AMDGPU/GlobalISel: cmp/select method for extract element

[rampitec]

Pre-commited test and rebased.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUPostLegalizerCombiner.cpp

230–231 ↗

(On Diff #266978)

For the 2 element case, we should do this in the pre-legalizer combiner (I guess the 16-bit case code would be different, since it would be bitcast and shift)

252 ↗

(On Diff #266978)

Register Vec = ...?

258 ↗

(On Diff #266978)

It would be better to produce a single G_UNMERGE_VALUES and index out of the result list than to produce a sequence of G_EXTRACTs.

Also FYI I'm planning on making 16-bit vector indexing bitcast promote to 32-bit vectors, so this would only see 32 and 64-bit element vectors

264–267 ↗

(On Diff #266978)

You can fold these like
auto B = buildConstant(LLT::scalar(32), ...)
auto Cmp = B.buildICmp(LLT::scalar(1), CmpInst::...)

268 ↗

(On Diff #266978)

Weird spacing, NumElem - 1?

[rampitec]

llvm/lib/Target/AMDGPU/AMDGPUPostLegalizerCombiner.cpp

230–231 ↗

(On Diff #266978)

Why is that? The problem is we need to know if index is divergent or not in general, so we avoid creating loops. I assume if we need some special case for 2 elements, that shall be a separate combine.

258 ↗

(On Diff #266978)

Thanks. Apparently there is a bug in the CombinerHelper::matchEqualDefs() when it works with an opcode producing multiple values. I will have to fix it first.

[rampitec]

Addressed review comments.
It also contains fix for CombinerHelper::matchEqualDefs(), it likely needs a separate change.

[rampitec]

Extracted CombinerHelper::matchEqualDefs() fix into D80767.

[arsenm]

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

6967–6968 ↗

(On Diff #267067)

This kind of thing belongs in TargetLowering still, it just needs to be made public

[rampitec]

Moved predicate function back to TLI, it is static anyway.

[rampitec]

Moved to the RegBankSelect itself. It is too late to do it past RegBankSelect because waterfall loops are already created, and it is too early to do it before because we need divergence info.

[rampitec]

Changed scalar condition type to S32.

[arsenm]

LGTM. You could use the ApplyRegBankMapping observer to avoid all the explicit setRegBankCalls, but not everywhere is consistently using it yet anyway

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h

public:
   const RegisterBank &getRegBankFromRegClass(const TargetRegisterClass &RC,
                                              LLT) const override;
 
   InstructionMappings
   getInstrAlternativeMappings(const MachineInstr &MI) const override;
 
   const InstructionMapping &
   getInstrMapping(const MachineInstr &MI) const override;
+
+private:
+
+  bool foldExtractEltToCmpSelect(MachineInstr &MI,
+                                 MachineRegisterInfo &MRI,
+                                 const OperandsMapper &OpdMapper) const;
 };
 } // End llvm namespace.
 #endif

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

if (IsBooleanSrc) {
       B.buildAShr(Hi32Reg, Lo32Reg, ShiftAmt);
     }
   } else {
     assert(ExtOpc == AMDGPU::G_ANYEXT && "not an integer extension");
     B.buildUndef(Hi32Reg);
   }
 }
 
+bool AMDGPURegisterBankInfo::foldExtractEltToCmpSelect(
+  MachineInstr &MI, MachineRegisterInfo &MRI,
+  const OperandsMapper &OpdMapper) const {
+
+  Register VecReg = MI.getOperand(1).getReg();
+  Register Idx = MI.getOperand(2).getReg();
+
+  const RegisterBank &IdxBank =
+    *OpdMapper.getInstrMapping().getOperandMapping(2).BreakDown[0].RegBank;
+
+  bool IsDivergentIdx = IdxBank == AMDGPU::VGPRRegBank;
+
+  LLT VecTy = MRI.getType(VecReg);
+  unsigned EltSize = VecTy.getScalarSizeInBits();
+  unsigned NumElem = VecTy.getNumElements();
+
+  if (!SITargetLowering::shouldExpandVectorDynExt(EltSize, NumElem,
+                                                  IsDivergentIdx))
+    return false;
+
+  MachineIRBuilder B(MI);
+  LLT S32 = LLT::scalar(32);
+
+  const RegisterBank &DstBank =
+    *OpdMapper.getInstrMapping().getOperandMapping(0).BreakDown[0].RegBank;
+  const RegisterBank &SrcBank =
+    *OpdMapper.getInstrMapping().getOperandMapping(1).BreakDown[0].RegBank;
+
+  const RegisterBank &CCBank =
+    (DstBank == AMDGPU::SGPRRegBank &&
+     SrcBank == AMDGPU::SGPRRegBank &&
+     IdxBank == AMDGPU::SGPRRegBank) ? AMDGPU::SGPRRegBank
+                                     : AMDGPU::VCCRegBank;
+  LLT CCTy = (CCBank == AMDGPU::SGPRRegBank) ? S32 : LLT::scalar(1);
+
+  if (CCBank == AMDGPU::VCCRegBank && IdxBank == AMDGPU::SGPRRegBank) {
+    Idx = B.buildCopy(S32, Idx)->getOperand(0).getReg();
+    MRI.setRegBank(Idx, AMDGPU::VGPRRegBank);
+  }
+
+  LLT EltTy = VecTy.getScalarType();
+  SmallVector<Register, 2> DstRegs(OpdMapper.getVRegs(0));
+  unsigned NumLanes = DstRegs.size();
+  if (!NumLanes)
+    NumLanes = 1;
+  else
+    EltTy = MRI.getType(DstRegs[0]);
+
+  auto UnmergeToEltTy = B.buildUnmerge(EltTy, VecReg);
+  SmallVector<Register, 2> Res(NumLanes);
+  for (unsigned L = 0; L < NumLanes; ++L)
+    Res[L] = UnmergeToEltTy.getReg(L);
+
+  for (unsigned I = 1; I < NumElem; ++I) {
+    auto IC = B.buildConstant(S32, I);
+    MRI.setRegBank(IC->getOperand(0).getReg(), AMDGPU::SGPRRegBank);
+    auto Cmp = B.buildICmp(CmpInst::ICMP_EQ, CCTy, Idx, IC);
+    MRI.setRegBank(Cmp->getOperand(0).getReg(), CCBank);
+
+    for (unsigned L = 0; L < NumLanes; ++L) {
+      auto S = B.buildSelect(EltTy, Cmp,
+                             UnmergeToEltTy.getReg(I * NumLanes + L), Res[L]);
+
+      for (unsigned N : { 0, 2, 3 })
+        MRI.setRegBank(S->getOperand(N).getReg(), DstBank);
+
+      Res[L] = S->getOperand(0).getReg();
+    }
+  }
+
+  for (unsigned L = 0; L < NumLanes; ++L) {
+    Register DstReg = (NumLanes == 1) ? MI.getOperand(0).getReg() : DstRegs[L];
+    B.buildCopy(DstReg, Res[L]);
+    MRI.setRegBank(DstReg, DstBank);
+  }
+
+  MRI.setRegBank(MI.getOperand(0).getReg(), DstBank);
+  MI.eraseFromParent();
+
+  return true;
+}
+
 void AMDGPURegisterBankInfo::applyMappingImpl(
     const OperandsMapper &OpdMapper) const {
   MachineInstr &MI = OpdMapper.getMI();
   unsigned Opc = MI.getOpcode();
   MachineRegisterInfo &MRI = OpdMapper.getMRI();
   switch (Opc) {
   case AMDGPU::G_PHI: {
     Register DstReg = MI.getOperand(0).getReg();
case AMDGPU::G_EXTRACT_VECTOR_ELT: {
 
     Register DstReg = MI.getOperand(0).getReg();
     Register SrcReg = MI.getOperand(1).getReg();
 
     const LLT S32 = LLT::scalar(32);
     LLT DstTy = MRI.getType(DstReg);
     LLT SrcTy = MRI.getType(SrcReg);
 
+    if (foldExtractEltToCmpSelect(MI, MRI, OpdMapper))
+      return;
+
     MachineIRBuilder B(MI);
 
     const ValueMapping &DstMapping
       = OpdMapper.getInstrMapping().getOperandMapping(0);
     const RegisterBank *DstBank = DstMapping.BreakDown[0].RegBank;
     const RegisterBank *SrcBank =
       OpdMapper.getInstrMapping().getOperandMapping(1).BreakDown[0].RegBank;
     const RegisterBank *IdxBank =

llvm/lib/Target/AMDGPU/SIISelLowering.h

public:
   /// \returns True if PC-relative relocation needs to be emitted for given
   /// global value \p GV, false otherwise.
   bool shouldEmitPCReloc(const GlobalValue *GV) const;
 
   /// \returns true if this should use a literal constant for an LDS address,
   /// and not emit a relocation for an LDS global.
   bool shouldUseLDSConstAddress(const GlobalValue *GV) const;
 
+  /// Check if EXTRACT_VECTOR_ELT/INSERT_VECTOR_ELT (<n x e>, var-idx) should be
+  /// expanded into a set of cmp/select instructions.
+  static bool shouldExpandVectorDynExt(unsigned EltSize, unsigned NumElem,
+                                       bool IsDivergentIdx);
+
 private:
   // Analyze a combined offset from an amdgcn_buffer_ intrinsic and store the
   // three offsets (voffset, soffset and instoffset) into the SDValue[3] array
   // pointed to by Offsets.
   /// \returns 0 If there is a non-constant offset or if the offset is 0.
   /// Otherwise returns the constant offset.
   unsigned setBufferOffsets(SDValue CombinedOffset, SelectionDAG &DAG,
                            SDValue *Offsets, unsigned Align = 4) const;

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

SDValue SITargetLowering::performCvtPkRTZCombine(SDNode *N,
   SDValue Src1 = N->getOperand(1);
   if (Src0.isUndef() && Src1.isUndef())
     return DCI.DAG.getUNDEF(N->getValueType(0));
   return SDValue();
 }
 
 // Check if EXTRACT_VECTOR_ELT/INSERT_VECTOR_ELT (<n x e>, var-idx) should be
 // expanded into a set of cmp/select instructions.
-static bool shouldExpandVectorDynExt(SDNode *N) {
-  SDValue Idx = N->getOperand(N->getNumOperands() - 1);
-  if (UseDivergentRegisterIndexing || isa<ConstantSDNode>(Idx))
+bool SITargetLowering::shouldExpandVectorDynExt(unsigned EltSize,
+                                                unsigned NumElem,
+                                                bool IsDivergentIdx) {
+  if (UseDivergentRegisterIndexing)
     return false;
 
-  SDValue Vec = N->getOperand(0);
-  EVT VecVT = Vec.getValueType();
-  EVT EltVT = VecVT.getVectorElementType();
-  unsigned VecSize = VecVT.getSizeInBits();
-  unsigned EltSize = EltVT.getSizeInBits();
-  unsigned NumElem = VecVT.getVectorNumElements();
+  unsigned VecSize = EltSize * NumElem;
 
   // Sub-dword vectors of size 2 dword or less have better implementation.
   if (VecSize <= 64 && EltSize < 32)
     return false;
 
   // Always expand the rest of sub-dword instructions, otherwise it will be
   // lowered via memory.
   if (EltSize < 32)
     return true;
 
   // Always do this if var-idx is divergent, otherwise it will become a loop.
-  if (Idx->isDivergent())
+  if (IsDivergentIdx)
     return true;
 
   // Large vectors would yield too many compares and v_cndmask_b32 instructions.
   unsigned NumInsts = NumElem /* Number of compares */ +
                       ((EltSize + 31) / 32) * NumElem /* Number of cndmasks */;
   return NumInsts <= 16;
 }
 
+static bool shouldExpandVectorDynExt(SDNode *N) {
+  SDValue Idx = N->getOperand(N->getNumOperands() - 1);
+  if (isa<ConstantSDNode>(Idx))
+    return false;
+
+  SDValue Vec = N->getOperand(0);
+  EVT VecVT = Vec.getValueType();
+  EVT EltVT = VecVT.getVectorElementType();
+  unsigned EltSize = EltVT.getSizeInBits();
+  unsigned NumElem = VecVT.getVectorNumElements();
+
+  return SITargetLowering::shouldExpandVectorDynExt(EltSize, NumElem,
+                                                    Idx->isDivergent());
+}
+
 SDValue SITargetLowering::performExtractVectorEltCombine(
   SDNode *N, DAGCombinerInfo &DCI) const {
   SDValue Vec = N->getOperand(0);
   SelectionDAG &DAG = DCI.DAG;
 
   EVT VecVT = Vec.getValueType();
   EVT EltVT = VecVT.getVectorElementType();
 
if (Vec.hasOneUse() && DCI.isBeforeLegalize()) {
     }
     }
   }
 
   unsigned VecSize = VecVT.getSizeInBits();
   unsigned EltSize = EltVT.getSizeInBits();
 
   // EXTRACT_VECTOR_ELT (<n x e>, var-idx) => n x select (e, const-idx)
-  if (shouldExpandVectorDynExt(N)) {
+  if (::shouldExpandVectorDynExt(N)) {
     SDLoc SL(N);
     SDValue Idx = N->getOperand(1);
     SDValue V;
     for (unsigned I = 0, E = VecVT.getVectorNumElements(); I < E; ++I) {
       SDValue IC = DAG.getVectorIdxConstant(I, SL);
       SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, EltVT, Vec, IC);
       if (I == 0)
         V = Elt;
SITargetLowering::performInsertVectorEltCombine(SDNode *N,
                                                 DAGCombinerInfo &DCI) const {
   SDValue Vec = N->getOperand(0);
   SDValue Idx = N->getOperand(2);
   EVT VecVT = Vec.getValueType();
   EVT EltVT = VecVT.getVectorElementType();
 
   // INSERT_VECTOR_ELT (<n x e>, var-idx)
   // => BUILD_VECTOR n x select (e, const-idx)
-  if (!shouldExpandVectorDynExt(N))
+  if (!::shouldExpandVectorDynExt(N))
     return SDValue();
 
   SelectionDAG &DAG = DCI.DAG;
   SDLoc SL(N);
   SDValue Ins = N->getOperand(1);
   EVT IdxVT = Idx.getValueType();
 
   SmallVector<SDValue, 16> Ops;

llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GPRIDX %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,MOVREL %s
 
 define float @dyn_extract_v8f32_const_s_v(i32 %sel) {
 ; GCN-LABEL: dyn_extract_v8f32_const_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s11, 0x41000000
-; GCN-NEXT:    s_mov_b32 s10, 0x40e00000
-; GCN-NEXT:    s_mov_b32 s9, 0x40c00000
-; GCN-NEXT:    s_mov_b32 s8, 0x40a00000
-; GCN-NEXT:    s_mov_b32 s7, 4.0
-; GCN-NEXT:    s_mov_b32 s6, 0x40400000
-; GCN-NEXT:    s_mov_b32 s5, 2.0
-; GCN-NEXT:    s_mov_b32 s4, 1.0
-; GCN-NEXT:    s_mov_b64 s[12:13], exec
-; GCN-NEXT:  BB0_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s14, v0
-; GCN-NEXT:    s_mov_b32 m0, s14
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s14, v0
-; GCN-NEXT:    s_movrels_b32 s14, s4
-; GCN-NEXT:    v_mov_b32_e32 v1, s14
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB0_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[12:13]
-; GCN-NEXT:    v_mov_b32_e32 v0, v1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_cndmask_b32_e64 v6, 1.0, 2.0, vcc
+; GCN-NEXT:    v_mov_b32_e32 v1, 0x40400000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v6, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, 4.0, vcc
+; GCN-NEXT:    v_mov_b32_e32 v2, 0x40a00000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GCN-NEXT:    v_mov_b32_e32 v3, 0x40c00000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_mov_b32_e32 v4, 0x40e00000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_mov_b32_e32 v5, 0x41000000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v5, vcc
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <8 x float> <float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0, float 8.0>, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v8f32_const_s_s(i32 inreg %sel) {
 ; GCN-LABEL: dyn_extract_v8f32_const_s_s:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s4, 1.0
-; GCN-NEXT:    s_mov_b32 m0, s2
-; GCN-NEXT:    s_mov_b32 s11, 0x41000000
-; GCN-NEXT:    s_mov_b32 s10, 0x40e00000
-; GCN-NEXT:    s_mov_b32 s9, 0x40c00000
-; GCN-NEXT:    s_mov_b32 s8, 0x40a00000
-; GCN-NEXT:    s_mov_b32 s7, 4.0
-; GCN-NEXT:    s_mov_b32 s6, 0x40400000
-; GCN-NEXT:    s_mov_b32 s5, 2.0
-; GCN-NEXT:    s_movrels_b32 s0, s4
+; GCN-NEXT:    s_cmp_eq_u32 s2, 1
+; GCN-NEXT:    s_cselect_b32 s0, 2.0, 1.0
+; GCN-NEXT:    s_cmp_eq_u32 s2, 2
+; GCN-NEXT:    s_cselect_b32 s0, 0x40400000, s0
+; GCN-NEXT:    s_cmp_eq_u32 s2, 3
+; GCN-NEXT:    s_cselect_b32 s0, 4.0, s0
+; GCN-NEXT:    s_cmp_eq_u32 s2, 4
+; GCN-NEXT:    s_cselect_b32 s0, 0x40a00000, s0
+; GCN-NEXT:    s_cmp_eq_u32 s2, 5
+; GCN-NEXT:    s_cselect_b32 s0, 0x40c00000, s0
+; GCN-NEXT:    s_cmp_eq_u32 s2, 6
+; GCN-NEXT:    s_cselect_b32 s0, 0x40e00000, s0
+; GCN-NEXT:    s_cmp_eq_u32 s2, 7
+; GCN-NEXT:    s_cselect_b32 s0, 0x41000000, s0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <8 x float> <float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0, float 8.0>, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v8f32_s_v(<8 x float> inreg %vec, i32 %sel) {
 ; GCN-LABEL: dyn_extract_v8f32_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_mov_b32 s0, s2
 ; GCN-NEXT:    s_mov_b32 s1, s3
 ; GCN-NEXT:    s_mov_b32 s2, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s0
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
 ; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_mov_b32 s6, s8
-; GCN-NEXT:    s_mov_b32 s7, s9
-; GCN-NEXT:    s_mov_b64 s[8:9], exec
-; GCN-NEXT:  BB2_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s10, v0
-; GCN-NEXT:    s_mov_b32 m0, s10
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s10, v0
-; GCN-NEXT:    s_movrels_b32 s10, s0
-; GCN-NEXT:    v_mov_b32_e32 v1, s10
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB2_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[8:9]
-; GCN-NEXT:    v_mov_b32_e32 v0, v1
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GCN-NEXT:    v_mov_b32_e32 v3, s2
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_mov_b32_e32 v4, s3
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_mov_b32_e32 v5, s6
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_mov_b32_e32 v6, s7
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
+; GCN-NEXT:    v_mov_b32_e32 v7, s8
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_mov_b32_e32 v8, s9
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v8, vcc
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <8 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define float @dyn_extract_v8f32_v_v(<8 x float> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v8f32_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB3_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v8
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v8
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v9, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB3_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v9
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v8f32_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB3_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v8
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v8
-; MOVREL-NEXT:    v_movrels_b32_e32 v9, v0
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB3_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v9
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v8f32_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <8 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v8f32_v_s(<8 x float> %vec, i32 inreg %sel) {
-; GPRIDX-LABEL: dyn_extract_v8f32_v_s:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    ; return to shader part epilog
-;
-; MOVREL-LABEL: dyn_extract_v8f32_v_s:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_mov_b32 m0, s2
-; MOVREL-NEXT:    v_movrels_b32_e32 v0, v0
-; MOVREL-NEXT:    ; return to shader part epilog
+; GCN-LABEL: dyn_extract_v8f32_v_s:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 3
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 4
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 5
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
+; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <8 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v8f32_s_s(<8 x float> inreg %vec, i32 inreg %sel) {
 ; GCN-LABEL: dyn_extract_v8f32_s_s:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s0, s2
-; GCN-NEXT:    s_mov_b32 m0, s10
-; GCN-NEXT:    s_mov_b32 s1, s3
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_mov_b32 s6, s8
-; GCN-NEXT:    s_mov_b32 s7, s9
-; GCN-NEXT:    s_movrels_b32 s0, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 1
+; GCN-NEXT:    s_cselect_b32 s0, s3, s2
+; GCN-NEXT:    s_cmp_eq_u32 s10, 2
+; GCN-NEXT:    s_cselect_b32 s0, s4, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 3
+; GCN-NEXT:    s_cselect_b32 s0, s5, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 4
+; GCN-NEXT:    s_cselect_b32 s0, s6, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 5
+; GCN-NEXT:    s_cselect_b32 s0, s7, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 6
+; GCN-NEXT:    s_cselect_b32 s0, s8, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 7
+; GCN-NEXT:    s_cselect_b32 s0, s9, s0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <8 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define i64 @dyn_extract_v8i64_const_s_v(i32 %sel) {
 ; GCN-LABEL: dyn_extract_v8i64_const_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b64 s[18:19], 8
-; GCN-NEXT:    s_mov_b64 s[16:17], 7
-; GCN-NEXT:    s_mov_b64 s[14:15], 6
-; GCN-NEXT:    s_mov_b64 s[12:13], 5
-; GCN-NEXT:    s_mov_b64 s[10:11], 4
-; GCN-NEXT:    s_mov_b64 s[8:9], 3
-; GCN-NEXT:    s_mov_b64 s[6:7], 2
 ; GCN-NEXT:    s_mov_b64 s[4:5], 1
-; GCN-NEXT:    s_mov_b64 s[20:21], exec
-; GCN-NEXT:  BB6_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s22, v0
-; GCN-NEXT:    s_lshl_b32 m0, s22, 1
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s22, v0
-; GCN-NEXT:    s_movrels_b32 s22, s4
-; GCN-NEXT:    s_movrels_b32 s23, s5
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB6_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[20:21]
-; GCN-NEXT:    v_mov_b32_e32 v0, s22
-; GCN-NEXT:    v_mov_b32_e32 v1, s23
+; GCN-NEXT:    s_mov_b64 s[6:7], 2
+; GCN-NEXT:    s_mov_b64 s[8:9], 3
+; GCN-NEXT:    v_mov_b32_e32 v1, s4
+; GCN-NEXT:    v_mov_b32_e32 v2, s5
+; GCN-NEXT:    v_mov_b32_e32 v3, s6
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_mov_b32_e32 v4, s7
+; GCN-NEXT:    s_mov_b64 s[10:11], 4
+; GCN-NEXT:    v_mov_b32_e32 v5, s8
+; GCN-NEXT:    v_mov_b32_e32 v6, s9
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    s_mov_b64 s[12:13], 5
+; GCN-NEXT:    v_mov_b32_e32 v7, s10
+; GCN-NEXT:    v_mov_b32_e32 v8, s11
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    s_mov_b64 s[14:15], 6
+; GCN-NEXT:    v_mov_b32_e32 v9, s12
+; GCN-NEXT:    v_mov_b32_e32 v10, s13
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    s_mov_b64 s[16:17], 7
+; GCN-NEXT:    v_mov_b32_e32 v11, s14
+; GCN-NEXT:    v_mov_b32_e32 v12, s15
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v10, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    s_mov_b64 s[18:19], 8
+; GCN-NEXT:    v_mov_b32_e32 v13, s16
+; GCN-NEXT:    v_mov_b32_e32 v14, s17
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v12, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GCN-NEXT:    v_mov_b32_e32 v15, s18
+; GCN-NEXT:    v_mov_b32_e32 v16, s19
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v14, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v15, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v2, v16, vcc
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <8 x i64> <i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8>, i32 %sel
   ret i64 %ext
 }
 
 define amdgpu_ps void @dyn_extract_v8i64_const_s_s(i32 inreg %sel) {
 ; GPRIDX-LABEL: dyn_extract_v8i64_const_s_s:

 ; GPRIDX-LABEL: dyn_extract_v8i64_s_v:
 ; GPRIDX:       ; %bb.0: ; %entry
 ; GPRIDX-NEXT:    s_mov_b32 s0, s2
 ; GPRIDX-NEXT:    s_mov_b32 s1, s3
 ; GPRIDX-NEXT:    s_mov_b32 s2, s4
 ; GPRIDX-NEXT:    s_mov_b32 s3, s5
 ; GPRIDX-NEXT:    s_mov_b32 s4, s6
 ; GPRIDX-NEXT:    s_mov_b32 s5, s7
+; GPRIDX-NEXT:    v_mov_b32_e32 v1, s0
+; GPRIDX-NEXT:    v_mov_b32_e32 v2, s1
+; GPRIDX-NEXT:    v_mov_b32_e32 v3, s2
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GPRIDX-NEXT:    v_mov_b32_e32 v4, s3
 ; GPRIDX-NEXT:    s_mov_b32 s6, s8
 ; GPRIDX-NEXT:    s_mov_b32 s7, s9
+; GPRIDX-NEXT:    v_mov_b32_e32 v5, s4
+; GPRIDX-NEXT:    v_mov_b32_e32 v6, s5
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
 ; GPRIDX-NEXT:    s_mov_b32 s8, s10
 ; GPRIDX-NEXT:    s_mov_b32 s9, s11
+; GPRIDX-NEXT:    v_mov_b32_e32 v7, s6
+; GPRIDX-NEXT:    v_mov_b32_e32 v8, s7
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
 ; GPRIDX-NEXT:    s_mov_b32 s10, s12
 ; GPRIDX-NEXT:    s_mov_b32 s11, s13
-; GPRIDX-NEXT:    s_mov_b32 s12, s14
-; GPRIDX-NEXT:    s_mov_b32 s13, s15
-; GPRIDX-NEXT:    s_mov_b32 s14, s16
-; GPRIDX-NEXT:    s_mov_b32 s15, s17
-; GPRIDX-NEXT:    s_mov_b64 s[16:17], exec
-; GPRIDX-NEXT:  BB8_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s18, v0
-; GPRIDX-NEXT:    s_lshl_b32 m0, s18, 1
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s18, v0
-; GPRIDX-NEXT:    s_movrels_b32 s18, s0
-; GPRIDX-NEXT:    s_movrels_b32 s19, s1
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, s18
-; GPRIDX-NEXT:    v_mov_b32_e32 v2, s19
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB8_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[16:17]
-; GPRIDX-NEXT:    global_store_dwordx2 v[0:1], v[1:2], off
+; GPRIDX-NEXT:    v_mov_b32_e32 v9, s8
+; GPRIDX-NEXT:    v_mov_b32_e32 v10, s9
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GPRIDX-NEXT:    v_mov_b32_e32 v11, s10
+; GPRIDX-NEXT:    v_mov_b32_e32 v12, s11
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v2, v2, v10, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GPRIDX-NEXT:    v_mov_b32_e32 v13, s14
+; GPRIDX-NEXT:    v_mov_b32_e32 v14, s15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v2, v2, v12, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GPRIDX-NEXT:    v_mov_b32_e32 v15, s16
+; GPRIDX-NEXT:    v_mov_b32_e32 v16, s17
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v2, v2, v14, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v1, v15, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v2, v16, vcc
+; GPRIDX-NEXT:    global_store_dwordx2 v[0:1], v[0:1], off
 ; GPRIDX-NEXT:    s_endpgm
 ;
 ; MOVREL-LABEL: dyn_extract_v8i64_s_v:
 ; MOVREL:       ; %bb.0: ; %entry
 ; MOVREL-NEXT:    s_mov_b32 s0, s2
 ; MOVREL-NEXT:    s_mov_b32 s1, s3
 ; MOVREL-NEXT:    s_mov_b32 s2, s4
 ; MOVREL-NEXT:    s_mov_b32 s3, s5
 ; MOVREL-NEXT:    s_mov_b32 s4, s6
 ; MOVREL-NEXT:    s_mov_b32 s5, s7
+; MOVREL-NEXT:    v_mov_b32_e32 v1, s0
+; MOVREL-NEXT:    v_mov_b32_e32 v2, s1
+; MOVREL-NEXT:    v_mov_b32_e32 v3, s2
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; MOVREL-NEXT:    v_mov_b32_e32 v4, s3
 ; MOVREL-NEXT:    s_mov_b32 s6, s8
 ; MOVREL-NEXT:    s_mov_b32 s7, s9
+; MOVREL-NEXT:    v_mov_b32_e32 v5, s4
+; MOVREL-NEXT:    v_mov_b32_e32 v6, s5
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
 ; MOVREL-NEXT:    s_mov_b32 s8, s10
 ; MOVREL-NEXT:    s_mov_b32 s9, s11
+; MOVREL-NEXT:    v_mov_b32_e32 v7, s6
+; MOVREL-NEXT:    v_mov_b32_e32 v8, s7
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
 ; MOVREL-NEXT:    s_mov_b32 s10, s12
 ; MOVREL-NEXT:    s_mov_b32 s11, s13
-; MOVREL-NEXT:    s_mov_b32 s12, s14
-; MOVREL-NEXT:    s_mov_b32 s13, s15
-; MOVREL-NEXT:    s_mov_b32 s14, s16
-; MOVREL-NEXT:    s_mov_b32 s15, s17
-; MOVREL-NEXT:    s_mov_b64 s[16:17], exec
-; MOVREL-NEXT:  BB8_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s18, v0
-; MOVREL-NEXT:    s_lshl_b32 m0, s18, 1
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s18, v0
-; MOVREL-NEXT:    s_movrels_b32 s18, s0
-; MOVREL-NEXT:    s_movrels_b32 s19, s1
-; MOVREL-NEXT:    v_mov_b32_e32 v1, s18
-; MOVREL-NEXT:    v_mov_b32_e32 v2, s19
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB8_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[16:17]
-; MOVREL-NEXT:    flat_store_dwordx2 v[0:1], v[1:2]
+; MOVREL-NEXT:    v_mov_b32_e32 v9, s8
+; MOVREL-NEXT:    v_mov_b32_e32 v10, s9
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; MOVREL-NEXT:    v_mov_b32_e32 v11, s10
+; MOVREL-NEXT:    v_mov_b32_e32 v12, s11
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v2, v2, v10, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; MOVREL-NEXT:    v_mov_b32_e32 v13, s14
+; MOVREL-NEXT:    v_mov_b32_e32 v14, s15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v2, v2, v12, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; MOVREL-NEXT:    v_mov_b32_e32 v15, s16
+; MOVREL-NEXT:    v_mov_b32_e32 v16, s17
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v2, v2, v14, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v1, v15, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v2, v16, vcc
+; MOVREL-NEXT:    flat_store_dwordx2 v[0:1], v[0:1]
 ; MOVREL-NEXT:    s_endpgm
 entry:
   %ext = extractelement <8 x i64> %vec, i32 %sel
   store i64 %ext, i64 addrspace(1)* undef
   ret void
 }
 
 define i64 @dyn_extract_v8i64_v_v(<8 x i64> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v8i64_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB9_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v16
-; GPRIDX-NEXT:    s_lshl_b32 s7, s6, 1
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v17, v0
-; GPRIDX-NEXT:    v_mov_b32_e32 v18, v1
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB9_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v17
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, v18
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v8i64_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB9_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v16
-; MOVREL-NEXT:    s_lshl_b32 m0, s6, 1
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
-; MOVREL-NEXT:    v_movrels_b32_e32 v17, v0
-; MOVREL-NEXT:    v_movrels_b32_e32 v18, v1
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB9_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v17
-; MOVREL-NEXT:    v_mov_b32_e32 v1, v18
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v8i64_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v15, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <8 x i64> %vec, i32 %sel
   ret i64 %ext
 }
 
 define amdgpu_ps void @dyn_extract_v8i64_v_s(<8 x i64> %vec, i32 inreg %sel) {
 ; GPRIDX-LABEL: dyn_extract_v8i64_v_s:
 ; GPRIDX:       ; %bb.0: ; %entry
entry:
   %ext = extractelement <8 x i64> %vec, i32 %sel
   store i64 %ext, i64 addrspace(1)* undef
   ret void
 }
 
 define amdgpu_ps float @dyn_extract_v8f32_s_s_offset3(<8 x float> inreg %vec, i32 inreg %sel) {
 ; GCN-LABEL: dyn_extract_v8f32_s_s_offset3:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s0, s2
-; GCN-NEXT:    s_mov_b32 s1, s3
-; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 m0, s10
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_mov_b32 s6, s8
-; GCN-NEXT:    s_mov_b32 s7, s9
-; GCN-NEXT:    s_movrels_b32 s0, s3
+; GCN-NEXT:    s_add_i32 s10, s10, 3
+; GCN-NEXT:    s_cmp_eq_u32 s10, 1
+; GCN-NEXT:    s_cselect_b32 s0, s3, s2
+; GCN-NEXT:    s_cmp_eq_u32 s10, 2
+; GCN-NEXT:    s_cselect_b32 s0, s4, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 3
+; GCN-NEXT:    s_cselect_b32 s0, s5, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 4
+; GCN-NEXT:    s_cselect_b32 s0, s6, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 5
+; GCN-NEXT:    s_cselect_b32 s0, s7, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 6
+; GCN-NEXT:    s_cselect_b32 s0, s8, s0
+; GCN-NEXT:    s_cmp_eq_u32 s10, 7
+; GCN-NEXT:    s_cselect_b32 s0, s9, s0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %add = add i32 %sel, 3
   %ext = extractelement <8 x float> %vec, i32 %add
   ret float %ext
 }
 
 define float @dyn_extract_v8f32_v_v_offset3(<8 x float> %vec, i32 %sel) {
 ; GPRIDX-LABEL: dyn_extract_v8f32_v_v_offset3:
 ; GPRIDX:       ; %bb.0: ; %entry
 ; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB13_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v8
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v8
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v9, v3
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB13_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v9
+; GPRIDX-NEXT:    v_add_u32_e32 v8, 3, v8
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v8
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
 ; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; MOVREL-LABEL: dyn_extract_v8f32_v_v_offset3:
 ; MOVREL:       ; %bb.0: ; %entry
 ; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB13_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v8
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v8
-; MOVREL-NEXT:    v_movrels_b32_e32 v9, v3
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB13_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v9
+; MOVREL-NEXT:    v_add_u32_e32 v8, vcc, 3, v8
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v8
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
 ; MOVREL-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %add = add i32 %sel, 3
   %ext = extractelement <8 x float> %vec, i32 %add
   ret float %ext
 }
 
 define amdgpu_ps double @dyn_extract_v8f64_s_s_offset1(<8 x double> inreg %vec, i32 inreg %sel) {
entry:
   %ext = extractelement <8 x double> %vec, i32 %add
   ret double %ext
 }
 
 define double @dyn_extract_v8f64_v_v_offset3(<8 x double> %vec, i32 %sel) {
 ; GPRIDX-LABEL: dyn_extract_v8f64_v_v_offset3:
 ; GPRIDX:       ; %bb.0: ; %entry
 ; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB22_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v16
-; GPRIDX-NEXT:    s_add_i32 s7, s6, 3
-; GPRIDX-NEXT:    s_lshl_b32 s7, s7, 1
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v17, v0
-; GPRIDX-NEXT:    v_mov_b32_e32 v18, v1
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB22_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v17
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, v18
+; GPRIDX-NEXT:    v_add_u32_e32 v16, 3, v16
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v16
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v1, v1, v15, vcc
 ; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; MOVREL-LABEL: dyn_extract_v8f64_v_v_offset3:
 ; MOVREL:       ; %bb.0: ; %entry
 ; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB22_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v16
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
-; MOVREL-NEXT:    s_add_i32 s6, s6, 3
-; MOVREL-NEXT:    s_lshl_b32 m0, s6, 1
-; MOVREL-NEXT:    v_movrels_b32_e32 v17, v0
-; MOVREL-NEXT:    v_movrels_b32_e32 v18, v1
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB22_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v17
-; MOVREL-NEXT:    v_mov_b32_e32 v1, v18
+; MOVREL-NEXT:    v_add_u32_e32 v16, vcc, 3, v16
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v16
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
+; MOVREL-NEXT:    v_cndmask_b32_e32 v1, v1, v15, vcc
 ; MOVREL-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %add = add i32 %sel, 3
   %ext = extractelement <8 x double> %vec, i32 %add
   ret double %ext
 }
 
 define i8 addrspace(3)* @dyn_extract_v8p3_v_v(<8 x i8 addrspace(3)*> %vec, i32 %idx) {
-; GPRIDX-LABEL: dyn_extract_v8p3_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB23_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v8
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v8
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v9, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB23_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v9
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v8p3_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB23_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v8
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v8
-; MOVREL-NEXT:    v_movrels_b32_e32 v9, v0
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB23_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v9
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v8p3_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v8
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <8 x i8 addrspace(3)*> %vec, i32 %idx
   ret i8 addrspace(3)* %ext
 }
 
 define amdgpu_ps void @dyn_extract_v8p3_s_s(<8 x i8 addrspace(3)*> inreg %vec, i32 inreg %idx) {
 ; GPRIDX-LABEL: dyn_extract_v8p3_s_s:
 ; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_mov_b32 s0, s2
-; GPRIDX-NEXT:    s_mov_b32 m0, s10
-; GPRIDX-NEXT:    s_mov_b32 s1, s3
-; GPRIDX-NEXT:    s_mov_b32 s2, s4
-; GPRIDX-NEXT:    s_mov_b32 s3, s5
-; GPRIDX-NEXT:    s_mov_b32 s4, s6
-; GPRIDX-NEXT:    s_mov_b32 s5, s7
-; GPRIDX-NEXT:    s_mov_b32 s6, s8
-; GPRIDX-NEXT:    s_mov_b32 s7, s9
-; GPRIDX-NEXT:    s_movrels_b32 s0, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 1
+; GPRIDX-NEXT:    s_cselect_b32 s0, s3, s2
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 2
+; GPRIDX-NEXT:    s_cselect_b32 s0, s4, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 3
+; GPRIDX-NEXT:    s_cselect_b32 s0, s5, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 4
+; GPRIDX-NEXT:    s_cselect_b32 s0, s6, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 5
+; GPRIDX-NEXT:    s_cselect_b32 s0, s7, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 6
+; GPRIDX-NEXT:    s_cselect_b32 s0, s8, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s10, 7
+; GPRIDX-NEXT:    s_cselect_b32 s0, s9, s0
 ; GPRIDX-NEXT:    v_mov_b32_e32 v0, s0
 ; GPRIDX-NEXT:    ds_write_b32 v0, v0
 ; GPRIDX-NEXT:    s_endpgm
 ;
 ; MOVREL-LABEL: dyn_extract_v8p3_s_s:
 ; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_mov_b32 s0, s2
-; MOVREL-NEXT:    s_mov_b32 m0, s10
-; MOVREL-NEXT:    s_mov_b32 s1, s3
-; MOVREL-NEXT:    s_mov_b32 s2, s4
-; MOVREL-NEXT:    s_mov_b32 s3, s5
-; MOVREL-NEXT:    s_mov_b32 s4, s6
-; MOVREL-NEXT:    s_mov_b32 s5, s7
-; MOVREL-NEXT:    s_mov_b32 s6, s8
-; MOVREL-NEXT:    s_mov_b32 s7, s9
-; MOVREL-NEXT:    s_movrels_b32 s0, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 1
+; MOVREL-NEXT:    s_cselect_b32 s0, s3, s2
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 2
+; MOVREL-NEXT:    s_cselect_b32 s0, s4, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 3
+; MOVREL-NEXT:    s_cselect_b32 s0, s5, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 4
+; MOVREL-NEXT:    s_cselect_b32 s0, s6, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 5
+; MOVREL-NEXT:    s_cselect_b32 s0, s7, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 6
+; MOVREL-NEXT:    s_cselect_b32 s0, s8, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s10, 7
+; MOVREL-NEXT:    s_cselect_b32 s0, s9, s0
 ; MOVREL-NEXT:    v_mov_b32_e32 v0, s0
 ; MOVREL-NEXT:    s_mov_b32 m0, -1
 ; MOVREL-NEXT:    ds_write_b32 v0, v0
 ; MOVREL-NEXT:    s_endpgm
 entry:
   %ext = extractelement <8 x i8 addrspace(3)*> %vec, i32 %idx
   store i8 addrspace(3)* %ext, i8 addrspace(3)* addrspace(3)* undef
   ret void
 }
 
 define i8 addrspace(1)* @dyn_extract_v8p1_v_v(<8 x i8 addrspace(1)*> %vec, i32 %idx) {
-; GPRIDX-LABEL: dyn_extract_v8p1_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB25_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v16
-; GPRIDX-NEXT:    s_lshl_b32 s7, s6, 1
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v17, v0
-; GPRIDX-NEXT:    v_mov_b32_e32 v18, v1
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB25_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v17
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, v18
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v8p1_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB25_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v16
-; MOVREL-NEXT:    s_lshl_b32 m0, s6, 1
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
-; MOVREL-NEXT:    v_movrels_b32_e32 v17, v0
-; MOVREL-NEXT:    v_movrels_b32_e32 v18, v1
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB25_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v17
-; MOVREL-NEXT:    v_mov_b32_e32 v1, v18
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v8p1_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v16
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v15, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <8 x i8 addrspace(1)*> %vec, i32 %idx
   ret i8 addrspace(1)* %ext
 }
 
 define amdgpu_ps void @dyn_extract_v8p1_s_s(<8 x i8 addrspace(1)*> inreg %vec, i32 inreg %idx) {
 ; GPRIDX-LABEL: dyn_extract_v8p1_s_s:
 ; GPRIDX:       ; %bb.0: ; %entry
entry:
   ret double %ext
 }
 
 define amdgpu_ps float @dyn_extract_v6f32_s_v(<6 x float> inreg %vec, i32 %sel) {
 ; GCN-LABEL: dyn_extract_v6f32_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_mov_b32 s0, s2
 ; GCN-NEXT:    s_mov_b32 s1, s3
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_mov_b64 s[6:7], exec
-; GCN-NEXT:  BB33_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s8, v0
-; GCN-NEXT:    s_mov_b32 m0, s8
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s8, v0
-; GCN-NEXT:    s_movrels_b32 s8, s0
-; GCN-NEXT:    v_mov_b32_e32 v1, s8
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB33_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[6:7]
-; GCN-NEXT:    v_mov_b32_e32 v0, v1
+; GCN-NEXT:    v_mov_b32_e32 v1, s0
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GCN-NEXT:    v_mov_b32_e32 v3, s4
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_mov_b32_e32 v4, s5
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_mov_b32_e32 v5, s6
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_mov_b32_e32 v6, s7
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v6, vcc
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <6 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define float @dyn_extract_v6f32_v_v(<6 x float> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v6f32_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB34_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v6
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v6
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v7, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB34_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v7
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v6f32_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB34_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v6
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v6
-; MOVREL-NEXT:    v_movrels_b32_e32 v7, v0
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB34_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v7
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v6f32_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <6 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v6f32_v_s(<6 x float> %vec, i32 inreg %sel) {
-; GPRIDX-LABEL: dyn_extract_v6f32_v_s:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    ; return to shader part epilog
-;
-; MOVREL-LABEL: dyn_extract_v6f32_v_s:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_mov_b32 m0, s2
-; MOVREL-NEXT:    v_movrels_b32_e32 v0, v0
-; MOVREL-NEXT:    ; return to shader part epilog
+; GCN-LABEL: dyn_extract_v6f32_v_s:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 3
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 4
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 5
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <6 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v6f32_s_s(<6 x float> inreg %vec, i32 inreg %sel) {
 ; GCN-LABEL: dyn_extract_v6f32_s_s:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s0, s2
-; GCN-NEXT:    s_mov_b32 m0, s8
-; GCN-NEXT:    s_mov_b32 s1, s3
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_movrels_b32 s0, s0
+; GCN-NEXT:    s_cmp_eq_u32 s8, 1
+; GCN-NEXT:    s_cselect_b32 s0, s3, s2
+; GCN-NEXT:    s_cmp_eq_u32 s8, 2
+; GCN-NEXT:    s_cselect_b32 s0, s4, s0
+; GCN-NEXT:    s_cmp_eq_u32 s8, 3
+; GCN-NEXT:    s_cselect_b32 s0, s5, s0
+; GCN-NEXT:    s_cmp_eq_u32 s8, 4
+; GCN-NEXT:    s_cselect_b32 s0, s6, s0
+; GCN-NEXT:    s_cmp_eq_u32 s8, 5
+; GCN-NEXT:    s_cselect_b32 s0, s7, s0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <6 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v7f32_s_v(<7 x float> inreg %vec, i32 %sel) {
 ; GCN-LABEL: dyn_extract_v7f32_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_mov_b32 s0, s2
 ; GCN-NEXT:    s_mov_b32 s1, s3
 ; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s6, s8
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_mov_b64 s[8:9], exec
-; GCN-NEXT:  BB37_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s7, v0
-; GCN-NEXT:    s_mov_b32 m0, s7
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s7, v0
-; GCN-NEXT:    s_movrels_b32 s7, s0
-; GCN-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB37_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[8:9]
-; GCN-NEXT:    v_mov_b32_e32 v0, v1
+; GCN-NEXT:    v_mov_b32_e32 v1, s0
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GCN-NEXT:    v_mov_b32_e32 v3, s2
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_mov_b32_e32 v4, s5
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_mov_b32_e32 v5, s6
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_mov_b32_e32 v6, s7
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
+; GCN-NEXT:    v_mov_b32_e32 v7, s8
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v7, vcc
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <7 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define float @dyn_extract_v7f32_v_v(<7 x float> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v7f32_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB38_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v7
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v7
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v8, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB38_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v8
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v7f32_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB38_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v7
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v7
-; MOVREL-NEXT:    v_movrels_b32_e32 v8, v0
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB38_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v8
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v7f32_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v7
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <7 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v7f32_v_s(<7 x float> %vec, i32 inreg %sel) {
-; GPRIDX-LABEL: dyn_extract_v7f32_v_s:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_set_gpr_idx_on s2, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    ; return to shader part epilog
-;
-; MOVREL-LABEL: dyn_extract_v7f32_v_s:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_mov_b32 m0, s2
-; MOVREL-NEXT:    v_movrels_b32_e32 v0, v0
-; MOVREL-NEXT:    ; return to shader part epilog
+; GCN-LABEL: dyn_extract_v7f32_v_s:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 3
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 4
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 5
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <7 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v7f32_s_s(<7 x float> inreg %vec, i32 inreg %sel) {
 ; GCN-LABEL: dyn_extract_v7f32_s_s:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s0, s2
-; GCN-NEXT:    s_mov_b32 m0, s9
-; GCN-NEXT:    s_mov_b32 s1, s3
-; GCN-NEXT:    s_mov_b32 s2, s4
-; GCN-NEXT:    s_mov_b32 s3, s5
-; GCN-NEXT:    s_mov_b32 s4, s6
-; GCN-NEXT:    s_mov_b32 s5, s7
-; GCN-NEXT:    s_mov_b32 s6, s8
-; GCN-NEXT:    s_movrels_b32 s0, s0
+; GCN-NEXT:    s_cmp_eq_u32 s9, 1
+; GCN-NEXT:    s_cselect_b32 s0, s3, s2
+; GCN-NEXT:    s_cmp_eq_u32 s9, 2
+; GCN-NEXT:    s_cselect_b32 s0, s4, s0
+; GCN-NEXT:    s_cmp_eq_u32 s9, 3
+; GCN-NEXT:    s_cselect_b32 s0, s5, s0
+; GCN-NEXT:    s_cmp_eq_u32 s9, 4
+; GCN-NEXT:    s_cselect_b32 s0, s6, s0
+; GCN-NEXT:    s_cmp_eq_u32 s9, 5
+; GCN-NEXT:    s_cselect_b32 s0, s7, s0
+; GCN-NEXT:    s_cmp_eq_u32 s9, 6
+; GCN-NEXT:    s_cselect_b32 s0, s8, s0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <7 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps double @dyn_extract_v6f64_s_v(<6 x double> inreg %vec, i32 %sel) {
 ; GCN-LABEL: dyn_extract_v6f64_s_v:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s16, s2
-; GCN-NEXT:    s_mov_b32 s17, s3
-; GCN-NEXT:    s_mov_b32 s18, s4
-; GCN-NEXT:    s_mov_b32 s19, s5
-; GCN-NEXT:    s_mov_b32 s20, s6
-; GCN-NEXT:    s_mov_b32 s21, s7
-; GCN-NEXT:    s_mov_b32 s22, s8
-; GCN-NEXT:    s_mov_b32 s23, s9
-; GCN-NEXT:    s_mov_b32 s24, s10
-; GCN-NEXT:    s_mov_b32 s25, s11
-; GCN-NEXT:    s_mov_b32 s26, s12
-; GCN-NEXT:    s_mov_b32 s27, s13
-; GCN-NEXT:    s_mov_b64 s[2:3], exec
-; GCN-NEXT:  BB41_1: ; =>This Inner Loop Header: Depth=1
+; GCN-NEXT:    s_mov_b32 s0, s2
+; GCN-NEXT:    s_mov_b32 s1, s3
+; GCN-NEXT:    s_mov_b32 s2, s4
+; GCN-NEXT:    s_mov_b32 s3, s5
+; GCN-NEXT:    s_mov_b32 s4, s6
+; GCN-NEXT:    s_mov_b32 s5, s7
+; GCN-NEXT:    v_mov_b32_e32 v1, s0
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_mov_b32_e32 v3, s2
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_mov_b32_e32 v4, s3
+; GCN-NEXT:    s_mov_b32 s6, s8
+; GCN-NEXT:    s_mov_b32 s7, s9
+; GCN-NEXT:    v_mov_b32_e32 v5, s4
+; GCN-NEXT:    v_mov_b32_e32 v6, s5
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    v_mov_b32_e32 v7, s6
+; GCN-NEXT:    v_mov_b32_e32 v8, s7
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_mov_b32_e32 v9, s10
+; GCN-NEXT:    v_mov_b32_e32 v10, s11
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_mov_b32_e32 v11, s12
+; GCN-NEXT:    v_mov_b32_e32 v12, s13
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v10, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v11, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v2, v12, vcc
 ; GCN-NEXT:    v_readfirstlane_b32 s0, v0
-; GCN-NEXT:    s_lshl_b32 m0, s0, 1
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s0, v0
-; GCN-NEXT:    s_movrels_b32 s0, s16
-; GCN-NEXT:    s_movrels_b32 s1, s17
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB41_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[2:3]
+; GCN-NEXT:    v_readfirstlane_b32 s1, v1
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <6 x double> %vec, i32 %sel
   ret double %ext
 }
 
 define double @dyn_extract_v6f64_v_v(<6 x double> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v6f64_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB42_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v12
-; GPRIDX-NEXT:    s_lshl_b32 s7, s6, 1
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v12
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v13, v0
-; GPRIDX-NEXT:    v_mov_b32_e32 v14, v1
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB42_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v13
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, v14
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v6f64_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB42_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v12
-; MOVREL-NEXT:    s_lshl_b32 m0, s6, 1
-; MOVREL-NEXT:    v_movrels_b32_e32 v13, v0
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v12
-; MOVREL-NEXT:    v_movrels_b32_e32 v14, v1
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB42_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v13
-; MOVREL-NEXT:    v_mov_b32_e32 v1, v14
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v6f64_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v12
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v12
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v12
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v12
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v12
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <6 x double> %vec, i32 %sel
   ret double %ext
 }
 
 define amdgpu_ps double @dyn_extract_v6f64_v_s(<6 x double> %vec, i32 inreg %sel) {
 ; GPRIDX-LABEL: dyn_extract_v6f64_v_s:
 ; GPRIDX:       ; %bb.0: ; %entry

 entry:
   %ext = extractelement <6 x double> %vec, i32 %sel
   ret double %ext
 }
 
 define amdgpu_ps double @dyn_extract_v7f64_s_v(<7 x double> inreg %vec, i32 %sel) {
 ; GCN-LABEL: dyn_extract_v7f64_s_v:
 ; GCN:       ; %bb.0: ; %entry
-; GCN-NEXT:    s_mov_b32 s16, s2
-; GCN-NEXT:    s_mov_b32 s17, s3
-; GCN-NEXT:    s_mov_b32 s18, s4
-; GCN-NEXT:    s_mov_b32 s19, s5
-; GCN-NEXT:    s_mov_b32 s20, s6
-; GCN-NEXT:    s_mov_b32 s21, s7
-; GCN-NEXT:    s_mov_b32 s22, s8
-; GCN-NEXT:    s_mov_b32 s23, s9
-; GCN-NEXT:    s_mov_b32 s24, s10
-; GCN-NEXT:    s_mov_b32 s25, s11
-; GCN-NEXT:    s_mov_b32 s26, s12
-; GCN-NEXT:    s_mov_b32 s27, s13
-; GCN-NEXT:    s_mov_b32 s28, s14
-; GCN-NEXT:    s_mov_b32 s29, s15
-; GCN-NEXT:    s_mov_b64 s[2:3], exec
-; GCN-NEXT:  BB45_1: ; =>This Inner Loop Header: Depth=1
+; GCN-NEXT:    s_mov_b32 s0, s2
+; GCN-NEXT:    s_mov_b32 s1, s3
+; GCN-NEXT:    s_mov_b32 s2, s4
+; GCN-NEXT:    s_mov_b32 s3, s5
+; GCN-NEXT:    s_mov_b32 s4, s6
+; GCN-NEXT:    s_mov_b32 s5, s7
+; GCN-NEXT:    v_mov_b32_e32 v1, s0
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_mov_b32_e32 v3, s2
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_mov_b32_e32 v4, s3
+; GCN-NEXT:    s_mov_b32 s6, s8
+; GCN-NEXT:    s_mov_b32 s7, s9
+; GCN-NEXT:    v_mov_b32_e32 v5, s4
+; GCN-NEXT:    v_mov_b32_e32 v6, s5
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    s_mov_b32 s8, s10
+; GCN-NEXT:    s_mov_b32 s9, s11
+; GCN-NEXT:    v_mov_b32_e32 v7, s6
+; GCN-NEXT:    v_mov_b32_e32 v8, s7
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_mov_b32_e32 v9, s8
+; GCN-NEXT:    v_mov_b32_e32 v10, s9
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_mov_b32_e32 v11, s12
+; GCN-NEXT:    v_mov_b32_e32 v12, s13
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v10, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_mov_b32_e32 v13, s14
+; GCN-NEXT:    v_mov_b32_e32 v14, s15
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v12, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v13, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v2, v14, vcc
 ; GCN-NEXT:    v_readfirstlane_b32 s0, v0
-; GCN-NEXT:    s_lshl_b32 m0, s0, 1
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s0, v0
-; GCN-NEXT:    s_movrels_b32 s0, s16
-; GCN-NEXT:    s_movrels_b32 s1, s17
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB45_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[2:3]
+; GCN-NEXT:    v_readfirstlane_b32 s1, v1
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <7 x double> %vec, i32 %sel
   ret double %ext
 }
 
 define double @dyn_extract_v7f64_v_v(<7 x double> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v7f64_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB46_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v14
-; GPRIDX-NEXT:    s_lshl_b32 s7, s6, 1
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v14
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v15, v0
-; GPRIDX-NEXT:    v_mov_b32_e32 v16, v1
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB46_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v15
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, v16
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v7f64_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB46_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v14
-; MOVREL-NEXT:    s_lshl_b32 m0, s6, 1
-; MOVREL-NEXT:    v_movrels_b32_e32 v15, v0
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v14
-; MOVREL-NEXT:    v_movrels_b32_e32 v16, v1
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB46_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v15
-; MOVREL-NEXT:    v_mov_b32_e32 v1, v16
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v7f64_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v14
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v14
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v14
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v14
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v14
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v14
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <7 x double> %vec, i32 %sel
   ret double %ext
 }
 
 define amdgpu_ps double @dyn_extract_v7f64_v_s(<7 x double> %vec, i32 inreg %sel) {
 ; GPRIDX-LABEL: dyn_extract_v7f64_v_s:
 ; GPRIDX:       ; %bb.0: ; %entry

 ; GPRIDX-NEXT:     amd_code_version_minor = 2
 ; GPRIDX-NEXT:     amd_machine_kind = 1
 ; GPRIDX-NEXT:     amd_machine_version_major = 9
 ; GPRIDX-NEXT:     amd_machine_version_minor = 0
 ; GPRIDX-NEXT:     amd_machine_version_stepping = 0
 ; GPRIDX-NEXT:     kernel_code_entry_byte_offset = 256
 ; GPRIDX-NEXT:     kernel_code_prefetch_byte_size = 0
 ; GPRIDX-NEXT:     granulated_workitem_vgpr_count = 0
-; GPRIDX-NEXT:     granulated_wavefront_sgpr_count = 2
+; GPRIDX-NEXT:     granulated_wavefront_sgpr_count = 1
 ; GPRIDX-NEXT:     priority = 0
 ; GPRIDX-NEXT:     float_mode = 240
 ; GPRIDX-NEXT:     priv = 0
 ; GPRIDX-NEXT:     enable_dx10_clamp = 1
 ; GPRIDX-NEXT:     debug_mode = 0
 ; GPRIDX-NEXT:     enable_ieee_mode = 1
 ; GPRIDX-NEXT:     enable_wgp_mode = 0
 ; GPRIDX-NEXT:     enable_mem_ordered = 0

 ; GPRIDX-NEXT:     is_dynamic_callstack = 0
 ; GPRIDX-NEXT:     is_debug_enabled = 0
 ; GPRIDX-NEXT:     is_xnack_enabled = 0
 ; GPRIDX-NEXT:     workitem_private_segment_byte_size = 0
 ; GPRIDX-NEXT:     workgroup_group_segment_byte_size = 0
 ; GPRIDX-NEXT:     gds_segment_byte_size = 0
 ; GPRIDX-NEXT:     kernarg_segment_byte_size = 28
 ; GPRIDX-NEXT:     workgroup_fbarrier_count = 0
-; GPRIDX-NEXT:     wavefront_sgpr_count = 24
+; GPRIDX-NEXT:     wavefront_sgpr_count = 9
 ; GPRIDX-NEXT:     workitem_vgpr_count = 4
 ; GPRIDX-NEXT:     reserved_vgpr_first = 0
 ; GPRIDX-NEXT:     reserved_vgpr_count = 0
 ; GPRIDX-NEXT:     reserved_sgpr_first = 0
 ; GPRIDX-NEXT:     reserved_sgpr_count = 0
 ; GPRIDX-NEXT:     debug_wavefront_private_segment_offset_sgpr = 0
 ; GPRIDX-NEXT:     debug_private_segment_buffer_sgpr = 0
 ; GPRIDX-NEXT:     kernarg_segment_alignment = 4
 ; GPRIDX-NEXT:     group_segment_alignment = 4
 ; GPRIDX-NEXT:     private_segment_alignment = 4
 ; GPRIDX-NEXT:     wavefront_size = 6
 ; GPRIDX-NEXT:     call_convention = -1
 ; GPRIDX-NEXT:     runtime_loader_kernel_symbol = 0
 ; GPRIDX-NEXT:    .end_amd_kernel_code_t
 ; GPRIDX-NEXT:  ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
-; GPRIDX-NEXT:    s_load_dword s2, s[4:5], 0x8
-; GPRIDX-NEXT:    s_mov_b32 s16, 0
-; GPRIDX-NEXT:    s_mov_b64 s[8:9], 1.0
-; GPRIDX-NEXT:    s_mov_b32 s17, 0x40140000
-; GPRIDX-NEXT:    s_mov_b64 s[14:15], 4.0
+; GPRIDX-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x0
+; GPRIDX-NEXT:    s_load_dword s8, s[4:5], 0x8
+; GPRIDX-NEXT:    s_mov_b32 s0, 0
+; GPRIDX-NEXT:    s_mov_b32 s3, 0x40080000
+; GPRIDX-NEXT:    s_mov_b32 s2, s0
+; GPRIDX-NEXT:    s_mov_b32 s1, 0x40140000
 ; GPRIDX-NEXT:    s_waitcnt lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b32 m0, s2
-; GPRIDX-NEXT:    s_mov_b32 s13, 0x40080000
-; GPRIDX-NEXT:    s_mov_b32 s12, s16
-; GPRIDX-NEXT:    s_mov_b64 s[10:11], 2.0
-; GPRIDX-NEXT:    s_movrels_b64 s[2:3], s[8:9]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, s2
-; GPRIDX-NEXT:    v_mov_b32_e32 v3, s1
-; GPRIDX-NEXT:    v_mov_b32_e32 v1, s3
-; GPRIDX-NEXT:    v_mov_b32_e32 v2, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s8, 1
+; GPRIDX-NEXT:    s_cselect_b64 s[4:5], 2.0, 1.0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s8, 2
+; GPRIDX-NEXT:    s_cselect_b64 s[2:3], s[2:3], s[4:5]
+; GPRIDX-NEXT:    s_cmp_eq_u32 s8, 3
+; GPRIDX-NEXT:    s_cselect_b64 s[2:3], 4.0, s[2:3]
+; GPRIDX-NEXT:    s_cmp_eq_u32 s8, 4
+; GPRIDX-NEXT:    s_cselect_b64 s[0:1], s[0:1], s[2:3]
+; GPRIDX-NEXT:    v_mov_b32_e32 v0, s0
+; GPRIDX-NEXT:    v_mov_b32_e32 v2, s6
+; GPRIDX-NEXT:    v_mov_b32_e32 v1, s1
+; GPRIDX-NEXT:    v_mov_b32_e32 v3, s7
 ; GPRIDX-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
 ; GPRIDX-NEXT:    s_endpgm
 ;
 ; MOVREL-LABEL: dyn_extract_v5f64_s_s:
 ; MOVREL:         .amd_kernel_code_t
 ; MOVREL-NEXT:     amd_code_version_major = 1
 ; MOVREL-NEXT:     amd_code_version_minor = 2
 ; MOVREL-NEXT:     amd_machine_kind = 1
 ; MOVREL-NEXT:     amd_machine_version_major = 8
 ; MOVREL-NEXT:     amd_machine_version_minor = 0
 ; MOVREL-NEXT:     amd_machine_version_stepping = 3
 ; MOVREL-NEXT:     kernel_code_entry_byte_offset = 256
 ; MOVREL-NEXT:     kernel_code_prefetch_byte_size = 0
 ; MOVREL-NEXT:     granulated_workitem_vgpr_count = 0
-; MOVREL-NEXT:     granulated_wavefront_sgpr_count = 2
+; MOVREL-NEXT:     granulated_wavefront_sgpr_count = 1
 ; MOVREL-NEXT:     priority = 0
 ; MOVREL-NEXT:     float_mode = 240
 ; MOVREL-NEXT:     priv = 0
 ; MOVREL-NEXT:     enable_dx10_clamp = 1
 ; MOVREL-NEXT:     debug_mode = 0
 ; MOVREL-NEXT:     enable_ieee_mode = 1
 ; MOVREL-NEXT:     enable_wgp_mode = 0
 ; MOVREL-NEXT:     enable_mem_ordered = 0

 ; MOVREL-NEXT:     is_dynamic_callstack = 0
 ; MOVREL-NEXT:     is_debug_enabled = 0
 ; MOVREL-NEXT:     is_xnack_enabled = 0
 ; MOVREL-NEXT:     workitem_private_segment_byte_size = 0
 ; MOVREL-NEXT:     workgroup_group_segment_byte_size = 0
 ; MOVREL-NEXT:     gds_segment_byte_size = 0
 ; MOVREL-NEXT:     kernarg_segment_byte_size = 28
 ; MOVREL-NEXT:     workgroup_fbarrier_count = 0
-; MOVREL-NEXT:     wavefront_sgpr_count = 24
+; MOVREL-NEXT:     wavefront_sgpr_count = 9
 ; MOVREL-NEXT:     workitem_vgpr_count = 4
 ; MOVREL-NEXT:     reserved_vgpr_first = 0
 ; MOVREL-NEXT:     reserved_vgpr_count = 0
 ; MOVREL-NEXT:     reserved_sgpr_first = 0
 ; MOVREL-NEXT:     reserved_sgpr_count = 0
 ; MOVREL-NEXT:     debug_wavefront_private_segment_offset_sgpr = 0
 ; MOVREL-NEXT:     debug_private_segment_buffer_sgpr = 0
 ; MOVREL-NEXT:     kernarg_segment_alignment = 4
 ; MOVREL-NEXT:     group_segment_alignment = 4
 ; MOVREL-NEXT:     private_segment_alignment = 4
 ; MOVREL-NEXT:     wavefront_size = 6
 ; MOVREL-NEXT:     call_convention = -1
 ; MOVREL-NEXT:     runtime_loader_kernel_symbol = 0
 ; MOVREL-NEXT:    .end_amd_kernel_code_t
 ; MOVREL-NEXT:  ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
-; MOVREL-NEXT:    s_load_dword s2, s[4:5], 0x8
-; MOVREL-NEXT:    s_mov_b32 s16, 0
-; MOVREL-NEXT:    s_mov_b64 s[8:9], 1.0
-; MOVREL-NEXT:    s_mov_b32 s17, 0x40140000
-; MOVREL-NEXT:    s_mov_b64 s[14:15], 4.0
+; MOVREL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x0
+; MOVREL-NEXT:    s_load_dword s8, s[4:5], 0x8
+; MOVREL-NEXT:    s_mov_b32 s0, 0
+; MOVREL-NEXT:    s_mov_b32 s3, 0x40080000
+; MOVREL-NEXT:    s_mov_b32 s2, s0
+; MOVREL-NEXT:    s_mov_b32 s1, 0x40140000
 ; MOVREL-NEXT:    s_waitcnt lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b32 m0, s2
-; MOVREL-NEXT:    s_mov_b32 s13, 0x40080000
-; MOVREL-NEXT:    s_mov_b32 s12, s16
-; MOVREL-NEXT:    s_mov_b64 s[10:11], 2.0
-; MOVREL-NEXT:    s_movrels_b64 s[2:3], s[8:9]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, s2
-; MOVREL-NEXT:    v_mov_b32_e32 v3, s1
-; MOVREL-NEXT:    v_mov_b32_e32 v1, s3
-; MOVREL-NEXT:    v_mov_b32_e32 v2, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s8, 1
+; MOVREL-NEXT:    s_cselect_b64 s[4:5], 2.0, 1.0
+; MOVREL-NEXT:    s_cmp_eq_u32 s8, 2
+; MOVREL-NEXT:    s_cselect_b64 s[2:3], s[2:3], s[4:5]
+; MOVREL-NEXT:    s_cmp_eq_u32 s8, 3
+; MOVREL-NEXT:    s_cselect_b64 s[2:3], 4.0, s[2:3]
+; MOVREL-NEXT:    s_cmp_eq_u32 s8, 4
+; MOVREL-NEXT:    s_cselect_b64 s[0:1], s[0:1], s[2:3]
+; MOVREL-NEXT:    v_mov_b32_e32 v0, s0
+; MOVREL-NEXT:    v_mov_b32_e32 v2, s6
+; MOVREL-NEXT:    v_mov_b32_e32 v1, s1
+; MOVREL-NEXT:    v_mov_b32_e32 v3, s7
 ; MOVREL-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; MOVREL-NEXT:    s_endpgm
 entry:
   %ext = extractelement <5 x double> <double 1.0, double 2.0, double 3.0, double 4.0, double 5.0>, i32 %sel
   store double %ext, double addrspace(1)* %out
   ret void
 }
 
 define float @dyn_extract_v15f32_const_s_v(i32 %sel) {
 ; GCN-LABEL: dyn_extract_v15f32_const_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s18, 0x41700000
-; GCN-NEXT:    s_mov_b32 s17, 0x41600000
-; GCN-NEXT:    s_mov_b32 s16, 0x41500000
-; GCN-NEXT:    s_mov_b32 s15, 0x41400000
-; GCN-NEXT:    s_mov_b32 s14, 0x41300000
-; GCN-NEXT:    s_mov_b32 s13, 0x41200000
-; GCN-NEXT:    s_mov_b32 s12, 0x41100000
-; GCN-NEXT:    s_mov_b32 s11, 0x41000000
-; GCN-NEXT:    s_mov_b32 s10, 0x40e00000
-; GCN-NEXT:    s_mov_b32 s9, 0x40c00000
-; GCN-NEXT:    s_mov_b32 s8, 0x40a00000
-; GCN-NEXT:    s_mov_b32 s7, 4.0
-; GCN-NEXT:    s_mov_b32 s6, 0x40400000
-; GCN-NEXT:    s_mov_b32 s5, 2.0
-; GCN-NEXT:    s_mov_b32 s4, 1.0
-; GCN-NEXT:    s_mov_b64 s[20:21], exec
-; GCN-NEXT:  BB50_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s19, v0
-; GCN-NEXT:    s_mov_b32 m0, s19
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s19, v0
-; GCN-NEXT:    s_movrels_b32 s19, s4
-; GCN-NEXT:    v_mov_b32_e32 v1, s19
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB50_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[20:21]
-; GCN-NEXT:    v_mov_b32_e32 v0, v1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GCN-NEXT:    v_cndmask_b32_e64 v13, 1.0, 2.0, vcc
+; GCN-NEXT:    v_mov_b32_e32 v1, 0x40400000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v13, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
+; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, 4.0, vcc
+; GCN-NEXT:    v_mov_b32_e32 v2, 0x40a00000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GCN-NEXT:    v_mov_b32_e32 v3, 0x40c00000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_mov_b32_e32 v4, 0x40e00000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_mov_b32_e32 v5, 0x41000000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_mov_b32_e32 v6, 0x41100000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 8, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
+; GCN-NEXT:    v_mov_b32_e32 v7, 0x41200000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 9, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_mov_b32_e32 v8, 0x41300000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 10, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
+; GCN-NEXT:    v_mov_b32_e32 v9, 0x41400000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 11, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_mov_b32_e32 v10, 0x41500000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 12, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v10, vcc
+; GCN-NEXT:    v_mov_b32_e32 v11, 0x41600000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 13, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_mov_b32_e32 v12, 0x41700000
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 14, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v12, vcc
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <15 x float> <float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0, float 8.0, float 9.0, float 10.0, float 11.0, float 12.0, float 13.0, float 14.0, float 15.0>, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v15f32_const_s_s(i32 inreg %sel) {
 ; GCN-LABEL: dyn_extract_v15f32_const_s_s:

 }
 
 define amdgpu_ps float @dyn_extract_v15f32_s_v(<15 x float> inreg %vec, i32 %sel) {
 ; GCN-LABEL: dyn_extract_v15f32_s_v:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_mov_b32 s0, s2
 ; GCN-NEXT:    s_mov_b32 s1, s3
 ; GCN-NEXT:    s_mov_b32 s2, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s0
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
 ; GCN-NEXT:    s_mov_b32 s3, s5
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GCN-NEXT:    v_mov_b32_e32 v3, s2
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v0
 ; GCN-NEXT:    s_mov_b32 s4, s6
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_mov_b32_e32 v4, s3
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v0
 ; GCN-NEXT:    s_mov_b32 s5, s7
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_mov_b32_e32 v5, s4
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v0
 ; GCN-NEXT:    s_mov_b32 s6, s8
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_mov_b32_e32 v6, s5
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v0
 ; GCN-NEXT:    s_mov_b32 s7, s9
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
+; GCN-NEXT:    v_mov_b32_e32 v7, s6
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v0
 ; GCN-NEXT:    s_mov_b32 s8, s10
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
+; GCN-NEXT:    v_mov_b32_e32 v8, s7
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v0
 ; GCN-NEXT:    s_mov_b32 s9, s11
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
+; GCN-NEXT:    v_mov_b32_e32 v9, s8
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 8, v0
 ; GCN-NEXT:    s_mov_b32 s10, s12
-; GCN-NEXT:    s_mov_b32 s11, s13
-; GCN-NEXT:    s_mov_b32 s12, s14
-; GCN-NEXT:    s_mov_b32 s14, s16
-; GCN-NEXT:    s_mov_b32 s13, s15
-; GCN-NEXT:    s_mov_b64 s[16:17], exec
-; GCN-NEXT:  BB52_1: ; =>This Inner Loop Header: Depth=1
-; GCN-NEXT:    v_readfirstlane_b32 s15, v0
-; GCN-NEXT:    s_mov_b32 m0, s15
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s15, v0
-; GCN-NEXT:    s_movrels_b32 s15, s0
-; GCN-NEXT:    v_mov_b32_e32 v1, s15
-; GCN-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GCN-NEXT:    s_xor_b64 exec, exec, vcc
-; GCN-NEXT:    s_cbranch_execnz BB52_1
-; GCN-NEXT:  ; %bb.2:
-; GCN-NEXT:    s_mov_b64 exec, s[16:17]
-; GCN-NEXT:    v_mov_b32_e32 v0, v1
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GCN-NEXT:    v_mov_b32_e32 v10, s9
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 9, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v10, vcc
+; GCN-NEXT:    v_mov_b32_e32 v11, s10
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 10, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v11, vcc
+; GCN-NEXT:    v_mov_b32_e32 v12, s13
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 11, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v12, vcc
+; GCN-NEXT:    v_mov_b32_e32 v13, s14
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 12, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v13, vcc
+; GCN-NEXT:    v_mov_b32_e32 v14, s15
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 13, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v14, vcc
+; GCN-NEXT:    v_mov_b32_e32 v15, s16
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 14, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v15, vcc
 ; GCN-NEXT:    ; return to shader part epilog
 entry:
   %ext = extractelement <15 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define float @dyn_extract_v15f32_v_v(<15 x float> %vec, i32 %sel) {
-; GPRIDX-LABEL: dyn_extract_v15f32_v_v:
-; GPRIDX:       ; %bb.0: ; %entry
-; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB53_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v15
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v15
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v16, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB53_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v16
-; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
-;
-; MOVREL-LABEL: dyn_extract_v15f32_v_v:
-; MOVREL:       ; %bb.0: ; %entry
-; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB53_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v15
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v15
-; MOVREL-NEXT:    v_movrels_b32_e32 v16, v0
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB53_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v16
-; MOVREL-NEXT:    s_setpc_b64 s[30:31]
+; GCN-LABEL: dyn_extract_v15f32_v_v:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 8, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 9, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 10, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 11, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v11, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 12, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 13, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v13, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 14, v15
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
+; GCN-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %ext = extractelement <15 x float> %vec, i32 %sel
   ret float %ext
 }
 
 define amdgpu_ps float @dyn_extract_v15f32_v_s(<15 x float> %vec, i32 inreg %sel) {
 ; GPRIDX-LABEL: dyn_extract_v15f32_v_s:
 ; GPRIDX:       ; %bb.0: ; %entry
entry:
   %ext = extractelement <15 x float> %vec, i32 %add
   ret float %ext
 }
 
 define float @dyn_extract_v15f32_v_v_offset3(<15 x float> %vec, i32 %sel) {
 ; GPRIDX-LABEL: dyn_extract_v15f32_v_v_offset3:
 ; GPRIDX:       ; %bb.0: ; %entry
 ; GPRIDX-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GPRIDX-NEXT:    s_mov_b64 s[4:5], exec
-; GPRIDX-NEXT:  BB57_1: ; =>This Inner Loop Header: Depth=1
-; GPRIDX-NEXT:    v_readfirstlane_b32 s6, v15
-; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v15
-; GPRIDX-NEXT:    s_set_gpr_idx_on s6, gpr_idx(SRC0)
-; GPRIDX-NEXT:    v_mov_b32_e32 v16, v3
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
-; GPRIDX-NEXT:    s_xor_b64 exec, exec, vcc
-; GPRIDX-NEXT:    s_cbranch_execnz BB57_1
-; GPRIDX-NEXT:  ; %bb.2:
-; GPRIDX-NEXT:    s_mov_b64 exec, s[4:5]
-; GPRIDX-NEXT:    v_mov_b32_e32 v0, v16
+; GPRIDX-NEXT:    v_add_u32_e32 v15, 3, v15
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 8, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 9, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 10, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 11, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v11, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 12, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 13, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v13, vcc
+; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, 14, v15
+; GPRIDX-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
 ; GPRIDX-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; MOVREL-LABEL: dyn_extract_v15f32_v_v_offset3:
 ; MOVREL:       ; %bb.0: ; %entry
 ; MOVREL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; MOVREL-NEXT:    s_mov_b64 s[4:5], exec
-; MOVREL-NEXT:  BB57_1: ; =>This Inner Loop Header: Depth=1
-; MOVREL-NEXT:    v_readfirstlane_b32 s6, v15
-; MOVREL-NEXT:    s_mov_b32 m0, s6
-; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v15
-; MOVREL-NEXT:    v_movrels_b32_e32 v16, v3
-; MOVREL-NEXT:    s_and_saveexec_b64 vcc, vcc
-; MOVREL-NEXT:    s_xor_b64 exec, exec, vcc
-; MOVREL-NEXT:    s_cbranch_execnz BB57_1
-; MOVREL-NEXT:  ; %bb.2:
-; MOVREL-NEXT:    s_mov_b64 exec, s[4:5]
-; MOVREL-NEXT:    v_mov_b32_e32 v0, v16
+; MOVREL-NEXT:    v_add_u32_e32 v15, vcc, 3, v15
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 2, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 3, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 4, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 5, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 6, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 7, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 8, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 9, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 10, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v10, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 11, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v11, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 12, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v12, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 13, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v13, vcc
+; MOVREL-NEXT:    v_cmp_eq_u32_e32 vcc, 14, v15
+; MOVREL-NEXT:    v_cndmask_b32_e32 v0, v0, v14, vcc
 ; MOVREL-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %add = add i32 %sel, 3
   %ext = extractelement <15 x float> %vec, i32 %add
   ret float %ext
 }
+
+define amdgpu_kernel void @dyn_extract_v4f32_s_s_s(float addrspace(1)* %out, i32 %sel) {
+; GPRIDX-LABEL: dyn_extract_v4f32_s_s_s:
+; GPRIDX:         .amd_kernel_code_t
+; GPRIDX-NEXT:     amd_code_version_major = 1
+; GPRIDX-NEXT:     amd_code_version_minor = 2
+; GPRIDX-NEXT:     amd_machine_kind = 1
+; GPRIDX-NEXT:     amd_machine_version_major = 9
+; GPRIDX-NEXT:     amd_machine_version_minor = 0
+; GPRIDX-NEXT:     amd_machine_version_stepping = 0
+; GPRIDX-NEXT:     kernel_code_entry_byte_offset = 256
+; GPRIDX-NEXT:     kernel_code_prefetch_byte_size = 0
+; GPRIDX-NEXT:     granulated_workitem_vgpr_count = 0
+; GPRIDX-NEXT:     granulated_wavefront_sgpr_count = 0
+; GPRIDX-NEXT:     priority = 0
+; GPRIDX-NEXT:     float_mode = 240
+; GPRIDX-NEXT:     priv = 0
+; GPRIDX-NEXT:     enable_dx10_clamp = 1
+; GPRIDX-NEXT:     debug_mode = 0
+; GPRIDX-NEXT:     enable_ieee_mode = 1
+; GPRIDX-NEXT:     enable_wgp_mode = 0
+; GPRIDX-NEXT:     enable_mem_ordered = 0
+; GPRIDX-NEXT:     enable_fwd_progress = 0
+; GPRIDX-NEXT:     enable_sgpr_private_segment_wave_byte_offset = 0
+; GPRIDX-NEXT:     user_sgpr_count = 6
+; GPRIDX-NEXT:     enable_trap_handler = 0
+; GPRIDX-NEXT:     enable_sgpr_workgroup_id_x = 1
+; GPRIDX-NEXT:     enable_sgpr_workgroup_id_y = 0
+; GPRIDX-NEXT:     enable_sgpr_workgroup_id_z = 0
+; GPRIDX-NEXT:     enable_sgpr_workgroup_info = 0
+; GPRIDX-NEXT:     enable_vgpr_workitem_id = 0
+; GPRIDX-NEXT:     enable_exception_msb = 0
+; GPRIDX-NEXT:     granulated_lds_size = 0
+; GPRIDX-NEXT:     enable_exception = 0
+; GPRIDX-NEXT:     enable_sgpr_private_segment_buffer = 1
+; GPRIDX-NEXT:     enable_sgpr_dispatch_ptr = 0
+; GPRIDX-NEXT:     enable_sgpr_queue_ptr = 0
+; GPRIDX-NEXT:     enable_sgpr_kernarg_segment_ptr = 1
+; GPRIDX-NEXT:     enable_sgpr_dispatch_id = 0
+; GPRIDX-NEXT:     enable_sgpr_flat_scratch_init = 0
+; GPRIDX-NEXT:     enable_sgpr_private_segment_size = 0
+; GPRIDX-NEXT:     enable_sgpr_grid_workgroup_count_x = 0
+; GPRIDX-NEXT:     enable_sgpr_grid_workgroup_count_y = 0
+; GPRIDX-NEXT:     enable_sgpr_grid_workgroup_count_z = 0
+; GPRIDX-NEXT:     enable_wavefront_size32 = 0
+; GPRIDX-NEXT:     enable_ordered_append_gds = 0
+; GPRIDX-NEXT:     private_element_size = 1
+; GPRIDX-NEXT:     is_ptr64 = 1
+; GPRIDX-NEXT:     is_dynamic_callstack = 0
+; GPRIDX-NEXT:     is_debug_enabled = 0
+; GPRIDX-NEXT:     is_xnack_enabled = 0
+; GPRIDX-NEXT:     workitem_private_segment_byte_size = 0
+; GPRIDX-NEXT:     workgroup_group_segment_byte_size = 0
+; GPRIDX-NEXT:     gds_segment_byte_size = 0
+; GPRIDX-NEXT:     kernarg_segment_byte_size = 28
+; GPRIDX-NEXT:     workgroup_fbarrier_count = 0
+; GPRIDX-NEXT:     wavefront_sgpr_count = 6
+; GPRIDX-NEXT:     workitem_vgpr_count = 3
+; GPRIDX-NEXT:     reserved_vgpr_first = 0
+; GPRIDX-NEXT:     reserved_vgpr_count = 0
+; GPRIDX-NEXT:     reserved_sgpr_first = 0
+; GPRIDX-NEXT:     reserved_sgpr_count = 0
+; GPRIDX-NEXT:     debug_wavefront_private_segment_offset_sgpr = 0
+; GPRIDX-NEXT:     debug_private_segment_buffer_sgpr = 0
+; GPRIDX-NEXT:     kernarg_segment_alignment = 4
+; GPRIDX-NEXT:     group_segment_alignment = 4
+; GPRIDX-NEXT:     private_segment_alignment = 4
+; GPRIDX-NEXT:     wavefront_size = 6
+; GPRIDX-NEXT:     call_convention = -1
+; GPRIDX-NEXT:     runtime_loader_kernel_symbol = 0
+; GPRIDX-NEXT:    .end_amd_kernel_code_t
+; GPRIDX-NEXT:  ; %bb.0: ; %entry
+; GPRIDX-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GPRIDX-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GPRIDX-NEXT:    s_waitcnt lgkmcnt(0)
+; GPRIDX-NEXT:    v_mov_b32_e32 v0, s0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s2, 1
+; GPRIDX-NEXT:    s_cselect_b32 s3, 2.0, 1.0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s2, 2
+; GPRIDX-NEXT:    s_cselect_b32 s3, 0x40400000, s3
+; GPRIDX-NEXT:    s_cmp_eq_u32 s2, 3
+; GPRIDX-NEXT:    s_cselect_b32 s2, 4.0, s3
+; GPRIDX-NEXT:    v_mov_b32_e32 v2, s2
+; GPRIDX-NEXT:    v_mov_b32_e32 v1, s1
+; GPRIDX-NEXT:    global_store_dword v[0:1], v2, off
+; GPRIDX-NEXT:    s_endpgm
+;
+; MOVREL-LABEL: dyn_extract_v4f32_s_s_s:
+; MOVREL:         .amd_kernel_code_t
+; MOVREL-NEXT:     amd_code_version_major = 1
+; MOVREL-NEXT:     amd_code_version_minor = 2
+; MOVREL-NEXT:     amd_machine_kind = 1
+; MOVREL-NEXT:     amd_machine_version_major = 8
+; MOVREL-NEXT:     amd_machine_version_minor = 0
+; MOVREL-NEXT:     amd_machine_version_stepping = 3
+; MOVREL-NEXT:     kernel_code_entry_byte_offset = 256
+; MOVREL-NEXT:     kernel_code_prefetch_byte_size = 0
+; MOVREL-NEXT:     granulated_workitem_vgpr_count = 0
+; MOVREL-NEXT:     granulated_wavefront_sgpr_count = 0
+; MOVREL-NEXT:     priority = 0
+; MOVREL-NEXT:     float_mode = 240
+; MOVREL-NEXT:     priv = 0
+; MOVREL-NEXT:     enable_dx10_clamp = 1
+; MOVREL-NEXT:     debug_mode = 0
+; MOVREL-NEXT:     enable_ieee_mode = 1
+; MOVREL-NEXT:     enable_wgp_mode = 0
+; MOVREL-NEXT:     enable_mem_ordered = 0
+; MOVREL-NEXT:     enable_fwd_progress = 0
+; MOVREL-NEXT:     enable_sgpr_private_segment_wave_byte_offset = 0
+; MOVREL-NEXT:     user_sgpr_count = 6
+; MOVREL-NEXT:     enable_trap_handler = 0
+; MOVREL-NEXT:     enable_sgpr_workgroup_id_x = 1
+; MOVREL-NEXT:     enable_sgpr_workgroup_id_y = 0
+; MOVREL-NEXT:     enable_sgpr_workgroup_id_z = 0
+; MOVREL-NEXT:     enable_sgpr_workgroup_info = 0
+; MOVREL-NEXT:     enable_vgpr_workitem_id = 0
+; MOVREL-NEXT:     enable_exception_msb = 0
+; MOVREL-NEXT:     granulated_lds_size = 0
+; MOVREL-NEXT:     enable_exception = 0
+; MOVREL-NEXT:     enable_sgpr_private_segment_buffer = 1
+; MOVREL-NEXT:     enable_sgpr_dispatch_ptr = 0
+; MOVREL-NEXT:     enable_sgpr_queue_ptr = 0
+; MOVREL-NEXT:     enable_sgpr_kernarg_segment_ptr = 1
+; MOVREL-NEXT:     enable_sgpr_dispatch_id = 0
+; MOVREL-NEXT:     enable_sgpr_flat_scratch_init = 0
+; MOVREL-NEXT:     enable_sgpr_private_segment_size = 0
+; MOVREL-NEXT:     enable_sgpr_grid_workgroup_count_x = 0
+; MOVREL-NEXT:     enable_sgpr_grid_workgroup_count_y = 0
+; MOVREL-NEXT:     enable_sgpr_grid_workgroup_count_z = 0
+; MOVREL-NEXT:     enable_wavefront_size32 = 0
+; MOVREL-NEXT:     enable_ordered_append_gds = 0
+; MOVREL-NEXT:     private_element_size = 1
+; MOVREL-NEXT:     is_ptr64 = 1
+; MOVREL-NEXT:     is_dynamic_callstack = 0
+; MOVREL-NEXT:     is_debug_enabled = 0
+; MOVREL-NEXT:     is_xnack_enabled = 0
+; MOVREL-NEXT:     workitem_private_segment_byte_size = 0
+; MOVREL-NEXT:     workgroup_group_segment_byte_size = 0
+; MOVREL-NEXT:     gds_segment_byte_size = 0
+; MOVREL-NEXT:     kernarg_segment_byte_size = 28
+; MOVREL-NEXT:     workgroup_fbarrier_count = 0
+; MOVREL-NEXT:     wavefront_sgpr_count = 6
+; MOVREL-NEXT:     workitem_vgpr_count = 3
+; MOVREL-NEXT:     reserved_vgpr_first = 0
+; MOVREL-NEXT:     reserved_vgpr_count = 0
+; MOVREL-NEXT:     reserved_sgpr_first = 0
+; MOVREL-NEXT:     reserved_sgpr_count = 0
+; MOVREL-NEXT:     debug_wavefront_private_segment_offset_sgpr = 0
+; MOVREL-NEXT:     debug_private_segment_buffer_sgpr = 0
+; MOVREL-NEXT:     kernarg_segment_alignment = 4
+; MOVREL-NEXT:     group_segment_alignment = 4
+; MOVREL-NEXT:     private_segment_alignment = 4
+; MOVREL-NEXT:     wavefront_size = 6
+; MOVREL-NEXT:     call_convention = -1
+; MOVREL-NEXT:     runtime_loader_kernel_symbol = 0
+; MOVREL-NEXT:    .end_amd_kernel_code_t
+; MOVREL-NEXT:  ; %bb.0: ; %entry
+; MOVREL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; MOVREL-NEXT:    s_load_dword s2, s[4:5], 0x8
+; MOVREL-NEXT:    s_waitcnt lgkmcnt(0)
+; MOVREL-NEXT:    v_mov_b32_e32 v0, s0
+; MOVREL-NEXT:    s_cmp_eq_u32 s2, 1
+; MOVREL-NEXT:    s_cselect_b32 s3, 2.0, 1.0
+; MOVREL-NEXT:    s_cmp_eq_u32 s2, 2
+; MOVREL-NEXT:    s_cselect_b32 s3, 0x40400000, s3
+; MOVREL-NEXT:    s_cmp_eq_u32 s2, 3
+; MOVREL-NEXT:    s_cselect_b32 s2, 4.0, s3
+; MOVREL-NEXT:    v_mov_b32_e32 v2, s2
+; MOVREL-NEXT:    v_mov_b32_e32 v1, s1
+; MOVREL-NEXT:    flat_store_dword v[0:1], v2
+; MOVREL-NEXT:    s_endpgm
+entry:
+  %ext = extractelement <4 x float> <float 1.0, float 2.0, float 3.0, float 4.0>, i32 %sel
+  store float %ext, float addrspace(1)* %out
+  ret void
+}
+
+define amdgpu_kernel void @dyn_extract_v4f64_s_s_s(double addrspace(1)* %out, i32 %sel) {
+; GPRIDX-LABEL: dyn_extract_v4f64_s_s_s:
+; GPRIDX:         .amd_kernel_code_t
+; GPRIDX-NEXT:     amd_code_version_major = 1
+; GPRIDX-NEXT:     amd_code_version_minor = 2
+; GPRIDX-NEXT:     amd_machine_kind = 1
+; GPRIDX-NEXT:     amd_machine_version_major = 9
+; GPRIDX-NEXT:     amd_machine_version_minor = 0
+; GPRIDX-NEXT:     amd_machine_version_stepping = 0
+; GPRIDX-NEXT:     kernel_code_entry_byte_offset = 256
+; GPRIDX-NEXT:     kernel_code_prefetch_byte_size = 0
+; GPRIDX-NEXT:     granulated_workitem_vgpr_count = 0
+; GPRIDX-NEXT:     granulated_wavefront_sgpr_count = 0
+; GPRIDX-NEXT:     priority = 0
+; GPRIDX-NEXT:     float_mode = 240
+; GPRIDX-NEXT:     priv = 0
+; GPRIDX-NEXT:     enable_dx10_clamp = 1
+; GPRIDX-NEXT:     debug_mode = 0
+; GPRIDX-NEXT:     enable_ieee_mode = 1
+; GPRIDX-NEXT:     enable_wgp_mode = 0
+; GPRIDX-NEXT:     enable_mem_ordered = 0
+; GPRIDX-NEXT:     enable_fwd_progress = 0
+; GPRIDX-NEXT:     enable_sgpr_private_segment_wave_byte_offset = 0
+; GPRIDX-NEXT:     user_sgpr_count = 6
+; GPRIDX-NEXT:     enable_trap_handler = 0
+; GPRIDX-NEXT:     enable_sgpr_workgroup_id_x = 1
+; GPRIDX-NEXT:     enable_sgpr_workgroup_id_y = 0
+; GPRIDX-NEXT:     enable_sgpr_workgroup_id_z = 0
+; GPRIDX-NEXT:     enable_sgpr_workgroup_info = 0
+; GPRIDX-NEXT:     enable_vgpr_workitem_id = 0
+; GPRIDX-NEXT:     enable_exception_msb = 0
+; GPRIDX-NEXT:     granulated_lds_size = 0
+; GPRIDX-NEXT:     enable_exception = 0
+; GPRIDX-NEXT:     enable_sgpr_private_segment_buffer = 1
+; GPRIDX-NEXT:     enable_sgpr_dispatch_ptr = 0
+; GPRIDX-NEXT:     enable_sgpr_queue_ptr = 0
+; GPRIDX-NEXT:     enable_sgpr_kernarg_segment_ptr = 1
+; GPRIDX-NEXT:     enable_sgpr_dispatch_id = 0
+; GPRIDX-NEXT:     enable_sgpr_flat_scratch_init = 0
+; GPRIDX-NEXT:     enable_sgpr_private_segment_size = 0
+; GPRIDX-NEXT:     enable_sgpr_grid_workgroup_count_x = 0
+; GPRIDX-NEXT:     enable_sgpr_grid_workgroup_count_y = 0
+; GPRIDX-NEXT:     enable_sgpr_grid_workgroup_count_z = 0
+; GPRIDX-NEXT:     enable_wavefront_size32 = 0
+; GPRIDX-NEXT:     enable_ordered_append_gds = 0
+; GPRIDX-NEXT:     private_element_size = 1
+; GPRIDX-NEXT:     is_ptr64 = 1
+; GPRIDX-NEXT:     is_dynamic_callstack = 0
+; GPRIDX-NEXT:     is_debug_enabled = 0
+; GPRIDX-NEXT:     is_xnack_enabled = 0
+; GPRIDX-NEXT:     workitem_private_segment_byte_size = 0
+; GPRIDX-NEXT:     workgroup_group_segment_byte_size = 0
+; GPRIDX-NEXT:     gds_segment_byte_size = 0
+; GPRIDX-NEXT:     kernarg_segment_byte_size = 28
+; GPRIDX-NEXT:     workgroup_fbarrier_count = 0
+; GPRIDX-NEXT:     wavefront_sgpr_count = 7
+; GPRIDX-NEXT:     workitem_vgpr_count = 4
+; GPRIDX-NEXT:     reserved_vgpr_first = 0
+; GPRIDX-NEXT:     reserved_vgpr_count = 0
+; GPRIDX-NEXT:     reserved_sgpr_first = 0
+; GPRIDX-NEXT:     reserved_sgpr_count = 0
+; GPRIDX-NEXT:     debug_wavefront_private_segment_offset_sgpr = 0
+; GPRIDX-NEXT:     debug_private_segment_buffer_sgpr = 0
+; GPRIDX-NEXT:     kernarg_segment_alignment = 4
+; GPRIDX-NEXT:     group_segment_alignment = 4
+; GPRIDX-NEXT:     private_segment_alignment = 4
+; GPRIDX-NEXT:     wavefront_size = 6
+; GPRIDX-NEXT:     call_convention = -1
+; GPRIDX-NEXT:     runtime_loader_kernel_symbol = 0
+; GPRIDX-NEXT:    .end_amd_kernel_code_t
+; GPRIDX-NEXT:  ; %bb.0: ; %entry
+; GPRIDX-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
+; GPRIDX-NEXT:    s_load_dword s6, s[4:5], 0x8
+; GPRIDX-NEXT:    s_mov_b32 s0, 0
+; GPRIDX-NEXT:    s_mov_b32 s1, 0x40080000
+; GPRIDX-NEXT:    s_waitcnt lgkmcnt(0)
+; GPRIDX-NEXT:    v_mov_b32_e32 v2, s2
+; GPRIDX-NEXT:    s_cmp_eq_u32 s6, 1
+; GPRIDX-NEXT:    s_cselect_b64 s[4:5], 2.0, 1.0
+; GPRIDX-NEXT:    s_cmp_eq_u32 s6, 2
+; GPRIDX-NEXT:    s_cselect_b64 s[0:1], s[0:1], s[4:5]
+; GPRIDX-NEXT:    s_cmp_eq_u32 s6, 3
+; GPRIDX-NEXT:    s_cselect_b64 s[0:1], 4.0, s[0:1]
+; GPRIDX-NEXT:    v_mov_b32_e32 v0, s0
+; GPRIDX-NEXT:    v_mov_b32_e32 v1, s1
+; GPRIDX-NEXT:    v_mov_b32_e32 v3, s3
+; GPRIDX-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GPRIDX-NEXT:    s_endpgm
+;
+; MOVREL-LABEL: dyn_extract_v4f64_s_s_s:
+; MOVREL:         .amd_kernel_code_t
+; MOVREL-NEXT:     amd_code_version_major = 1
+; MOVREL-NEXT:     amd_code_version_minor = 2
+; MOVREL-NEXT:     amd_machine_kind = 1
+; MOVREL-NEXT:     amd_machine_version_major = 8
+; MOVREL-NEXT:     amd_machine_version_minor = 0
+; MOVREL-NEXT:     amd_machine_version_stepping = 3
+; MOVREL-NEXT:     kernel_code_entry_byte_offset = 256
+; MOVREL-NEXT:     kernel_code_prefetch_byte_size = 0
+; MOVREL-NEXT:     granulated_workitem_vgpr_count = 0
+; MOVREL-NEXT:     granulated_wavefront_sgpr_count = 0
+; MOVREL-NEXT:     priority = 0
+; MOVREL-NEXT:     float_mode = 240
+; MOVREL-NEXT:     priv = 0
+; MOVREL-NEXT:     enable_dx10_clamp = 1
+; MOVREL-NEXT:     debug_mode = 0
+; MOVREL-NEXT:     enable_ieee_mode = 1
+; MOVREL-NEXT:     enable_wgp_mode = 0
+; MOVREL-NEXT:     enable_mem_ordered = 0
+; MOVREL-NEXT:     enable_fwd_progress = 0
+; MOVREL-NEXT:     enable_sgpr_private_segment_wave_byte_offset = 0
+; MOVREL-NEXT:     user_sgpr_count = 6
+; MOVREL-NEXT:     enable_trap_handler = 0
+; MOVREL-NEXT:     enable_sgpr_workgroup_id_x = 1
+; MOVREL-NEXT:     enable_sgpr_workgroup_id_y = 0
+; MOVREL-NEXT:     enable_sgpr_workgroup_id_z = 0
+; MOVREL-NEXT:     enable_sgpr_workgroup_info = 0
+; MOVREL-NEXT:     enable_vgpr_workitem_id = 0
+; MOVREL-NEXT:     enable_exception_msb = 0
+; MOVREL-NEXT:     granulated_lds_size = 0
+; MOVREL-NEXT:     enable_exception = 0
+; MOVREL-NEXT:     enable_sgpr_private_segment_buffer = 1
+; MOVREL-NEXT:     enable_sgpr_dispatch_ptr = 0
+; MOVREL-NEXT:     enable_sgpr_queue_ptr = 0
+; MOVREL-NEXT:     enable_sgpr_kernarg_segment_ptr = 1
+; MOVREL-NEXT:     enable_sgpr_dispatch_id = 0
+; MOVREL-NEXT:     enable_sgpr_flat_scratch_init = 0
+; MOVREL-NEXT:     enable_sgpr_private_segment_size = 0
+; MOVREL-NEXT:     enable_sgpr_grid_workgroup_count_x = 0
+; MOVREL-NEXT:     enable_sgpr_grid_workgroup_count_y = 0
+; MOVREL-NEXT:     enable_sgpr_grid_workgroup_count_z = 0
+; MOVREL-NEXT:     enable_wavefront_size32 = 0
+; MOVREL-NEXT:     enable_ordered_append_gds = 0
+; MOVREL-NEXT:     private_element_size = 1
+; MOVREL-NEXT:     is_ptr64 = 1
+; MOVREL-NEXT:     is_dynamic_callstack = 0
+; MOVREL-NEXT:     is_debug_enabled = 0
+; MOVREL-NEXT:     is_xnack_enabled = 0
+; MOVREL-NEXT:     workitem_private_segment_byte_size = 0
+; MOVREL-NEXT:     workgroup_group_segment_byte_size = 0
+; MOVREL-NEXT:     gds_segment_byte_size = 0
+; MOVREL-NEXT:     kernarg_segment_byte_size = 28
+; MOVREL-NEXT:     workgroup_fbarrier_count = 0
+; MOVREL-NEXT:     wavefront_sgpr_count = 7
+; MOVREL-NEXT:     workitem_vgpr_count = 4
+; MOVREL-NEXT:     reserved_vgpr_first = 0
+; MOVREL-NEXT:     reserved_vgpr_count = 0
+; MOVREL-NEXT:     reserved_sgpr_first = 0
+; MOVREL-NEXT:     reserved_sgpr_count = 0
+; MOVREL-NEXT:     debug_wavefront_private_segment_offset_sgpr = 0
+; MOVREL-NEXT:     debug_private_segment_buffer_sgpr = 0
+; MOVREL-NEXT:     kernarg_segment_alignment = 4
+; MOVREL-NEXT:     group_segment_alignment = 4
+; MOVREL-NEXT:     private_segment_alignment = 4
+; MOVREL-NEXT:     wavefront_size = 6
+; MOVREL-NEXT:     call_convention = -1
+; MOVREL-NEXT:     runtime_loader_kernel_symbol = 0
+; MOVREL-NEXT:    .end_amd_kernel_code_t
+; MOVREL-NEXT:  ; %bb.0: ; %entry
+; MOVREL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
+; MOVREL-NEXT:    s_load_dword s6, s[4:5], 0x8
+; MOVREL-NEXT:    s_mov_b32 s0, 0
+; MOVREL-NEXT:    s_mov_b32 s1, 0x40080000
+; MOVREL-NEXT:    s_waitcnt lgkmcnt(0)
+; MOVREL-NEXT:    v_mov_b32_e32 v2, s2
+; MOVREL-NEXT:    s_cmp_eq_u32 s6, 1
+; MOVREL-NEXT:    s_cselect_b64 s[4:5], 2.0, 1.0
+; MOVREL-NEXT:    s_cmp_eq_u32 s6, 2
+; MOVREL-NEXT:    s_cselect_b64 s[0:1], s[0:1], s[4:5]
+; MOVREL-NEXT:    s_cmp_eq_u32 s6, 3
+; MOVREL-NEXT:    s_cselect_b64 s[0:1], 4.0, s[0:1]
+; MOVREL-NEXT:    v_mov_b32_e32 v0, s0
+; MOVREL-NEXT:    v_mov_b32_e32 v1, s1
+; MOVREL-NEXT:    v_mov_b32_e32 v3, s3
+; MOVREL-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; MOVREL-NEXT:    s_endpgm
+entry:
+  %ext = extractelement <4 x double> <double 1.0, double 2.0, double 3.0, double 4.0>, i32 %sel
+  store double %ext, double addrspace(1)* %out
+  ret void
+}

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

 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
 
     ; WAVE64-LABEL: name: extract_vector_elt_v16s32_sv
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE64: [[COPY:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
-    ; WAVE64: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF1]], %bb.0, %8, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %2(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE64: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: $vgpr0 = COPY [[V_MOV_B32_e32_]](s32)
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE64: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C7]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE64: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE64: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C8]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE64: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE64: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C9]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE64: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE64: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C10]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE64: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE64: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C11]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE64: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE64: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C12]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE64: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE64: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C13]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE64: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE64: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C14]]
+    ; WAVE64: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE64: $vgpr0 = COPY [[COPY2]](s32)
     ; WAVE32-LABEL: name: extract_vector_elt_v16s32_sv
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE32: [[COPY:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
-    ; WAVE32: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF1]], %bb.0, %8, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %2(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE32: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: $vgpr0 = COPY [[V_MOV_B32_e32_]](s32)
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE32: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C7]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE32: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE32: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C8]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE32: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE32: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C9]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE32: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE32: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C10]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE32: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE32: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C11]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE32: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE32: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C12]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE32: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE32: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C13]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE32: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE32: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C14]]
+    ; WAVE32: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE32: $vgpr0 = COPY [[COPY2]](s32)
     %0:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
     %1:_(s32) = COPY $vgpr0
     %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
     $vgpr0 = COPY %2
 ...
 
 ---
 name: extract_vector_elt_v16s32_vs

 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
 
     ; WAVE64-LABEL: name: extract_vector_elt_v16s32_vv
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
-    ; WAVE64: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF1]], %bb.0, %8, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %2(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE64: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C7]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE64: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE64: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C8]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE64: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE64: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C9]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE64: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE64: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C10]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE64: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE64: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C11]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE64: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE64: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C12]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE64: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE64: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C13]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE64: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE64: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C14]]
+    ; WAVE64: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE64: $vgpr0 = COPY [[COPY2]](s32)
     ; WAVE32-LABEL: name: extract_vector_elt_v16s32_vv
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE32: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
-    ; WAVE32: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF1]], %bb.0, %8, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %2(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE32: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C7]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE32: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE32: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C8]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE32: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE32: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C9]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE32: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE32: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C10]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE32: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE32: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C11]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE32: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE32: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C12]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE32: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE32: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C13]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE32: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE32: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C14]]
+    ; WAVE32: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE32: $vgpr0 = COPY [[COPY2]](s32)
     %0:_(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     %1:_(s32) = COPY $vgpr16
     %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
     $vgpr0 = COPY %2
 ...
 
 ---
 name: extract_vector_elt_v8s64_ss

 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
 
     ; WAVE64-LABEL: name: extract_vector_elt_v8s64_sv
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE64: [[COPY:%[0-9]+]]:sgpr(<8 x s64>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
-    ; WAVE64: [[BITCAST:%[0-9]+]]:sgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF4:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF4]], %bb.0, %20, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE64: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %8(s32), %bb.1
-    ; WAVE64: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %3(s32), %bb.1
-    ; WAVE64: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[V_READFIRSTLANE_B32_]], [[C]](s32)
-    ; WAVE64: [[ADD:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C]]
-    ; WAVE64: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE64: [[EVEC1:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD]](s32)
-    ; WAVE64: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE64: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC1]](s32), implicit $exec
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[V_MOV_B32_e32_]](s32), [[V_MOV_B32_e32_1]](s32)
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; WAVE64: $vgpr0_vgpr1 = COPY [[MV]](s64)
     ; WAVE32-LABEL: name: extract_vector_elt_v8s64_sv
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE32: [[COPY:%[0-9]+]]:sgpr(<8 x s64>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
-    ; WAVE32: [[BITCAST:%[0-9]+]]:sgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF4:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF4]], %bb.0, %20, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE32: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %8(s32), %bb.1
-    ; WAVE32: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %3(s32), %bb.1
-    ; WAVE32: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[V_READFIRSTLANE_B32_]], [[C]](s32)
-    ; WAVE32: [[ADD:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C]]
-    ; WAVE32: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE32: [[EVEC1:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD]](s32)
-    ; WAVE32: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE32: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC1]](s32), implicit $exec
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[V_MOV_B32_e32_]](s32), [[V_MOV_B32_e32_1]](s32)
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; WAVE32: $vgpr0_vgpr1 = COPY [[MV]](s64)
     %0:_(<8 x s64>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
     %1:_(s32) = COPY $vgpr0
     %2:_(s64) = G_EXTRACT_VECTOR_ELT %0, %1
     $vgpr0_vgpr1 = COPY %2
 ...
 
 ---
 name: extract_vector_elt_v8s64_vv
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
 
     ; WAVE64-LABEL: name: extract_vector_elt_v8s64_vv
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr(<8 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
-    ; WAVE64: [[BITCAST:%[0-9]+]]:vgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF4:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF4]], %bb.0, %20, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE64: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %8(s32), %bb.1
-    ; WAVE64: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %3(s32), %bb.1
-    ; WAVE64: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[V_READFIRSTLANE_B32_]], [[C]](s32)
-    ; WAVE64: [[ADD:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C]]
-    ; WAVE64: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE64: [[EVEC1:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD]](s32)
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[EVEC]](s32), [[EVEC1]](s32)
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; WAVE64: $vgpr0_vgpr1 = COPY [[MV]](s64)
     ; WAVE32-LABEL: name: extract_vector_elt_v8s64_vv
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE32: [[COPY:%[0-9]+]]:vgpr(<8 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
-    ; WAVE32: [[BITCAST:%[0-9]+]]:vgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF4:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF4]], %bb.0, %20, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE32: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %8(s32), %bb.1
-    ; WAVE32: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %3(s32), %bb.1
-    ; WAVE32: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[V_READFIRSTLANE_B32_]], [[C]](s32)
-    ; WAVE32: [[ADD:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C]]
-    ; WAVE32: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE32: [[EVEC1:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD]](s32)
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[EVEC]](s32), [[EVEC1]](s32)
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C1]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C2]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C3]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C4]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C5]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C6]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; WAVE32: $vgpr0_vgpr1 = COPY [[MV]](s64)
     %0:_(<8 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     %1:_(s32) = COPY $vgpr16
     %2:_(s64) = G_EXTRACT_VECTOR_ELT %0, %1
     $vgpr0_vgpr1 = COPY %2
 ...
 
 ---
 name: extract_vector_elt_v16s32_vv_idx_add1
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
 
     ; WAVE64-LABEL: name: extract_vector_elt_v16s32_vv_idx_add1
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE64: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE64: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
     ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C1]]
-    ; WAVE64: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[ADD1]](s32)
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE64: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE64: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE64: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE64: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE64: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE64: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE64: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE64: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE64: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE64: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE64: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE64: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE64: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE64: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE64: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE64: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE64: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE64: $vgpr0 = COPY [[COPY3]](s32)
     ; WAVE32-LABEL: name: extract_vector_elt_v16s32_vv_idx_add1
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE32: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE32: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE32: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
     ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C1]]
-    ; WAVE32: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[ADD1]](s32)
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE32: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE32: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE32: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE32: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE32: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE32: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE32: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE32: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE32: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE32: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE32: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE32: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE32: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE32: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE32: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE32: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE32: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE32: $vgpr0 = COPY [[COPY3]](s32)
     %0:_(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     %1:_(s32) = COPY $vgpr16
     %2:_(s32) = G_CONSTANT i32 1
     %3:_(s32) = G_ADD %1, %2
     %4:_(s32) = G_EXTRACT_VECTOR_ELT %0, %3
     $vgpr0 = COPY %4
 ...
 
 ---
 name: extract_vector_elt_v16s32_vv_idx_addm1
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
 
     ; WAVE64-LABEL: name: extract_vector_elt_v16s32_vv_idx_addm1
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -1
     ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
-    ; WAVE64: [[ADD:%[0-9]+]]:vgpr_32(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE64: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[ADD]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[ADD]](s32), implicit $exec
-    ; WAVE64: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE64: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE64: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE64: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE64: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE64: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE64: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE64: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE64: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE64: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE64: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE64: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE64: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE64: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE64: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE64: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE64: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE64: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE64: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE64: $vgpr0 = COPY [[COPY3]](s32)
     ; WAVE32-LABEL: name: extract_vector_elt_v16s32_vv_idx_addm1
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE32: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -1
     ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
-    ; WAVE32: [[ADD:%[0-9]+]]:vgpr_32(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE32: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[ADD]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[ADD]](s32), implicit $exec
-    ; WAVE32: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE32: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE32: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE32: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE32: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE32: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE32: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE32: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE32: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE32: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE32: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE32: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE32: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE32: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE32: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE32: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE32: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE32: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE32: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE32: $vgpr0 = COPY [[COPY3]](s32)
     %0:_(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     %1:_(s32) = COPY $vgpr16
     %2:_(s32) = G_CONSTANT i32 -1
     %3:_(s32) = G_ADD %1, %2
     %4:_(s32) = G_EXTRACT_VECTOR_ELT %0, %3
     $vgpr0 = COPY %4
 ...
 
 ---
 name: extract_vector_elt_v16s32_vv_idx_add16
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
 
     ; WAVE64-LABEL: name: extract_vector_elt_v16s32_vv_idx_add16
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
     ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
-    ; WAVE64: [[ADD:%[0-9]+]]:vgpr_32(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE64: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[ADD]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[ADD]](s32), implicit $exec
-    ; WAVE64: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE64: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE64: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE64: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE64: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE64: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE64: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE64: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE64: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE64: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE64: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE64: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE64: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE64: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE64: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE64: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE64: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE64: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE64: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE64: $vgpr0 = COPY [[COPY3]](s32)
     ; WAVE32-LABEL: name: extract_vector_elt_v16s32_vv_idx_add16
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE32: [[COPY:%[0-9]+]]:vgpr(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16
     ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
-    ; WAVE32: [[ADD:%[0-9]+]]:vgpr_32(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE32: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[ADD]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[ADD]](s32), implicit $exec
-    ; WAVE32: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[V_READFIRSTLANE_B32_]](s32)
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: $vgpr0 = COPY [[EVEC]](s32)
+    ; WAVE32: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
+    ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE32: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE32: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE32: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE32: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE32: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE32: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE32: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE32: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE32: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE32: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE32: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE32: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE32: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE32: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE32: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE32: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE32: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE32: $vgpr0 = COPY [[COPY3]](s32)
     %0:_(<16 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     %1:_(s32) = COPY $vgpr16
     %2:_(s32) = G_CONSTANT i32 16
     %3:_(s32) = G_ADD %1, %2
     %4:_(s32) = G_EXTRACT_VECTOR_ELT %0, %3
     $vgpr0 = COPY %4
 ...
 
 ---
 name: extract_vector_elt_v8s64_vv_idx_add1
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
 
     ; WAVE64-LABEL: name: extract_vector_elt_v8s64_vv_idx_add1
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr(<8 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE64: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE64: [[BITCAST:%[0-9]+]]:vgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF4:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF4]], %bb.0, %23, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %10(s32), %bb.1
-    ; WAVE64: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %11(s32), %bb.1
-    ; WAVE64: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %6(s32), %bb.1
-    ; WAVE64: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C2]]
-    ; WAVE64: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[ADD1]], [[C1]](s32)
-    ; WAVE64: [[ADD2:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C1]]
-    ; WAVE64: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE64: [[EVEC1:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD2]](s32)
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[EVEC]](s32), [[EVEC1]](s32)
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE64: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY4]](s32)
     ; WAVE64: $vgpr0_vgpr1 = COPY [[MV]](s64)
     ; WAVE32-LABEL: name: extract_vector_elt_v8s64_vv_idx_add1
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16
     ; WAVE32: [[COPY:%[0-9]+]]:vgpr(<8 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr16
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr16
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE32: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE32: [[BITCAST:%[0-9]+]]:vgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF4:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF4]], %bb.0, %23, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %10(s32), %bb.1
-    ; WAVE32: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %11(s32), %bb.1
-    ; WAVE32: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %6(s32), %bb.1
-    ; WAVE32: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C2]]
-    ; WAVE32: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[ADD1]], [[C1]](s32)
-    ; WAVE32: [[ADD2:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C1]]
-    ; WAVE32: [[EVEC:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE32: [[EVEC1:%[0-9]+]]:vgpr(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD2]](s32)
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[EVEC]](s32), [[EVEC1]](s32)
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE32: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY4]](s32)
     ; WAVE32: $vgpr0_vgpr1 = COPY [[MV]](s64)
     %0:_(<8 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
     %1:_(s32) = COPY $vgpr16
     %2:_(s32) = G_CONSTANT i32 1
     %3:_(s32) = G_ADD %1, %2
     %4:_(s64) = G_EXTRACT_VECTOR_ELT %0, %3
     $vgpr0_vgpr1 = COPY %4
 ...
 
 ---
 name: extract_vector_elt_v16s32_sv_idx_add1
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
 
     ; WAVE64-LABEL: name: extract_vector_elt_v16s32_sv_idx_add1
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE64: [[COPY:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE64: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE64: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
     ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C1]]
-    ; WAVE64: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[ADD1]](s32)
-    ; WAVE64: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: $vgpr0 = COPY [[V_MOV_B32_e32_]](s32)
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE64: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE64: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE64: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE64: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE64: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE64: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE64: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE64: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE64: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE64: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE64: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE64: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE64: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE64: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE64: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE64: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE64: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE64: $vgpr0 = COPY [[COPY3]](s32)
     ; WAVE32-LABEL: name: extract_vector_elt_v16s32_sv_idx_add1
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE32: [[COPY:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE32: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE32: [[DEF:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF1]], %bb.0, %11, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %4(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<16 x s32>)
     ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C1]]
-    ; WAVE32: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<16 x s32>), [[ADD1]](s32)
-    ; WAVE32: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: $vgpr0 = COPY [[V_MOV_B32_e32_]](s32)
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV1]], [[UV]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV2]], [[SELECT]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV3]], [[SELECT1]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV4]], [[SELECT2]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV5]], [[SELECT3]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV6]], [[SELECT4]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV7]], [[SELECT5]]
+    ; WAVE32: [[C8:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 8
+    ; WAVE32: [[ICMP7:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C8]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP7]](s1), [[UV8]], [[SELECT6]]
+    ; WAVE32: [[C9:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 9
+    ; WAVE32: [[ICMP8:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C9]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP8]](s1), [[UV9]], [[SELECT7]]
+    ; WAVE32: [[C10:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 10
+    ; WAVE32: [[ICMP9:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C10]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP9]](s1), [[UV10]], [[SELECT8]]
+    ; WAVE32: [[C11:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 11
+    ; WAVE32: [[ICMP10:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C11]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP10]](s1), [[UV11]], [[SELECT9]]
+    ; WAVE32: [[C12:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
+    ; WAVE32: [[ICMP11:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C12]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP11]](s1), [[UV12]], [[SELECT10]]
+    ; WAVE32: [[C13:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 13
+    ; WAVE32: [[ICMP12:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C13]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP12]](s1), [[UV13]], [[SELECT11]]
+    ; WAVE32: [[C14:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 14
+    ; WAVE32: [[ICMP13:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C14]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP13]](s1), [[UV14]], [[SELECT12]]
+    ; WAVE32: [[C15:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 15
+    ; WAVE32: [[ICMP14:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C15]]
+    ; WAVE32: [[SELECT14:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP14]](s1), [[UV15]], [[SELECT13]]
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT14]](s32)
+    ; WAVE32: $vgpr0 = COPY [[COPY3]](s32)
     %0:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
     %1:_(s32) = COPY $vgpr0
     %2:_(s32) = G_CONSTANT i32 1
     %3:_(s32) = G_ADD %1, %2
     %4:_(s32) = G_EXTRACT_VECTOR_ELT %0, %3
     $vgpr0 = COPY %4
 ...
 
 ---
 name: extract_vector_elt_v8s64_sv_add1
 legalized: true
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
 
     ; WAVE64-LABEL: name: extract_vector_elt_v8s64_sv_add1
-    ; WAVE64: successors: %bb.1(0x80000000)
     ; WAVE64: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE64: [[COPY:%[0-9]+]]:sgpr(<8 x s64>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; WAVE64: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE64: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE64: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE64: [[BITCAST:%[0-9]+]]:sgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE64: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE64: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE64: [[DEF4:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-    ; WAVE64: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-    ; WAVE64: .1:
-    ; WAVE64: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE64: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF4]], %bb.0, %23, %bb.1
-    ; WAVE64: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %10(s32), %bb.1
-    ; WAVE64: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %11(s32), %bb.1
-    ; WAVE64: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %6(s32), %bb.1
-    ; WAVE64: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE64: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE64: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C2]]
-    ; WAVE64: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[ADD1]], [[C1]](s32)
-    ; WAVE64: [[ADD2:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C1]]
-    ; WAVE64: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE64: [[EVEC1:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD2]](s32)
-    ; WAVE64: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE64: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC1]](s32), implicit $exec
-    ; WAVE64: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE64: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-    ; WAVE64: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE64: .2:
-    ; WAVE64: successors: %bb.3(0x80000000)
-    ; WAVE64: $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-    ; WAVE64: .3:
-    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[V_MOV_B32_e32_]](s32), [[V_MOV_B32_e32_1]](s32)
+    ; WAVE64: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE64: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE64: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE64: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE64: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE64: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE64: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE64: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE64: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE64: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE64: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE64: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE64: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE64: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE64: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE64: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE64: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE64: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE64: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE64: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE64: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE64: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE64: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE64: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE64: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE64: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE64: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE64: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE64: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE64: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY4]](s32)
     ; WAVE64: $vgpr0_vgpr1 = COPY [[MV]](s64)
     ; WAVE32-LABEL: name: extract_vector_elt_v8s64_sv_add1
-    ; WAVE32: successors: %bb.1(0x80000000)
     ; WAVE32: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, $vgpr0
     ; WAVE32: [[COPY:%[0-9]+]]:sgpr(<8 x s64>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
-    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; WAVE32: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
     ; WAVE32: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; WAVE32: [[ADD:%[0-9]+]]:vgpr(s32) = G_ADD [[COPY1]], [[COPY2]]
-    ; WAVE32: [[BITCAST:%[0-9]+]]:sgpr(<16 x s32>) = G_BITCAST [[COPY]](<8 x s64>)
+    ; WAVE32: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32), [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32), [[UV4:%[0-9]+]]:vgpr(s32), [[UV5:%[0-9]+]]:vgpr(s32), [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32), [[UV8:%[0-9]+]]:vgpr(s32), [[UV9:%[0-9]+]]:vgpr(s32), [[UV10:%[0-9]+]]:vgpr(s32), [[UV11:%[0-9]+]]:vgpr(s32), [[UV12:%[0-9]+]]:vgpr(s32), [[UV13:%[0-9]+]]:vgpr(s32), [[UV14:%[0-9]+]]:vgpr(s32), [[UV15:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY]](<8 x s64>)
     ; WAVE32: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[DEF:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF1:%[0-9]+]]:sgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF2:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF3:%[0-9]+]]:vgpr(s32) = G_IMPLICIT_DEF
-    ; WAVE32: [[DEF4:%[0-9]+]]:sreg_32_xm0_xexec = IMPLICIT_DEF
-    ; WAVE32: [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
-    ; WAVE32: .1:
-    ; WAVE32: successors: %bb.2(0x40000000), %bb.1(0x40000000)
-    ; WAVE32: [[PHI:%[0-9]+]]:sreg_32_xm0_xexec = PHI [[DEF4]], %bb.0, %23, %bb.1
-    ; WAVE32: [[PHI1:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF]](s32), %bb.0, %10(s32), %bb.1
-    ; WAVE32: [[PHI2:%[0-9]+]]:sgpr(s32) = G_PHI [[DEF1]](s32), %bb.0, %11(s32), %bb.1
-    ; WAVE32: [[PHI3:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF2]](s32), %bb.0, %6(s32), %bb.1
-    ; WAVE32: [[PHI4:%[0-9]+]]:vgpr(s32) = G_PHI [[DEF3]](s32), %bb.0, %7(s32), %bb.1
-    ; WAVE32: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY1]](s32), implicit $exec
-    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
-    ; WAVE32: [[ADD1:%[0-9]+]]:sgpr(s32) = G_ADD [[V_READFIRSTLANE_B32_]], [[C2]]
-    ; WAVE32: [[SHL:%[0-9]+]]:sgpr(s32) = G_SHL [[ADD1]], [[C1]](s32)
-    ; WAVE32: [[ADD2:%[0-9]+]]:sgpr(s32) = G_ADD [[SHL]], [[C1]]
-    ; WAVE32: [[EVEC:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[SHL]](s32)
-    ; WAVE32: [[EVEC1:%[0-9]+]]:sreg_32(s32) = G_EXTRACT_VECTOR_ELT [[BITCAST]](<16 x s32>), [[ADD2]](s32)
-    ; WAVE32: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC]](s32), implicit $exec
-    ; WAVE32: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32(s32) = V_MOV_B32_e32 [[EVEC1]](s32), implicit $exec
-    ; WAVE32: [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-    ; WAVE32: $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
-    ; WAVE32: S_CBRANCH_EXECNZ %bb.1, implicit $exec
-    ; WAVE32: .2:
-    ; WAVE32: successors: %bb.3(0x80000000)
-    ; WAVE32: $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
-    ; WAVE32: .3:
-    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[V_MOV_B32_e32_]](s32), [[V_MOV_B32_e32_1]](s32)
+    ; WAVE32: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C1]]
+    ; WAVE32: [[SELECT:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV2]], [[UV]]
+    ; WAVE32: [[SELECT1:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP]](s1), [[UV3]], [[UV1]]
+    ; WAVE32: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 2
+    ; WAVE32: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C2]]
+    ; WAVE32: [[SELECT2:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV4]], [[SELECT]]
+    ; WAVE32: [[SELECT3:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP1]](s1), [[UV5]], [[SELECT1]]
+    ; WAVE32: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 3
+    ; WAVE32: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C3]]
+    ; WAVE32: [[SELECT4:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV6]], [[SELECT2]]
+    ; WAVE32: [[SELECT5:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP2]](s1), [[UV7]], [[SELECT3]]
+    ; WAVE32: [[C4:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 4
+    ; WAVE32: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C4]]
+    ; WAVE32: [[SELECT6:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV8]], [[SELECT4]]
+    ; WAVE32: [[SELECT7:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP3]](s1), [[UV9]], [[SELECT5]]
+    ; WAVE32: [[C5:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 5
+    ; WAVE32: [[ICMP4:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C5]]
+    ; WAVE32: [[SELECT8:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV10]], [[SELECT6]]
+    ; WAVE32: [[SELECT9:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP4]](s1), [[UV11]], [[SELECT7]]
+    ; WAVE32: [[C6:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 6
+    ; WAVE32: [[ICMP5:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C6]]
+    ; WAVE32: [[SELECT10:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV12]], [[SELECT8]]
+    ; WAVE32: [[SELECT11:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP5]](s1), [[UV13]], [[SELECT9]]
+    ; WAVE32: [[C7:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 7
+    ; WAVE32: [[ICMP6:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[ADD]](s32), [[C7]]
+    ; WAVE32: [[SELECT12:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV14]], [[SELECT10]]
+    ; WAVE32: [[SELECT13:%[0-9]+]]:vgpr(s32) = G_SELECT [[ICMP6]](s1), [[UV15]], [[SELECT11]]
+    ; WAVE32: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[SELECT12]](s32)
+    ; WAVE32: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[SELECT13]](s32)
+    ; WAVE32: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY4]](s32)
     ; WAVE32: $vgpr0_vgpr1 = COPY [[MV]](s64)
     %0:_(<8 x s64>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
     %1:_(s32) = COPY $vgpr0
     %2:_(s32) = G_CONSTANT i32 1
     %3:_(s32) = G_ADD %1, %2
     %4:_(s64) = G_EXTRACT_VECTOR_ELT %0, %3
     $vgpr0_vgpr1 = COPY %4
 ...