack.0 + 136, align 256, addrspace 5)
+    ; CHECK: [[C37:%[0-9]+]]:_(s32) = G_CONSTANT i32 140
+    ; CHECK: [[PTR_ADD37:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C37]](s32)
+    ; CHECK: G_STORE [[UV35]](s32), [[PTR_ADD37]](p5) :: (store 4 into %stack.0 + 140, align 256, addrspace 5)
+    ; CHECK: [[C38:%[0-9]+]]:_(s32) = G_CONSTANT i32 144
+    ; CHECK: [[PTR_ADD38:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C38]](s32)
+    ; CHECK: G_STORE [[UV36]](s32), [[PTR_ADD38]](p5) :: (store 4 into %stack.0 + 144, align 256, addrspace 5)
+    ; CHECK: [[C39:%[0-9]+]]:_(s32) = G_CONSTANT i32 148
+    ; CHECK: [[PTR_ADD39:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C39]](s32)
+    ; CHECK: G_STORE [[UV37]](s32), [[PTR_ADD39]](p5) :: (store 4 into %stack.0 + 148, align 256, addrspace 5)
+    ; CHECK: [[C40:%[0-9]+]]:_(s32) = G_CONSTANT i32 152
+    ; CHECK: [[PTR_ADD40:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C40]](s32)
+    ; CHECK: G_STORE [[UV38]](s32), [[PTR_ADD40]](p5) :: (store 4 into %stack.0 + 152, align 256, addrspace 5)
+    ; CHECK: [[C41:%[0-9]+]]:_(s32) = G_CONSTANT i32 156
+    ; CHECK: [[PTR_ADD41:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C41]](s32)
+    ; CHECK: G_STORE [[UV39]](s32), [[PTR_ADD41]](p5) :: (store 4 into %stack.0 + 156, align 256, addrspace 5)
+    ; CHECK: [[C42:%[0-9]+]]:_(s32) = G_CONSTANT i32 160
+    ; CHECK: [[PTR_ADD42:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C42]](s32)
+    ; CHECK: G_STORE [[UV40]](s32), [[PTR_ADD42]](p5) :: (store 4 into %stack.0 + 160, align 256, addrspace 5)
+    ; CHECK: [[C43:%[0-9]+]]:_(s32) = G_CONSTANT i32 164
+    ; CHECK: [[PTR_ADD43:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C43]](s32)
+    ; CHECK: G_STORE [[UV41]](s32), [[PTR_ADD43]](p5) :: (store 4 into %stack.0 + 164, align 256, addrspace 5)
+    ; CHECK: [[C44:%[0-9]+]]:_(s32) = G_CONSTANT i32 168
+    ; CHECK: [[PTR_ADD44:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C44]](s32)
+    ; CHECK: G_STORE [[UV42]](s32), [[PTR_ADD44]](p5) :: (store 4 into %stack.0 + 168, align 256, addrspace 5)
+    ; CHECK: [[C45:%[0-9]+]]:_(s32) = G_CONSTANT i32 172
+    ; CHECK: [[PTR_ADD45:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C45]](s32)
+    ; CHECK: G_STORE [[UV43]](s32), [[PTR_ADD45]](p5) :: (store 4 into %stack.0 + 172, align 256, addrspace 5)
+    ; CHECK: [[C46:%[0-9]+]]:_(s32) = G_CONSTANT i32 176
+    ; CHECK: [[PTR_ADD46:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C46]](s32)
+    ; CHECK: G_STORE [[UV44]](s32), [[PTR_ADD46]](p5) :: (store 4 into %stack.0 + 176, align 256, addrspace 5)
+    ; CHECK: [[C47:%[0-9]+]]:_(s32) = G_CONSTANT i32 180
+    ; CHECK: [[PTR_ADD47:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C47]](s32)
+    ; CHECK: G_STORE [[UV45]](s32), [[PTR_ADD47]](p5) :: (store 4 into %stack.0 + 180, align 256, addrspace 5)
+    ; CHECK: [[C48:%[0-9]+]]:_(s32) = G_CONSTANT i32 184
+    ; CHECK: [[PTR_ADD48:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C48]](s32)
+    ; CHECK: G_STORE [[UV46]](s32), [[PTR_ADD48]](p5) :: (store 4 into %stack.0 + 184, align 256, addrspace 5)
+    ; CHECK: [[C49:%[0-9]+]]:_(s32) = G_CONSTANT i32 188
+    ; CHECK: [[PTR_ADD49:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C49]](s32)
+    ; CHECK: G_STORE [[UV47]](s32), [[PTR_ADD49]](p5) :: (store 4 into %stack.0 + 188, align 256, addrspace 5)
+    ; CHECK: [[C50:%[0-9]+]]:_(s32) = G_CONSTANT i32 192
+    ; CHECK: [[PTR_ADD50:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C50]](s32)
+    ; CHECK: G_STORE [[UV48]](s32), [[PTR_ADD50]](p5) :: (store 4 into %stack.0 + 192, align 256, addrspace 5)
+    ; CHECK: [[C51:%[0-9]+]]:_(s32) = G_CONSTANT i32 196
+    ; CHECK: [[PTR_ADD51:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C51]](s32)
+    ; CHECK: G_STORE [[UV49]](s32), [[PTR_ADD51]](p5) :: (store 4 into %stack.0 + 196, align 256, addrspace 5)
+    ; CHECK: [[C52:%[0-9]+]]:_(s32) = G_CONSTANT i32 200
+    ; CHECK: [[PTR_ADD52:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C52]](s32)
+    ; CHECK: G_STORE [[UV50]](s32), [[PTR_ADD52]](p5) :: (store 4 into %stack.0 + 200, align 256, addrspace 5)
+    ; CHECK: [[C53:%[0-9]+]]:_(s32) = G_CONSTANT i32 204
+    ; CHECK: [[PTR_ADD53:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C53]](s32)
+    ; CHECK: G_STORE [[UV51]](s32), [[PTR_ADD53]](p5) :: (store 4 into %stack.0 + 204, align 256, addrspace 5)
+    ; CHECK: [[C54:%[0-9]+]]:_(s32) = G_CONSTANT i32 208
+    ; CHECK: [[PTR_ADD54:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C54]](s32)
+    ; CHECK: G_STORE [[UV52]](s32), [[PTR_ADD54]](p5) :: (store 4 into %stack.0 + 208, align 256, addrspace 5)
+    ; CHECK: [[C55:%[0-9]+]]:_(s32) = G_CONSTANT i32 212
+    ; CHECK: [[PTR_ADD55:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C55]](s32)
+    ; CHECK: G_STORE [[UV53]](s32), [[PTR_ADD55]](p5) :: (store 4 into %stack.0 + 212, align 256, addrspace 5)
+    ; CHECK: [[C56:%[0-9]+]]:_(s32) = G_CONSTANT i32 216
+    ; CHECK: [[PTR_ADD56:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C56]](s32)
+    ; CHECK: G_STORE [[UV54]](s32), [[PTR_ADD56]](p5) :: (store 4 into %stack.0 + 216, align 256, addrspace 5)
+    ; CHECK: [[C57:%[0-9]+]]:_(s32) = G_CONSTANT i32 220
+    ; CHECK: [[PTR_ADD57:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C57]](s32)
+    ; CHECK: G_STORE [[UV55]](s32), [[PTR_ADD57]](p5) :: (store 4 into %stack.0 + 220, align 256, addrspace 5)
+    ; CHECK: [[C58:%[0-9]+]]:_(s32) = G_CONSTANT i32 224
+    ; CHECK: [[PTR_ADD58:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C58]](s32)
+    ; CHECK: G_STORE [[UV56]](s32), [[PTR_ADD58]](p5) :: (store 4 into %stack.0 + 224, align 256, addrspace 5)
+    ; CHECK: [[C59:%[0-9]+]]:_(s32) = G_CONSTANT i32 228
+    ; CHECK: [[PTR_ADD59:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C59]](s32)
+    ; CHECK: G_STORE [[UV57]](s32), [[PTR_ADD59]](p5) :: (store 4 into %stack.0 + 228, align 256, addrspace 5)
+    ; CHECK: [[C60:%[0-9]+]]:_(s32) = G_CONSTANT i32 232
+    ; CHECK: [[PTR_ADD60:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C60]](s32)
+    ; CHECK: G_STORE [[UV58]](s32), [[PTR_ADD60]](p5) :: (store 4 into %stack.0 + 232, align 256, addrspace 5)
+    ; CHECK: [[C61:%[0-9]+]]:_(s32) = G_CONSTANT i32 236
+    ; CHECK: [[PTR_ADD61:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C61]](s32)
+    ; CHECK: G_STORE [[UV59]](s32), [[PTR_ADD61]](p5) :: (store 4 into %stack.0 + 236, align 256, addrspace 5)
+    ; CHECK: [[C62:%[0-9]+]]:_(s32) = G_CONSTANT i32 240
+    ; CHECK: [[PTR_ADD62:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C62]](s32)
+    ; CHECK: G_STORE [[UV60]](s32), [[PTR_ADD62]](p5) :: (store 4 into %stack.0 + 240, align 256, addrspace 5)
+    ; CHECK: [[C63:%[0-9]+]]:_(s32) = G_CONSTANT i32 244
+    ; CHECK: [[PTR_ADD63:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C63]](s32)
+    ; CHECK: G_STORE [[UV61]](s32), [[PTR_ADD63]](p5) :: (store 4 into %stack.0 + 244, align 256, addrspace 5)
+    ; CHECK: [[C64:%[0-9]+]]:_(s32) = G_CONSTANT i32 248
+    ; CHECK: [[PTR_ADD64:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C64]](s32)
+    ; CHECK: G_STORE [[UV62]](s32), [[PTR_ADD64]](p5) :: (store 4 into %stack.0 + 248, align 256, addrspace 5)
+    ; CHECK: [[C65:%[0-9]+]]:_(s32) = G_CONSTANT i32 252
+    ; CHECK: [[PTR_ADD65:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[C65]](s32)
+    ; CHECK: G_STORE [[UV63]](s32), [[PTR_ADD65]](p5) :: (store 4 into %stack.0 + 252, align 256, addrspace 5)
+    ; CHECK: [[C66:%[0-9]+]]:_(s32) = G_CONSTANT i32 63
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C66]]
+    ; CHECK: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[AND]], [[C3]]
+    ; CHECK: [[PTR_ADD66:%[0-9]+]]:_(p5) = G_PTR_ADD [[FRAME_INDEX]], [[MUL]](s32)
+    ; CHECK: [[LOAD4:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD66]](p5) :: (load 4, addrspace 5)
+    ; CHECK: S_ENDPGM 0, implicit [[LOAD4]](s32)
+    %0:_(p1) = COPY $sgpr0_sgpr1
+    %1:_(s32) = COPY $sgpr2
+    %2:_(<64 x s32>) = G_LOAD %0 :: (load 256, align 4, addrspace 4)
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %2, %1
+    S_ENDPGM 0, implicit %3
+...