[AMDGPU][GlobalISel] Improve BFI Pattern Matching

[Pierre-vh]

Summary

• Introduce a simple "Copy Hoisting" post RegBankAssign combine to prevent stray COP¨Y instructions from interfering with pattern matching.
• Prevent introduction of useless UNMERGE_VALUEs when splitting i64 values that also interfere with pattern matching.

[jsilvanus]

llvm/lib/Target/AMDGPU/AMDGPURegBankCombiner.cpp

401

This line seems to be wrong.

[Pierre-vh]

Tweak codegen

[Pierre-vh]

The copy hoisting works for BFI, but it's not an ideal solution I think because for other cases (like the insertelement tests) it can worsen codegen. I tried to make it as local as possible but I couldn't get rid of all the cases where it's unprofitable to move the copy "out".

If copy hoisting is an acceptable solution and we want to move forward with it maybe it needs to be made smarter, e.g. look at the whole expression tree, try to compute how many copies could be inserted and where and choose the solution that inserts the least copies? Perhaps it could even move copies downwards, e.g. currently it can transform a tree of expressions from SGPR to VGPR, but maybe it could do the opposite as well if it introduces less copies?

Thoughts? Should I keep going with this approach and try to make it smarter and better, or give it up?
Ideally I'd really like to be able to just fix the tablegen but I haven't found a way to do it properly.

There's also another annoying case in one of the BFI tests where the RegBankSelect adds 2 identical copies.
It prevents BFI from being selected because %6/%7 aren't identical (despite referencing the same physical register).
This could be fixed by another combine (?) or we could maybe change GISe's GIM_CheckIsSameOperand l so it looks through copies of physregs to vregs?

%6:vgpr(s32) = COPY %2:sgpr(s32)
%3:vgpr(s32) = G_XOR %1:vgpr, %6:vgpr
%4:vgpr(s32) = G_AND %0:vgpr, %3:vgpr
%7:vgpr(s32) = COPY %2:sgpr(s32)
%5:vgpr(s32) = G_XOR %7:vgpr, %4:vgpr

[foad]

My gut feeling is that this is the wrong approach, because it will convert SALU to VALU operations regardless of whether they can actually be matched by some larger VALU pattern. This is bad for all sorts of reasons especially on GFX10+: latency, occupancy, power consumption.

Instead I think the pattern matching for instructions like BFI needs to be able to look through cross-regbank copies. There has been some discussion in the past about this. Please read through https://discourse.llvm.org/t/globalisel-cross-bank-constant-propagation/57927 for a start.

[Pierre-vh]

In D132483#3745119, @foad wrote:

My gut feeling is that this is the wrong approach, because it will convert SALU to VALU operations regardless of whether they can actually be matched by some larger VALU pattern. This is bad for all sorts of reasons especially on GFX10+: latency, occupancy, power consumption.

Instead I think the pattern matching for instructions like BFI needs to be able to look through cross-regbank copies. There has been some discussion in the past about this. Please read through https://discourse.llvm.org/t/globalisel-cross-bank-constant-propagation/57927 for a start.

I agree, this approach is very hacky and I started noticing its limits/drawbacks while I was trying to fix the unintended consequences of the new combine.

Do you think it'd be worthwhile to restart a new discussion on the Discourse? It seems like there's a lot of different opinions on how to solve this.
I'm personally not a fan of adding something like g_maybe_cross_reg_bank_copy - it feels like we'll need it everywhere sooner or later and it's making TableGen even more confusing.
Ignoring copies all the time also seems wrong because it'll just move the problem somewhere else. Later someone might want to match copies explicitly and then we'll end up with the same discussion.

Recently I've also had a lot of issues like this, where the DAG pattern is straightforward but the MIR one has extra additions making it harder to match.
I feel like there's an opportunity to improve pattern matching for those cases. On top of my head, maybe we could add a way to create special complex "PatFrags" for GISel that:

• In DAGISel, Map to one or more nodes (like current PatFrags)
• In GISel, run a C++ function to perform the matching

Then the C++ function could check the instruction & look through copies as needed. It'd have an advantage over ComplexPatterns as it could be used as a non-leaf node and won't require having an equivalent DAGISel function.

[foad]

Do you think it'd be worthwhile to restart a new discussion on the Discourse?

Sure!

llvm/lib/Target/AMDGPU/AMDGPUCombine.td

def fp_minmax_to_clamp : GICombineRule<
   (apply [{ RegBankHelper.applyClamp(*${min_or_max}, ${matchinfo}); }])>;
 
 def fmed3_intrinsic_to_clamp : GICombineRule<
   (defs root:$fmed3, register_matchinfo:$matchinfo),
   (match (wip_match_opcode G_INTRINSIC):$fmed3,
          [{ return RegBankHelper.matchFPMed3ToClamp(*${fmed3}, ${matchinfo}); }]),
   (apply [{ RegBankHelper.applyClamp(*${fmed3}, ${matchinfo}); }])>;
 
+def simple_copy_hoisting_matchdata : GIDefMatchData<"MachineInstr *">;
+
+def simple_copy_hoisting : GICombineRule<
+  (defs root:$ffn, simple_copy_hoisting_matchdata:$matchinfo),
+  (match (wip_match_opcode COPY):$ffn,
+         [{ return RegBankHelper.matchSimpleCopyHoisting(*${ffn}, ${matchinfo}); }]),
+  (apply [{ RegBankHelper.applySimpleCopyHoisting(*${ffn}, ${matchinfo}); }])>;
+
 def remove_fcanonicalize_matchinfo : GIDefMatchData<"Register">;
 
 def remove_fcanonicalize : GICombineRule<
   (defs root:$fcanonicalize, remove_fcanonicalize_matchinfo:$matchinfo),
   (match (wip_match_opcode G_FCANONICALIZE):$fcanonicalize,
          [{ return PostLegalizerHelper.matchRemoveFcanonicalize(*${fcanonicalize}, ${matchinfo}); }]),
   (apply [{ Helper.replaceSingleDefInstWithReg(*${fcanonicalize}, ${matchinfo}); }])>;
 
def AMDGPUPostLegalizerCombinerHelper: GICombinerHelper<
   let DisableRuleOption = "amdgpupostlegalizercombiner-disable-rule";
   let StateClass = "AMDGPUPostLegalizerCombinerHelperState";
   let AdditionalArguments = [];
 }
 
 def AMDGPURegBankCombinerHelper : GICombinerHelper<
   "AMDGPUGenRegBankCombinerHelper",
   [zext_trunc_fold, int_minmax_to_med3, ptr_add_immed_chain,
-   fp_minmax_to_clamp, fp_minmax_to_med3, fmed3_intrinsic_to_clamp]> {
+   fp_minmax_to_clamp, fp_minmax_to_med3, fmed3_intrinsic_to_clamp,
+   simple_copy_hoisting]> {
   let DisableRuleOption = "amdgpuregbankcombiner-disable-rule";
   let StateClass = "AMDGPURegBankCombinerHelperState";
   let AdditionalArguments = [];
 }

llvm/lib/Target/AMDGPU/AMDGPURegBankCombiner.cpp

 public:
   AMDGPURegBankCombinerHelper(MachineIRBuilder &B, CombinerHelper &Helper)
       : B(B), MF(B.getMF()), MRI(*B.getMRI()),
         Subtarget(MF.getSubtarget<GCNSubtarget>()),
         RBI(*Subtarget.getRegBankInfo()), TRI(*Subtarget.getRegisterInfo()),
         TII(*Subtarget.getInstrInfo()), Helper(Helper){};
 
   bool isVgprRegBank(Register Reg);
+  bool isSgprRegBank(Register Reg);
   Register getAsVgpr(Register Reg);
 
+  bool isVgprToSgprCopy(const MachineInstr &MI);
+  bool isSgprToVgprCopy(const MachineInstr &MI);
+
   struct MinMaxMedOpc {
     unsigned Min, Max, Med;
   };
 
   struct Med3MatchInfo {
     unsigned Opc;
     Register Val0, Val1, Val2;
   };
 
   MinMaxMedOpc getMinMaxPair(unsigned Opc);
 
   template <class m_Cst, typename CstTy>
   bool matchMed(MachineInstr &MI, MachineRegisterInfo &MRI, MinMaxMedOpc MMMOpc,
                 Register &Val, CstTy &K0, CstTy &K1);
 
   bool matchIntMinMaxToMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo);
   bool matchFPMinMaxToMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo);
   bool matchFPMinMaxToClamp(MachineInstr &MI, Register &Reg);
   bool matchFPMed3ToClamp(MachineInstr &MI, Register &Reg);
   void applyMed3(MachineInstr &MI, Med3MatchInfo &MatchInfo);
   void applyClamp(MachineInstr &MI, Register &Reg);
 
+  bool matchSimpleCopyHoisting(MachineInstr &MI, MachineInstr *&MatchInfo);
+  void applySimpleCopyHoisting(MachineInstr &MI, MachineInstr *&MatchInfo);
+
 private:
   AMDGPU::SIModeRegisterDefaults getMode();
   bool getIEEE();
   bool getDX10Clamp();
   bool isFminnumIeee(const MachineInstr &MI);
   bool isFCst(MachineInstr *MI);
   bool isClampZeroToOne(MachineInstr *K0, MachineInstr *K1);
 };
 
 bool AMDGPURegBankCombinerHelper::isVgprRegBank(Register Reg) {
   return RBI.getRegBank(Reg, MRI, TRI)->getID() == AMDGPU::VGPRRegBankID;
 }
 
+bool AMDGPURegBankCombinerHelper::isSgprRegBank(Register Reg) {
+  return RBI.getRegBank(Reg, MRI, TRI)->getID() == AMDGPU::SGPRRegBankID;
+}
+
 Register AMDGPURegBankCombinerHelper::getAsVgpr(Register Reg) {
   if (isVgprRegBank(Reg))
     return Reg;
 
   // Search for existing copy of Reg to vgpr.
   for (MachineInstr &Use : MRI.use_instructions(Reg)) {
     Register Def = Use.getOperand(0).getReg();
-    if (Use.getOpcode() == AMDGPU::COPY && isVgprRegBank(Def))
+    if (Use.getOpcode() == AMDGPU::COPY && isVgprRegBank(Def)) {
+      // Make sure the use dominates the insertion point.
+      Use.moveBefore(&*B.getInsertPt());
       return Def;
     }
+  }
 
   // Copy Reg to vgpr.
   Register VgprReg = B.buildCopy(MRI.getType(Reg), Reg).getReg(0);
   MRI.setRegBank(VgprReg, RBI.getRegBank(AMDGPU::VGPRRegBankID));
   return VgprReg;
 }
 
+bool AMDGPURegBankCombinerHelper::isVgprToSgprCopy(const MachineInstr &MI) {
+  if (MI.getOpcode() != AMDGPU::COPY) {
+    return false;
+  }
+
+  Register DstReg = MI.getOperand(0).getReg();
+  Register SrcReg = MI.getOperand(1).getReg();
+  return isVgprRegBank(SrcReg) && isSgprRegBank(DstReg) &&
+         MRI.getType(DstReg) == MRI.getType(SrcReg);
+}
+
+bool AMDGPURegBankCombinerHelper::isSgprToVgprCopy(const MachineInstr &MI) {
+  if (MI.getOpcode() != AMDGPU::COPY) {
+    return false;
+  }
+
+  Register DstReg = MI.getOperand(0).getReg();
+  Register SrcReg = MI.getOperand(1).getReg();
+  return isSgprRegBank(SrcReg) && isVgprRegBank(DstReg) &&
+         MRI.getType(DstReg) == MRI.getType(SrcReg);
+}
+
 AMDGPURegBankCombinerHelper::MinMaxMedOpc
 AMDGPURegBankCombinerHelper::getMinMaxPair(unsigned Opc) {
   switch (Opc) {
   default:
     llvm_unreachable("Unsupported opcode");
   case AMDGPU::G_SMAX:
   case AMDGPU::G_SMIN:
     return {AMDGPU::G_SMIN, AMDGPU::G_SMAX, AMDGPU::G_AMDGPU_SMED3};
void AMDGPURegBankCombinerHelper::applyMed3(MachineInstr &MI,
   B.setInstrAndDebugLoc(MI);
   B.buildInstr(MatchInfo.Opc, {MI.getOperand(0)},
                {getAsVgpr(MatchInfo.Val0), getAsVgpr(MatchInfo.Val1),
                 getAsVgpr(MatchInfo.Val2)},
                MI.getFlags());
   MI.eraseFromParent();
 }
 
+static bool IsEligibleForSimpleCopyHoisting(MachineInstr &MI) {
+
+  // This combine is currently targeted at improving generation of BFI
+  // instructions, where copies can be inserted in-between a chain of
+  // bitwise operations, preventing pattern matching.
+  //
+  // More operations can be added if needed for other purposes, as long
+  // as the copy hoisting combine (which can transform an operation that
+  // uses/returns a SGPR to use VGPRs instead) won't break them.
+  //
+  // In short, those instructions should be able to run on both the SALU/VALU.
+  switch (MI.getOpcode()) {
+  case AMDGPU::G_XOR:
+  case AMDGPU::G_AND:
+  case AMDGPU::G_OR:
+    return true;
+  default:
+    return false;
+  }
+}
+
+bool AMDGPURegBankCombinerHelper::matchSimpleCopyHoisting(
+    MachineInstr &MI, MachineInstr *&MatchInfo) {
+  // When we have a SGPR -> VGPR copy where the input is defined by an eligible
+  // instruction, we may be able to hoist the copy to the operands of
+  // instruction that defines the COPY's input.
+  //
+  // For this combine to be applied, the following criterias must be satisfied:
+  //    - The COPY input must be eligible according to
+  //      `IsEligibleForSimpleCopyHoisting`.
+  //    - The COPY must be the only user of its input register.
+  //    - The COPY's output register must only be used by another eligible
+  //      instruction.
+  //    - There are also some additional constraints on the COPY input
+  //    instruction's
+  //      operands. For instance, some specific instructions are not allowed to
+  //      try and avoid cases where hoisting the copy would worsen codegen.
+  //
+  // For example, in:
+  //
+  //   %4:sgpr(s32) = G_XOR %1:sgpr, %2:sgpr
+  //   %7:vgpr(s32) = COPY %4:sgpr(s32)
+  //   %5:vgpr(s32) = G_AND %0:vgpr, %7:vgpr
+  //
+  // %4 is only used by the COPY, G_XOR has only SGPR operands and
+  // G_AND has only VGPR operand. We can hoist the copy out into the G_XOR
+  // operands to make both instructions use all VGPR operands, making
+  // matching easier in GISel.
+  //
+  //   %4:vgpr(s32) = COPY %1:sgpr
+  //   %5:vgpr(s32) = COPY %2:sgpr
+  //   %6:vgpr(s32) = G_XOR %4:vgpr, %5:vgpr
+  //   %7:vgpr(s32) = G_AND %0:vgpr, %6:vgpr
+  //
+  // Now G_XOR and G_AND chain neatly together, making pattern matching
+  // easier.
+
+  Register DstReg = MI.getOperand(0).getReg();
+  Register InputReg = MI.getOperand(1).getReg();
+
+  // Check that this is a simple SGPR -> VGPR copy.
+  if (!isSgprToVgprCopy(MI)) {
+    return false;
+  }
+
+  // Input/Output register must have exactly one user.
+  if (!MRI.hasOneNonDBGUse(InputReg) || !MRI.hasOneNonDBGUse(DstReg)) {
+    return false;
+  }
+
+  // The instruction that defines the input register AND the instruction
+  // that uses the output register must both be considered eligible.
+  MachineInstr &InputInstr = *getDefIgnoringCopies(InputReg, MRI);
+  MachineInstr &DefUserInstr = *MRI.use_instr_begin(DstReg);
+  if (!IsEligibleForSimpleCopyHoisting(InputInstr) ||
+      !IsEligibleForSimpleCopyHoisting(DefUserInstr)) {
+    return false;
+  }
+
+  // Check all instructions are in the same basic block.
+  if (InputInstr.getParent() != MI.getParent() ||
+      MI.getParent() != DefUserInstr.getParent()) {
+    return false;
+  }
+
+  // Check all input operands of InputInstr are SGPRs, and check
+  // for undesirable patterns.
+  for (std::size_t k = 1; k < InputInstr.getNumOperands(); ++k) {
+    MachineOperand &MO = InputInstr.getOperand(k);
+    Register Reg = MO.getReg();
+    if (!isSgprRegBank(Reg)) {
+      return false;
+    }
+
+    // G_CONSTANT is not allowed because it can prevent matching of
+    // common instructions, like s_not.
+    //
+    // G_BITCAST is not allowed because we assume that it's there for a
+    // reason, and forcing its result to be a VGPR when the destination
+    // of the BITCAST is a SGPR worsens codegen.
+    //
+    // VGPR->SGPR G_COPY is not allowed for similar reasons (likely there
+    // for a reason + affects codegen), but also because it'd introduce
+    // a pair of useless copies.
+    MachineInstr *Def = MRI.getVRegDef(Reg);
+    if (Def->getOpcode() == AMDGPU::G_CONSTANT ||
+        Def->getOpcode() == AMDGPU::G_BITCAST || isVgprToSgprCopy(*Def)) {
+      return false;
+    }
+  }
+
+  MatchInfo = &InputInstr;
+  return true;
+}
+
+void AMDGPURegBankCombinerHelper::applySimpleCopyHoisting(
+    MachineInstr &MI, MachineInstr *&MatchInfo) {
+  assert(&MI != MatchInfo);
+
+  B.setInstrAndDebugLoc(*MatchInfo);
+  const RegisterBank &VGPRRegBank = RBI.getRegBank(AMDGPU::VGPRRegBankID);
+
+  // Make all operands of MatchInfo into VGPRs.
+  for (std::size_t k = 1; k < MatchInfo->getNumOperands(); ++k) {
+    MachineOperand &MO = MatchInfo->getOperand(k);
+    Register Reg = MO.getReg();
+    assert(isSgprRegBank(Reg));
+    MO.setReg(getAsVgpr(Reg));
+  }
+
+  // Make the result of MatchInfo into a VGPR.
+  Register MatchInfoDst = MatchInfo->getOperand(0).getReg();
+  MRI.setRegBank(MatchInfoDst, VGPRRegBank);
+
+  // Replace all users of the Copy's result with MatchInfoDst.
+  MRI.replaceRegWith(MI.getOperand(0).getReg(), MatchInfoDst);
+
+  // Eliminate the copy.
+  MI.removeFromParent();
+}
+
 AMDGPU::SIModeRegisterDefaults AMDGPURegBankCombinerHelper::getMode() {
   return MF.getInfo<SIMachineFunctionInfo>()->getMode();
 }
 
 bool AMDGPURegBankCombinerHelper::getIEEE() { return getMode().IEEE; }
 
 bool AMDGPURegBankCombinerHelper::getDX10Clamp() { return getMode().DX10Clamp; }
 

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

 
 void AMDGPURegisterBankInfo::split64BitValueForMapping(
   MachineIRBuilder &B,
   SmallVector<Register, 2> &Regs,
   LLT HalfTy,
   Register Reg) const {
   assert(HalfTy.getSizeInBits() == 32);
   MachineRegisterInfo *MRI = B.getMRI();
+  const RegisterBank *Bank = getRegBank(Reg, *MRI, *TRI);
+
+  // Check if the Reg is already defined by a MERGE_VALUE, if so
+  // don't bother generating a UNMERGE_VALUE/BUILD_VECTOR and just
+  // reuse its operands.
+  MachineInstr *RegInst = getDefIgnoringCopies(Reg, *MRI);
+  if ((RegInst->getOpcode() == AMDGPU::G_MERGE_VALUES ||
+       RegInst->getOpcode() == AMDGPU::G_BUILD_VECTOR) &&
+      RegInst->getNumOperands() == 3) {
+    Register Lo = RegInst->getOperand(1).getReg();
+    Register Hi = RegInst->getOperand(2).getReg();
+
+    // FIXME: Do we need to insert copies to fix types?
+    assert(MRI->getType(Lo) == HalfTy && MRI->getType(Hi) == HalfTy);
+
+    if (MRI->getRegBankOrNull(Lo) == Bank &&
+        MRI->getRegBankOrNull(Hi) == Bank) {
+      Regs.push_back(Lo);
+      Regs.push_back(Hi);
+      return;
+    }
+  }
+
   Register LoLHS = MRI->createGenericVirtualRegister(HalfTy);
   Register HiLHS = MRI->createGenericVirtualRegister(HalfTy);
-  const RegisterBank *Bank = getRegBank(Reg, *MRI, *TRI);
   MRI->setRegBank(LoLHS, *Bank);
   MRI->setRegBank(HiLHS, *Bank);
 
   Regs.push_back(LoLHS);
   Regs.push_back(HiLHS);
 
   B.buildInstr(AMDGPU::G_UNMERGE_VALUES)
     .addDef(LoLHS)

llvm/lib/Target/AMDGPU/SIInstructions.td

(V_BFI_B32_e64 (i32 (EXTRACT_SUBREG VReg_64:$x, sub1)),
                (i32 (EXTRACT_SUBREG VReg_64:$y, sub1)),
                (i32 (EXTRACT_SUBREG VReg_64:$z, sub1))), sub1)
 >;
 
 // SHA-256 Ch function
 // z ^ (x & (y ^ z))
 def : AMDGPUPat <
   (DivergentBinFrag<xor> i32:$z, (and i32:$x, (xor i32:$y, i32:$z))),
-  (V_BFI_B32_e64 VSrc_b32:$x, VSrc_b32:$y, VSrc_b32:$z)
+  (V_BFI_B32_e64 (COPY_TO_REGCLASS VSrc_b32:$x, VGPR_32),
+                 (COPY_TO_REGCLASS VSrc_b32:$y, VGPR_32),
+                 (COPY_TO_REGCLASS VSrc_b32:$z, VGPR_32))
 >;
 
 // 64-bit version
 def : AMDGPUPat <
   (DivergentBinFrag<xor> i64:$z, (and i64:$x, (xor i64:$y, i64:$z))),
   (REG_SEQUENCE VReg_64,
     (V_BFI_B32_e64 (i32 (EXTRACT_SUBREG VReg_64:$x, sub0)),
                (i32 (EXTRACT_SUBREG VReg_64:$y, sub0)),

 >;
 
 // SHA-256 Ma patterns
 
 // ((x & z) | (y & (x | z))) -> BFI (XOR x, y), z, y
 def : AMDGPUPat <
   (DivergentBinFrag<or> (and i32:$x, i32:$z),
                         (and i32:$y, (or i32:$x, i32:$z))),
-  (V_BFI_B32_e64 (V_XOR_B32_e64 VSrc_b32:$x, VSrc_b32:$y), VSrc_b32:$z, VSrc_b32:$y)
+  (V_BFI_B32_e64 (V_XOR_B32_e64 (COPY_TO_REGCLASS VSrc_b32:$x, VGPR_32),
+                                (COPY_TO_REGCLASS VSrc_b32:$y, VGPR_32)), 
+                  (COPY_TO_REGCLASS VSrc_b32:$z, VGPR_32), 
+                  (COPY_TO_REGCLASS VSrc_b32:$y, VGPR_32))
 >;
 
 def : AMDGPUPat <
   (DivergentBinFrag<or> (and i64:$x, i64:$z),
                         (and i64:$y, (or i64:$x, i64:$z))),
   (REG_SEQUENCE VReg_64,
     (V_BFI_B32_e64 (V_XOR_B32_e64 (i32 (EXTRACT_SUBREG VReg_64:$x, sub0)),
                     (i32 (EXTRACT_SUBREG VReg_64:$y, sub0))),

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

 ; GFX8-NEXT:    v_or_b32_e32 v1, v2, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshl_i64_48:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX9-NEXT:    v_lshrrev_b64 v[0:1], 16, v[2:3]
-; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v4
-; GFX9-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_fshl_i64_48:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX10-NEXT:    v_lshrrev_b64 v[0:1], 16, v[2:3]
-; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 16, v4
-; GFX10-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX10-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_fshl_i64_48:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX11-NEXT:    v_lshrrev_b64 v[0:1], 16, v[2:3]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 16, v4
-; GFX11-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %result = call i64 @llvm.fshl.i64(i64 %lhs, i64 %rhs, i64 48)
   ret i64 %result
 }
 
 define amdgpu_ps <2 x float> @v_fshl_i64_ssv(i64 inreg %lhs, i64 inreg %rhs, i64 %amt) {
 ; GFX6-LABEL: v_fshl_i64_ssv:
 ; GFX6:       ; %bb.0:

 ; GFX9-NEXT:    v_lshlrev_b64 v[12:13], v14, v[0:1]
 ; GFX9-NEXT:    v_lshlrev_b64 v[0:1], v16, v[0:1]
 ; GFX9-NEXT:    v_or_b32_e32 v8, v8, v10
 ; GFX9-NEXT:    v_or_b32_e32 v9, v9, v11
 ; GFX9-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v14
 ; GFX9-NEXT:    v_cndmask_b32_e32 v10, 0, v12, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v11, 0, v13, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v8, v1, v9, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v14
 ; GFX9-NEXT:    v_cndmask_b32_e32 v12, v0, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v13, v1, v3, vcc
 ; GFX9-NEXT:    v_lshrrev_b64 v[0:1], 1, v[4:5]
-; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 31, v6
-; GFX9-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v13, v8, v3, vcc
 ; GFX9-NEXT:    v_lshrrev_b64 v[2:3], 1, v[6:7]
+; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 31, v1
 ; GFX9-NEXT:    v_sub_u32_e32 v6, 64, v15
 ; GFX9-NEXT:    v_subrev_u32_e32 v14, 64, v15
 ; GFX9-NEXT:    v_lshrrev_b64 v[4:5], v15, v[0:1]
 ; GFX9-NEXT:    v_lshlrev_b64 v[6:7], v6, v[2:3]
 ; GFX9-NEXT:    v_lshrrev_b64 v[8:9], v15, v[2:3]
 ; GFX9-NEXT:    v_lshrrev_b64 v[2:3], v14, v[2:3]
 ; GFX9-NEXT:    v_or_b32_e32 v4, v4, v6
 ; GFX9-NEXT:    v_or_b32_e32 v5, v5, v7

 ;
 ; GFX10-LABEL: v_fshl_i128:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v18, 0x7f, v8
 ; GFX10-NEXT:    v_xor_b32_e32 v8, -1, v8
 ; GFX10-NEXT:    v_lshrrev_b64 v[4:5], 1, v[4:5]
-; GFX10-NEXT:    v_lshlrev_b32_e32 v12, 31, v6
-; GFX10-NEXT:    v_lshrrev_b64 v[6:7], 1, v[6:7]
-; GFX10-NEXT:    v_sub_nc_u32_e32 v9, 64, v18
+; GFX10-NEXT:    v_lshrrev_b64 v[12:13], 1, v[6:7]
+; GFX10-NEXT:    v_sub_nc_u32_e32 v10, 64, v18
 ; GFX10-NEXT:    v_and_b32_e32 v19, 0x7f, v8
-; GFX10-NEXT:    v_lshlrev_b64 v[10:11], v18, v[2:3]
-; GFX10-NEXT:    v_or_b32_e32 v5, v5, v12
+; GFX10-NEXT:    v_lshlrev_b64 v[8:9], v18, v[2:3]
+; GFX10-NEXT:    v_lshl_or_b32 v5, v6, 31, v5
 ; GFX10-NEXT:    v_subrev_nc_u32_e32 v20, 64, v18
-; GFX10-NEXT:    v_lshrrev_b64 v[8:9], v9, v[0:1]
+; GFX10-NEXT:    v_lshrrev_b64 v[10:11], v10, v[0:1]
 ; GFX10-NEXT:    v_sub_nc_u32_e32 v16, 64, v19
-; GFX10-NEXT:    v_lshlrev_b64 v[12:13], v18, v[0:1]
+; GFX10-NEXT:    v_lshlrev_b64 v[6:7], v18, v[0:1]
 ; GFX10-NEXT:    v_lshrrev_b64 v[14:15], v19, v[4:5]
 ; GFX10-NEXT:    v_lshlrev_b64 v[0:1], v20, v[0:1]
 ; GFX10-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v18
-; GFX10-NEXT:    v_or_b32_e32 v10, v8, v10
+; GFX10-NEXT:    v_or_b32_e32 v10, v10, v8
 ; GFX10-NEXT:    v_subrev_nc_u32_e32 v8, 64, v19
-; GFX10-NEXT:    v_lshlrev_b64 v[16:17], v16, v[6:7]
-; GFX10-NEXT:    v_or_b32_e32 v11, v9, v11
+; GFX10-NEXT:    v_lshlrev_b64 v[16:17], v16, v[12:13]
+; GFX10-NEXT:    v_or_b32_e32 v11, v11, v9
 ; GFX10-NEXT:    v_cmp_gt_u32_e64 s4, 64, v19
 ; GFX10-NEXT:    v_cndmask_b32_e32 v10, v0, v10, vcc_lo
-; GFX10-NEXT:    v_lshrrev_b64 v[8:9], v8, v[6:7]
+; GFX10-NEXT:    v_lshrrev_b64 v[8:9], v8, v[12:13]
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 s5, 0, v19
 ; GFX10-NEXT:    v_or_b32_e32 v14, v14, v16
 ; GFX10-NEXT:    v_or_b32_e32 v15, v15, v17
 ; GFX10-NEXT:    v_cndmask_b32_e32 v11, v1, v11, vcc_lo
-; GFX10-NEXT:    v_lshrrev_b64 v[0:1], v19, v[6:7]
+; GFX10-NEXT:    v_lshrrev_b64 v[0:1], v19, v[12:13]
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 s6, 0, v18
 ; GFX10-NEXT:    v_cndmask_b32_e64 v8, v8, v14, s4
-; GFX10-NEXT:    v_cndmask_b32_e64 v6, v9, v15, s4
-; GFX10-NEXT:    v_cndmask_b32_e32 v12, 0, v12, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e32 v7, 0, v13, vcc_lo
+; GFX10-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s4
+; GFX10-NEXT:    v_cndmask_b32_e32 v6, 0, v6, vcc_lo
+; GFX10-NEXT:    v_cndmask_b32_e32 v7, 0, v7, vcc_lo
 ; GFX10-NEXT:    v_cndmask_b32_e64 v2, v10, v2, s6
 ; GFX10-NEXT:    v_cndmask_b32_e64 v3, v11, v3, s6
 ; GFX10-NEXT:    v_cndmask_b32_e64 v4, v8, v4, s5
-; GFX10-NEXT:    v_cndmask_b32_e64 v5, v6, v5, s5
-; GFX10-NEXT:    v_cndmask_b32_e64 v6, 0, v0, s4
-; GFX10-NEXT:    v_cndmask_b32_e64 v8, 0, v1, s4
-; GFX10-NEXT:    v_or_b32_e32 v0, v12, v4
+; GFX10-NEXT:    v_cndmask_b32_e64 v5, v9, v5, s5
+; GFX10-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s4
+; GFX10-NEXT:    v_cndmask_b32_e64 v9, 0, v1, s4
+; GFX10-NEXT:    v_or_b32_e32 v0, v6, v4
 ; GFX10-NEXT:    v_or_b32_e32 v1, v7, v5
-; GFX10-NEXT:    v_or_b32_e32 v2, v2, v6
-; GFX10-NEXT:    v_or_b32_e32 v3, v3, v8
+; GFX10-NEXT:    v_or_b32_e32 v2, v2, v8
+; GFX10-NEXT:    v_or_b32_e32 v3, v3, v9
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_fshl_i128:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    v_lshrrev_b64 v[4:5], 1, v[4:5]
 ; GFX11-NEXT:    v_and_b32_e32 v18, 0x7f, v8
 ; GFX11-NEXT:    v_xor_b32_e32 v8, -1, v8
-; GFX11-NEXT:    v_lshrrev_b64 v[4:5], 1, v[4:5]
+; GFX11-NEXT:    v_lshrrev_b64 v[12:13], 1, v[6:7]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_sub_nc_u32_e32 v9, 64, v18
-; GFX11-NEXT:    v_lshlrev_b64 v[10:11], v18, v[2:3]
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v20, 64, v18
+; GFX11-NEXT:    v_sub_nc_u32_e32 v10, 64, v18
+; GFX11-NEXT:    v_lshl_or_b32 v5, v6, 31, v5
+; GFX11-NEXT:    v_lshlrev_b64 v[6:7], v18, v[0:1]
 ; GFX11-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v18
-; GFX11-NEXT:    v_lshlrev_b32_e32 v12, 31, v6
 ; GFX11-NEXT:    v_and_b32_e32 v19, 0x7f, v8
-; GFX11-NEXT:    v_lshrrev_b64 v[8:9], v9, v[0:1]
-; GFX11-NEXT:    v_lshrrev_b64 v[6:7], 1, v[6:7]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_or_b32_e32 v5, v5, v12
-; GFX11-NEXT:    v_lshlrev_b64 v[12:13], v18, v[0:1]
-; GFX11-NEXT:    v_lshlrev_b64 v[0:1], v20, v[0:1]
-; GFX11-NEXT:    v_or_b32_e32 v10, v8, v10
-; GFX11-NEXT:    v_or_b32_e32 v11, v9, v11
-; GFX11-NEXT:    v_cndmask_b32_e32 v12, 0, v12, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cndmask_b32_e32 v10, v0, v10, vcc_lo
+; GFX11-NEXT:    v_lshlrev_b64 v[8:9], v18, v[2:3]
+; GFX11-NEXT:    v_lshrrev_b64 v[10:11], v10, v[0:1]
+; GFX11-NEXT:    v_subrev_nc_u32_e32 v20, 64, v18
+; GFX11-NEXT:    v_cndmask_b32_e32 v6, 0, v6, vcc_lo
 ; GFX11-NEXT:    v_sub_nc_u32_e32 v16, 64, v19
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v8, 64, v19
 ; GFX11-NEXT:    v_lshrrev_b64 v[14:15], v19, v[4:5]
 ; GFX11-NEXT:    v_cmp_gt_u32_e64 s0, 64, v19
-; GFX11-NEXT:    v_cndmask_b32_e32 v11, v1, v11, vcc_lo
-; GFX11-NEXT:    v_lshlrev_b64 v[16:17], v16, v[6:7]
-; GFX11-NEXT:    v_lshrrev_b64 v[8:9], v8, v[6:7]
-; GFX11-NEXT:    v_lshrrev_b64 v[0:1], v19, v[6:7]
+; GFX11-NEXT:    v_or_b32_e32 v10, v10, v8
+; GFX11-NEXT:    v_subrev_nc_u32_e32 v8, 64, v19
+; GFX11-NEXT:    v_lshlrev_b64 v[16:17], v16, v[12:13]
+; GFX11-NEXT:    v_lshlrev_b64 v[0:1], v20, v[0:1]
+; GFX11-NEXT:    v_or_b32_e32 v11, v11, v9
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 s1, 0, v19
-; GFX11-NEXT:    v_cmp_eq_u32_e64 s2, 0, v18
-; GFX11-NEXT:    v_cndmask_b32_e32 v7, 0, v13, vcc_lo
+; GFX11-NEXT:    v_lshrrev_b64 v[8:9], v8, v[12:13]
+; GFX11-NEXT:    v_cndmask_b32_e32 v7, 0, v7, vcc_lo
 ; GFX11-NEXT:    v_or_b32_e32 v14, v14, v16
 ; GFX11-NEXT:    v_or_b32_e32 v15, v15, v17
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_cndmask_b32_e64 v2, v10, v2, s2
-; GFX11-NEXT:    v_cndmask_b32_e64 v3, v11, v3, s2
+; GFX11-NEXT:    v_dual_cndmask_b32 v10, v0, v10 :: v_dual_cndmask_b32 v11, v1, v11
+; GFX11-NEXT:    v_lshrrev_b64 v[0:1], v19, v[12:13]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_2) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_cndmask_b32_e64 v8, v8, v14, s0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_cndmask_b32_e64 v6, v9, v15, s0
+; GFX11-NEXT:    v_cmp_eq_u32_e64 s2, 0, v18
+; GFX11-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s0
 ; GFX11-NEXT:    v_cndmask_b32_e64 v4, v8, v4, s1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_cndmask_b32_e64 v5, v6, v5, s1
-; GFX11-NEXT:    v_cndmask_b32_e64 v6, 0, v0, s0
-; GFX11-NEXT:    v_cndmask_b32_e64 v8, 0, v1, s0
-; GFX11-NEXT:    v_or_b32_e32 v0, v12, v4
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
+; GFX11-NEXT:    v_cndmask_b32_e64 v2, v10, v2, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v3, v11, v3, s2
+; GFX11-NEXT:    v_cndmask_b32_e64 v5, v9, v5, s1
+; GFX11-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s0
+; GFX11-NEXT:    v_cndmask_b32_e64 v9, 0, v1, s0
+; GFX11-NEXT:    v_or_b32_e32 v0, v6, v4
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_or_b32_e32 v1, v7, v5
-; GFX11-NEXT:    v_or_b32_e32 v2, v2, v6
+; GFX11-NEXT:    v_or_b32_e32 v2, v2, v8
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
-; GFX11-NEXT:    v_or_b32_e32 v3, v3, v8
+; GFX11-NEXT:    v_or_b32_e32 v3, v3, v9
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %result = call i128 @llvm.fshl.i128(i128 %lhs, i128 %rhs, i128 %amt)
   ret i128 %result
 }
 
 define amdgpu_ps <4 x float> @v_fshl_i128_ssv(i128 inreg %lhs, i128 inreg %rhs, i128 %amt) {
 ; GFX6-LABEL: v_fshl_i128_ssv:
 ; GFX6:       ; %bb.0:

 ; GFX9-NEXT:    s_cmp_eq_u32 s8, 0
 ; GFX9-NEXT:    s_cselect_b32 s13, 1, 0
 ; GFX9-NEXT:    s_lshl_b64 s[6:7], s[0:1], s8
 ; GFX9-NEXT:    s_lshr_b64 s[10:11], s[0:1], s9
 ; GFX9-NEXT:    s_lshl_b64 s[8:9], s[2:3], s8
 ; GFX9-NEXT:    s_or_b64 s[8:9], s[10:11], s[8:9]
 ; GFX9-NEXT:    s_lshl_b64 s[0:1], s[0:1], s5
 ; GFX9-NEXT:    s_cmp_lg_u32 s12, 0
+; GFX9-NEXT:    v_lshrrev_b64 v[0:1], 1, v[0:1]
 ; GFX9-NEXT:    s_cselect_b64 s[6:7], s[6:7], 0
 ; GFX9-NEXT:    s_cselect_b64 s[0:1], s[8:9], s[0:1]
 ; GFX9-NEXT:    s_cmp_lg_u32 s13, 0
-; GFX9-NEXT:    v_lshrrev_b64 v[0:1], 1, v[0:1]
 ; GFX9-NEXT:    s_cselect_b64 s[0:1], s[2:3], s[0:1]
-; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 31, v2
+; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 31, v1
 ; GFX9-NEXT:    v_lshrrev_b64 v[2:3], 1, v[2:3]
 ; GFX9-NEXT:    s_sub_i32 s2, s4, 64
 ; GFX9-NEXT:    s_sub_i32 s3, 64, s4
-; GFX9-NEXT:    v_or_b32_e32 v1, v1, v4
 ; GFX9-NEXT:    s_cmp_lt_u32 s4, 64
 ; GFX9-NEXT:    s_cselect_b32 s5, 1, 0
 ; GFX9-NEXT:    s_cmp_eq_u32 s4, 0
 ; GFX9-NEXT:    v_lshrrev_b64 v[4:5], s4, v[0:1]
 ; GFX9-NEXT:    v_lshlrev_b64 v[6:7], s3, v[2:3]
 ; GFX9-NEXT:    s_cselect_b32 s8, 1, 0
 ; GFX9-NEXT:    v_lshrrev_b64 v[8:9], s4, v[2:3]
 ; GFX9-NEXT:    v_lshrrev_b64 v[2:3], s2, v[2:3]

 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_mov_b64 s[6:7], 0x7f
 ; GFX10-NEXT:    v_lshrrev_b64 v[0:1], 1, v[0:1]
 ; GFX10-NEXT:    s_and_b64 s[8:9], s[4:5], s[6:7]
 ; GFX10-NEXT:    s_andn2_b64 s[4:5], s[6:7], s[4:5]
 ; GFX10-NEXT:    s_sub_i32 s5, s8, 64
 ; GFX10-NEXT:    s_sub_i32 s6, 64, s8
 ; GFX10-NEXT:    s_cmp_lt_u32 s8, 64
-; GFX10-NEXT:    v_lshlrev_b32_e32 v4, 31, v2
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 31, v1
 ; GFX10-NEXT:    s_cselect_b32 s12, 1, 0
 ; GFX10-NEXT:    s_cmp_eq_u32 s8, 0
 ; GFX10-NEXT:    v_lshrrev_b64 v[2:3], 1, v[2:3]
 ; GFX10-NEXT:    s_cselect_b32 s13, 1, 0
 ; GFX10-NEXT:    s_lshr_b64 s[6:7], s[0:1], s6
 ; GFX10-NEXT:    s_lshl_b64 s[10:11], s[2:3], s8
 ; GFX10-NEXT:    s_lshl_b64 s[8:9], s[0:1], s8
 ; GFX10-NEXT:    s_or_b64 s[6:7], s[6:7], s[10:11]
 ; GFX10-NEXT:    s_lshl_b64 s[0:1], s[0:1], s5
 ; GFX10-NEXT:    s_cmp_lg_u32 s12, 0
-; GFX10-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX10-NEXT:    v_lshrrev_b64 v[4:5], s4, v[0:1]
 ; GFX10-NEXT:    s_cselect_b64 s[8:9], s[8:9], 0
 ; GFX10-NEXT:    s_cselect_b64 s[0:1], s[6:7], s[0:1]
 ; GFX10-NEXT:    s_cmp_lg_u32 s13, 0
 ; GFX10-NEXT:    s_cselect_b64 s[2:3], s[2:3], s[0:1]
 ; GFX10-NEXT:    s_sub_i32 s0, 64, s4
-; GFX10-NEXT:    v_lshrrev_b64 v[4:5], s4, v[0:1]
 ; GFX10-NEXT:    v_lshlrev_b64 v[6:7], s0, v[2:3]
 ; GFX10-NEXT:    s_sub_i32 s0, s4, 64
 ; GFX10-NEXT:    s_cmp_lt_u32 s4, 64
 ; GFX10-NEXT:    v_lshrrev_b64 v[8:9], s0, v[2:3]
 ; GFX10-NEXT:    s_cselect_b32 s1, 1, 0
 ; GFX10-NEXT:    s_cmp_eq_u32 s4, 0
 ; GFX10-NEXT:    v_or_b32_e32 v4, v4, v6
 ; GFX10-NEXT:    s_cselect_b32 s5, 1, 0

 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_mov_b64 s[6:7], 0x7f
 ; GFX11-NEXT:    v_lshrrev_b64 v[0:1], 1, v[0:1]
 ; GFX11-NEXT:    s_and_b64 s[8:9], s[4:5], s[6:7]
 ; GFX11-NEXT:    s_and_not1_b64 s[4:5], s[6:7], s[4:5]
 ; GFX11-NEXT:    s_sub_i32 s5, s8, 64
 ; GFX11-NEXT:    s_sub_i32 s6, 64, s8
 ; GFX11-NEXT:    s_cmp_lt_u32 s8, 64
-; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 31, v2
+; GFX11-NEXT:    v_lshl_or_b32 v1, v2, 31, v1
 ; GFX11-NEXT:    s_cselect_b32 s12, 1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s8, 0
 ; GFX11-NEXT:    v_lshrrev_b64 v[2:3], 1, v[2:3]
 ; GFX11-NEXT:    s_cselect_b32 s13, 1, 0
 ; GFX11-NEXT:    s_lshr_b64 s[6:7], s[0:1], s6
 ; GFX11-NEXT:    s_lshl_b64 s[10:11], s[2:3], s8
 ; GFX11-NEXT:    s_lshl_b64 s[8:9], s[0:1], s8
 ; GFX11-NEXT:    s_or_b64 s[6:7], s[6:7], s[10:11]
 ; GFX11-NEXT:    s_lshl_b64 s[0:1], s[0:1], s5
 ; GFX11-NEXT:    s_cmp_lg_u32 s12, 0
-; GFX11-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX11-NEXT:    v_lshrrev_b64 v[4:5], s4, v[0:1]
 ; GFX11-NEXT:    s_cselect_b64 s[8:9], s[8:9], 0
 ; GFX11-NEXT:    s_cselect_b64 s[0:1], s[6:7], s[0:1]
 ; GFX11-NEXT:    s_cmp_lg_u32 s13, 0
 ; GFX11-NEXT:    s_cselect_b64 s[2:3], s[2:3], s[0:1]
 ; GFX11-NEXT:    s_sub_i32 s0, 64, s4
-; GFX11-NEXT:    v_lshrrev_b64 v[4:5], s4, v[0:1]
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    v_lshlrev_b64 v[6:7], s0, v[2:3]
 ; GFX11-NEXT:    s_sub_i32 s0, s4, 64
 ; GFX11-NEXT:    s_cmp_lt_u32 s4, 64
 ; GFX11-NEXT:    v_lshrrev_b64 v[8:9], s0, v[2:3]
 ; GFX11-NEXT:    s_cselect_b32 s1, 1, 0
 ; GFX11-NEXT:    s_cmp_eq_u32 s4, 0
 ; GFX11-NEXT:    v_or_b32_e32 v4, v4, v6
 ; GFX11-NEXT:    s_cselect_b32 s5, 1, 0
 ; GFX11-NEXT:    s_and_b32 s0, 1, s1
 ; GFX11-NEXT:    v_or_b32_e32 v5, v5, v7
 ; GFX11-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX11-NEXT:    s_and_b32 s0, 1, s5
 ; GFX11-NEXT:    s_and_b32 s1, 1, s1
 ; GFX11-NEXT:    v_lshrrev_b64 v[2:3], s4, v[2:3]
-; GFX11-NEXT:    v_dual_cndmask_b32 v5, v9, v5 :: v_dual_cndmask_b32 v4, v8, v4
+; GFX11-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v5, v9, v5
 ; GFX11-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX11-NEXT:    v_cmp_ne_u32_e64 s0, 0, s1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_dual_cndmask_b32 v1, v5, v1 :: v_dual_cndmask_b32 v0, v4, v0
+; GFX11-NEXT:    v_dual_cndmask_b32 v0, v4, v0 :: v_dual_cndmask_b32 v1, v5, v1
 ; GFX11-NEXT:    v_cndmask_b32_e64 v2, 0, v2, s0
 ; GFX11-NEXT:    v_cndmask_b32_e64 v3, 0, v3, s0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_or_b32_e32 v1, s9, v1
 ; GFX11-NEXT:    v_or_b32_e32 v0, s8, v0
+; GFX11-NEXT:    v_or_b32_e32 v1, s9, v1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_or_b32_e32 v2, s2, v2
 ; GFX11-NEXT:    v_or_b32_e32 v3, s3, v3
 ; GFX11-NEXT:    ; return to shader part epilog
   %result = call i128 @llvm.fshl.i128(i128 %lhs, i128 %rhs, i128 %amt)
   %cast.result = bitcast i128 %result to <4 x float>
   ret <4 x float> %cast.result
 }

 ; GFX8-NEXT:    v_or_b32_e32 v6, v16, v6
 ; GFX8-NEXT:    v_or_b32_e32 v7, v20, v7
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshl_v2i128:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_and_b32_e32 v23, 0x7f, v16
+; GFX9-NEXT:    v_lshrrev_b64 v[8:9], 1, v[8:9]
 ; GFX9-NEXT:    v_sub_u32_e32 v17, 64, v23
+; GFX9-NEXT:    v_xor_b32_e32 v16, -1, v16
 ; GFX9-NEXT:    v_lshrrev_b64 v[17:18], v17, v[0:1]
 ; GFX9-NEXT:    v_lshlrev_b64 v[21:22], v23, v[2:3]
-; GFX9-NEXT:    v_lshrrev_b64 v[8:9], 1, v[8:9]
-; GFX9-NEXT:    v_xor_b32_e32 v16, -1, v16
-; GFX9-NEXT:    v_or_b32_e32 v21, v17, v21
-; GFX9-NEXT:    v_lshlrev_b32_e32 v17, 31, v10
 ; GFX9-NEXT:    v_and_b32_e32 v24, 0x7f, v16
+; GFX9-NEXT:    v_lshl_or_b32 v9, v10, 31, v9
 ; GFX9-NEXT:    v_lshrrev_b64 v[10:11], 1, v[10:11]
-; GFX9-NEXT:    v_or_b32_e32 v9, v9, v17
 ; GFX9-NEXT:    v_sub_u32_e32 v16, 64, v24
+; GFX9-NEXT:    v_or_b32_e32 v21, v17, v21
 ; GFX9-NEXT:    v_or_b32_e32 v22, v18, v22
 ; GFX9-NEXT:    v_lshlrev_b64 v[16:17], v16, v[10:11]
 ; GFX9-NEXT:    v_lshrrev_b64 v[18:19], v24, v[8:9]
 ; GFX9-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v23
 ; GFX9-NEXT:    v_or_b32_e32 v18, v18, v16
 ; GFX9-NEXT:    v_subrev_u32_e32 v16, 64, v23
 ; GFX9-NEXT:    v_or_b32_e32 v19, v19, v17
 ; GFX9-NEXT:    v_lshlrev_b64 v[16:17], v16, v[0:1]

 ; GFX9-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GFX9-NEXT:    v_or_b32_e32 v11, v9, v11
 ; GFX9-NEXT:    v_lshlrev_b64 v[8:9], v16, v[4:5]
 ; GFX9-NEXT:    v_lshlrev_b64 v[4:5], v18, v[4:5]
 ; GFX9-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v16
 ; GFX9-NEXT:    v_cndmask_b32_e32 v18, 0, v8, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v19, 0, v9, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v4, v4, v10, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v11, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v8, v5, v11, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v16
 ; GFX9-NEXT:    v_cndmask_b32_e32 v16, v4, v6, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v20, v5, v7, vcc
 ; GFX9-NEXT:    v_lshrrev_b64 v[4:5], 1, v[12:13]
-; GFX9-NEXT:    v_lshlrev_b32_e32 v6, 31, v14
-; GFX9-NEXT:    v_or_b32_e32 v5, v5, v6
+; GFX9-NEXT:    v_cndmask_b32_e32 v12, v8, v7, vcc
 ; GFX9-NEXT:    v_lshrrev_b64 v[6:7], 1, v[14:15]
+; GFX9-NEXT:    v_lshl_or_b32 v5, v14, 31, v5
 ; GFX9-NEXT:    v_sub_u32_e32 v10, 64, v17
 ; GFX9-NEXT:    v_lshrrev_b64 v[8:9], v17, v[4:5]
 ; GFX9-NEXT:    v_lshlrev_b64 v[10:11], v10, v[6:7]
-; GFX9-NEXT:    v_subrev_u32_e32 v12, 64, v17
+; GFX9-NEXT:    v_subrev_u32_e32 v13, 64, v17
 ; GFX9-NEXT:    v_or_b32_e32 v10, v8, v10
 ; GFX9-NEXT:    v_or_b32_e32 v11, v9, v11
 ; GFX9-NEXT:    v_lshrrev_b64 v[8:9], v17, v[6:7]
-; GFX9-NEXT:    v_lshrrev_b64 v[6:7], v12, v[6:7]
+; GFX9-NEXT:    v_lshrrev_b64 v[6:7], v13, v[6:7]
 ; GFX9-NEXT:    v_cmp_gt_u32_e32 vcc, 64, v17
 ; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v10, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v7, v7, v11, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v17
 ; GFX9-NEXT:    v_cndmask_b32_e64 v4, v6, v4, s[4:5]
 ; GFX9-NEXT:    v_cndmask_b32_e64 v5, v7, v5, s[4:5]
 ; GFX9-NEXT:    v_cndmask_b32_e32 v6, 0, v8, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v7, 0, v9, vcc
 ; GFX9-NEXT:    v_or_b32_e32 v4, v18, v4
 ; GFX9-NEXT:    v_or_b32_e32 v5, v19, v5
 ; GFX9-NEXT:    v_or_b32_e32 v6, v16, v6
-; GFX9-NEXT:    v_or_b32_e32 v7, v20, v7
+; GFX9-NEXT:    v_or_b32_e32 v7, v12, v7
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_fshl_v2i128:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v27, 0x7f, v16
 ; GFX10-NEXT:    v_xor_b32_e32 v16, -1, v16
 ; GFX10-NEXT:    v_lshrrev_b64 v[8:9], 1, v[8:9]
-; GFX10-NEXT:    v_lshlrev_b32_e32 v21, 31, v10
-; GFX10-NEXT:    v_lshrrev_b64 v[10:11], 1, v[10:11]
 ; GFX10-NEXT:    v_sub_nc_u32_e32 v17, 64, v27
 ; GFX10-NEXT:    v_and_b32_e32 v28, 0x7f, v16
 ; GFX10-NEXT:    v_lshlrev_b64 v[18:19], v27, v[2:3]
-; GFX10-NEXT:    v_or_b32_e32 v9, v9, v21
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v29, 64, v27
+; GFX10-NEXT:    v_lshl_or_b32 v9, v10, 31, v9
+; GFX10-NEXT:    v_lshrrev_b64 v[10:11], 1, v[10:11]
 ; GFX10-NEXT:    v_lshrrev_b64 v[16:17], v17, v[0:1]
 ; GFX10-NEXT:    v_sub_nc_u32_e32 v25, 64, v28
-; GFX10-NEXT:    v_lshlrev_b64 v[21:22], v27, v[0:1]
+; GFX10-NEXT:    v_subrev_nc_u32_e32 v29, 64, v27
 ; GFX10-NEXT:    v_lshrrev_b64 v[23:24], v28, v[8:9]
-; GFX10-NEXT:    v_lshlrev_b64 v[0:1], v29, v[0:1]
+; GFX10-NEXT:    v_lshlrev_b64 v[21:22], v27, v[0:1]
 ; GFX10-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v27
 ; GFX10-NEXT:    v_or_b32_e32 v18, v16, v18
 ; GFX10-NEXT:    v_subrev_nc_u32_e32 v16, 64, v28
 ; GFX10-NEXT:    v_lshlrev_b64 v[25:26], v25, v[10:11]
+; GFX10-NEXT:    v_lshlrev_b64 v[0:1], v29, v[0:1]
 ; GFX10-NEXT:    v_or_b32_e32 v19, v17, v19
 ; GFX10-NEXT:    v_cmp_gt_u32_e64 s4, 64, v28
-; GFX10-NEXT:    v_cmp_eq_u32_e64 s5, 0, v28
 ; GFX10-NEXT:    v_lshrrev_b64 v[16:17], v16, v[10:11]
-; GFX10-NEXT:    v_cmp_eq_u32_e64 s6, 0, v27
+; GFX10-NEXT:    v_cndmask_b32_e32 v21, 0, v21, vcc_lo
 ; GFX10-NEXT:    v_or_b32_e32 v23, v23, v25
-; GFX10-NEXT:    v_or_b32_e32 v24, v24, v26
-; GFX10-NEXT:    v_cndmask_b32_e32 v19, v1, v19, vcc_lo
 ; GFX10-NEXT:    v_cndmask_b32_e32 v18, v0, v18, vcc_lo
-; GFX10-NEXT:    v_lshrrev_b64 v[0:1], v28, v[10:11]
+; GFX10-NEXT:    v_or_b32_e32 v0, v24, v26
+; GFX10-NEXT:    v_cndmask_b32_e32 v22, 0, v22, vcc_lo
+; GFX10-NEXT:    v_cndmask_b32_e32 v19, v1, v19, vcc_lo
 ; GFX10-NEXT:    v_cndmask_b32_e64 v16, v16, v23, s4
-; GFX10-NEXT:    v_cndmask_b32_e64 v10, v17, v24, s4
-; GFX10-NEXT:    v_cndmask_b32_e32 v21, 0, v21, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e32 v11, 0, v22, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e64 v22, v19, v3, s6
+; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v27
+; GFX10-NEXT:    v_cmp_eq_u32_e64 s5, 0, v28
+; GFX10-NEXT:    v_cndmask_b32_e64 v17, v17, v0, s4
+; GFX10-NEXT:    v_lshrrev_b64 v[0:1], v28, v[10:11]
+; GFX10-NEXT:    v_and_b32_e32 v24, 0x7f, v20
+; GFX10-NEXT:    v_cndmask_b32_e32 v23, v19, v3, vcc_lo
 ; GFX10-NEXT:    v_cndmask_b32_e64 v3, v16, v8, s5
-; GFX10-NEXT:    v_cndmask_b32_e64 v2, v18, v2, s6
-; GFX10-NEXT:    v_cndmask_b32_e64 v8, v10, v9, s5
-; GFX10-NEXT:    v_cndmask_b32_e64 v9, 0, v0, s4
-; GFX10-NEXT:    v_and_b32_e32 v23, 0x7f, v20
+; GFX10-NEXT:    v_cndmask_b32_e64 v8, v17, v9, s5
+; GFX10-NEXT:    v_cndmask_b32_e32 v2, v18, v2, vcc_lo
+; GFX10-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s4
+; GFX10-NEXT:    v_cndmask_b32_e64 v25, 0, v1, s4
 ; GFX10-NEXT:    v_or_b32_e32 v0, v21, v3
 ; GFX10-NEXT:    v_xor_b32_e32 v3, -1, v20
-; GFX10-NEXT:    v_cndmask_b32_e64 v24, 0, v1, s4
-; GFX10-NEXT:    v_or_b32_e32 v1, v11, v8
-; GFX10-NEXT:    v_sub_nc_u32_e32 v10, 64, v23
-; GFX10-NEXT:    v_or_b32_e32 v2, v2, v9
+; GFX10-NEXT:    v_or_b32_e32 v1, v22, v8
 ; GFX10-NEXT:    v_lshrrev_b64 v[8:9], 1, v[12:13]
-; GFX10-NEXT:    v_lshlrev_b32_e32 v16, 31, v14
-; GFX10-NEXT:    v_and_b32_e32 v25, 0x7f, v3
-; GFX10-NEXT:    v_lshrrev_b64 v[10:11], v10, v[4:5]
-; GFX10-NEXT:    v_lshlrev_b64 v[12:13], v23, v[6:7]
+; GFX10-NEXT:    v_sub_nc_u32_e32 v11, 64, v24
+; GFX10-NEXT:    v_or_b32_e32 v2, v2, v10
+; GFX10-NEXT:    v_and_b32_e32 v22, 0x7f, v3
+; GFX10-NEXT:    v_lshlrev_b64 v[12:13], v24, v[6:7]
+; GFX10-NEXT:    v_subrev_nc_u32_e32 v3, 64, v24
+; GFX10-NEXT:    v_lshrrev_b64 v[10:11], v11, v[4:5]
+; GFX10-NEXT:    v_lshl_or_b32 v9, v14, 31, v9
 ; GFX10-NEXT:    v_lshrrev_b64 v[14:15], 1, v[14:15]
-; GFX10-NEXT:    v_or_b32_e32 v9, v9, v16
-; GFX10-NEXT:    v_sub_nc_u32_e32 v20, 64, v25
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v3, 64, v23
-; GFX10-NEXT:    v_lshlrev_b64 v[16:17], v23, v[4:5]
+; GFX10-NEXT:    v_sub_nc_u32_e32 v20, 64, v22
+; GFX10-NEXT:    v_lshlrev_b64 v[16:17], v24, v[4:5]
+; GFX10-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v24
 ; GFX10-NEXT:    v_or_b32_e32 v12, v10, v12
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v10, 64, v25
-; GFX10-NEXT:    v_lshrrev_b64 v[18:19], v25, v[8:9]
+; GFX10-NEXT:    v_subrev_nc_u32_e32 v10, 64, v22
+; GFX10-NEXT:    v_lshrrev_b64 v[18:19], v22, v[8:9]
 ; GFX10-NEXT:    v_lshlrev_b64 v[20:21], v20, v[14:15]
-; GFX10-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v23
 ; GFX10-NEXT:    v_lshlrev_b64 v[3:4], v3, v[4:5]
 ; GFX10-NEXT:    v_or_b32_e32 v5, v11, v13
 ; GFX10-NEXT:    v_lshrrev_b64 v[10:11], v10, v[14:15]
-; GFX10-NEXT:    v_cmp_gt_u32_e64 s4, 64, v25
 ; GFX10-NEXT:    v_cndmask_b32_e32 v13, 0, v16, vcc_lo
+; GFX10-NEXT:    v_cmp_gt_u32_e64 s4, 64, v22
 ; GFX10-NEXT:    v_or_b32_e32 v16, v18, v20
 ; GFX10-NEXT:    v_or_b32_e32 v18, v19, v21
 ; GFX10-NEXT:    v_cndmask_b32_e32 v12, v3, v12, vcc_lo
 ; GFX10-NEXT:    v_cndmask_b32_e32 v5, v4, v5, vcc_lo
-; GFX10-NEXT:    v_lshrrev_b64 v[3:4], v25, v[14:15]
+; GFX10-NEXT:    v_lshrrev_b64 v[3:4], v22, v[14:15]
 ; GFX10-NEXT:    v_cndmask_b32_e64 v10, v10, v16, s4
-; GFX10-NEXT:    v_cmp_eq_u32_e64 s5, 0, v25
-; GFX10-NEXT:    v_cmp_eq_u32_e64 s6, 0, v23
+; GFX10-NEXT:    v_cmp_eq_u32_e64 s5, 0, v22
+; GFX10-NEXT:    v_cmp_eq_u32_e64 s6, 0, v24
 ; GFX10-NEXT:    v_cndmask_b32_e64 v11, v11, v18, s4
 ; GFX10-NEXT:    v_cndmask_b32_e32 v14, 0, v17, vcc_lo
 ; GFX10-NEXT:    v_cndmask_b32_e64 v6, v12, v6, s6
 ; GFX10-NEXT:    v_cndmask_b32_e64 v7, v5, v7, s6
 ; GFX10-NEXT:    v_cndmask_b32_e64 v5, v10, v8, s5
 ; GFX10-NEXT:    v_cndmask_b32_e64 v8, v11, v9, s5
 ; GFX10-NEXT:    v_cndmask_b32_e64 v9, 0, v3, s4
 ; GFX10-NEXT:    v_cndmask_b32_e64 v10, 0, v4, s4
-; GFX10-NEXT:    v_or_b32_e32 v3, v22, v24
+; GFX10-NEXT:    v_or_b32_e32 v3, v23, v25
 ; GFX10-NEXT:    v_or_b32_e32 v4, v13, v5
 ; GFX10-NEXT:    v_or_b32_e32 v5, v14, v8
 ; GFX10-NEXT:    v_or_b32_e32 v6, v6, v9
 ; GFX10-NEXT:    v_or_b32_e32 v7, v7, v10
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_fshl_v2i128:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    v_lshrrev_b64 v[8:9], 1, v[8:9]
 ; GFX11-NEXT:    v_and_b32_e32 v27, 0x7f, v16
-; GFX11-NEXT:    v_lshlrev_b32_e32 v21, 31, v10
 ; GFX11-NEXT:    v_xor_b32_e32 v16, -1, v16
-; GFX11-NEXT:    v_lshrrev_b64 v[10:11], 1, v[10:11]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_sub_nc_u32_e32 v17, 64, v27
-; GFX11-NEXT:    v_or_b32_e32 v9, v9, v21
+; GFX11-NEXT:    v_lshrrev_b64 v[8:9], 1, v[8:9]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_lshlrev_b64 v[21:22], v27, v[0:1]
 ; GFX11-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v27
 ; GFX11-NEXT:    v_and_b32_e32 v28, 0x7f, v16
-; GFX11-NEXT:    v_lshrrev_b64 v[16:17], v17, v[0:1]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX11-NEXT:    v_lshl_or_b32 v9, v10, 31, v9
+; GFX11-NEXT:    v_lshrrev_b64 v[10:11], 1, v[10:11]
+; GFX11-NEXT:    v_dual_cndmask_b32 v21, 0, v21 :: v_dual_cndmask_b32 v22, 0, v22
+; GFX11-NEXT:    v_sub_nc_u32_e32 v17, 64, v27
 ; GFX11-NEXT:    v_lshlrev_b64 v[18:19], v27, v[2:3]
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v29, 64, v27
-; GFX11-NEXT:    v_cndmask_b32_e32 v21, 0, v21, vcc_lo
 ; GFX11-NEXT:    v_sub_nc_u32_e32 v25, 64, v28
+; GFX11-NEXT:    v_subrev_nc_u32_e32 v29, 64, v27
 ; GFX11-NEXT:    v_lshrrev_b64 v[23:24], v28, v[8:9]
+; GFX11-NEXT:    v_lshrrev_b64 v[16:17], v17, v[0:1]
 ; GFX11-NEXT:    v_cmp_gt_u32_e64 s0, 64, v28
-; GFX11-NEXT:    v_or_b32_e32 v18, v16, v18
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v16, 64, v28
 ; GFX11-NEXT:    v_lshlrev_b64 v[25:26], v25, v[10:11]
 ; GFX11-NEXT:    v_lshlrev_b64 v[0:1], v29, v[0:1]
-; GFX11-NEXT:    v_or_b32_e32 v19, v17, v19
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 s1, 0, v28
-; GFX11-NEXT:    v_lshrrev_b64 v[16:17], v16, v[10:11]
+; GFX11-NEXT:    v_or_b32_e32 v18, v16, v18
+; GFX11-NEXT:    v_subrev_nc_u32_e32 v16, 64, v28
+; GFX11-NEXT:    v_or_b32_e32 v19, v17, v19
 ; GFX11-NEXT:    v_or_b32_e32 v23, v23, v25
-; GFX11-NEXT:    v_or_b32_e32 v24, v24, v26
-; GFX11-NEXT:    v_dual_cndmask_b32 v18, v0, v18 :: v_dual_cndmask_b32 v19, v1, v19
-; GFX11-NEXT:    v_cmp_eq_u32_e64 s2, 0, v27
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX11-NEXT:    v_cndmask_b32_e32 v18, v0, v18, vcc_lo
+; GFX11-NEXT:    v_lshrrev_b64 v[16:17], v16, v[10:11]
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_4) | instid1(VALU_DEP_4)
+; GFX11-NEXT:    v_cndmask_b32_e32 v19, v1, v19, vcc_lo
+; GFX11-NEXT:    v_or_b32_e32 v0, v24, v26
+; GFX11-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v27
+; GFX11-NEXT:    v_and_b32_e32 v24, 0x7f, v20
 ; GFX11-NEXT:    v_cndmask_b32_e64 v16, v16, v23, s0
+; GFX11-NEXT:    v_cndmask_b32_e64 v17, v17, v0, s0
 ; GFX11-NEXT:    v_lshrrev_b64 v[0:1], v28, v[10:11]
-; GFX11-NEXT:    v_cndmask_b32_e64 v10, v17, v24, s0
-; GFX11-NEXT:    v_cndmask_b32_e32 v11, 0, v22, vcc_lo
-; GFX11-NEXT:    v_cndmask_b32_e64 v22, v19, v3, s2
+; GFX11-NEXT:    v_dual_cndmask_b32 v2, v18, v2 :: v_dual_cndmask_b32 v23, v19, v3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_cndmask_b32_e64 v3, v16, v8, s1
-; GFX11-NEXT:    v_cndmask_b32_e64 v2, v18, v2, s2
-; GFX11-NEXT:    v_cndmask_b32_e64 v8, v10, v9, s1
-; GFX11-NEXT:    v_cndmask_b32_e64 v9, 0, v0, s0
-; GFX11-NEXT:    v_and_b32_e32 v23, 0x7f, v20
+; GFX11-NEXT:    v_cndmask_b32_e64 v8, v17, v9, s1
+; GFX11-NEXT:    v_sub_nc_u32_e32 v11, 64, v24
+; GFX11-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
+; GFX11-NEXT:    v_cndmask_b32_e64 v25, 0, v1, s0
 ; GFX11-NEXT:    v_or_b32_e32 v0, v21, v3
 ; GFX11-NEXT:    v_xor_b32_e32 v3, -1, v20
-; GFX11-NEXT:    v_cndmask_b32_e64 v24, 0, v1, s0
-; GFX11-NEXT:    v_or_b32_e32 v1, v11, v8
-; GFX11-NEXT:    v_sub_nc_u32_e32 v10, 64, v23
-; GFX11-NEXT:    v_or_b32_e32 v2, v2, v9
+; GFX11-NEXT:    v_or_b32_e32 v1, v22, v8
 ; GFX11-NEXT:    v_lshrrev_b64 v[8:9], 1, v[12:13]
-; GFX11-NEXT:    v_lshlrev_b32_e32 v16, 31, v14
-; GFX11-NEXT:    v_and_b32_e32 v25, 0x7f, v3
-; GFX11-NEXT:    v_lshrrev_b64 v[10:11], v10, v[4:5]
-; GFX11-NEXT:    v_lshlrev_b64 v[12:13], v23, v[6:7]
+; GFX11-NEXT:    v_or_b32_e32 v2, v2, v10
+; GFX11-NEXT:    v_lshrrev_b64 v[10:11], v11, v[4:5]
+; GFX11-NEXT:    v_lshlrev_b64 v[12:13], v24, v[6:7]
+; GFX11-NEXT:    v_lshlrev_b64 v[16:17], v24, v[4:5]
+; GFX11-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v24
+; GFX11-NEXT:    v_and_b32_e32 v22, 0x7f, v3
+; GFX11-NEXT:    v_subrev_nc_u32_e32 v3, 64, v24
+; GFX11-NEXT:    v_lshl_or_b32 v9, v14, 31, v9
 ; GFX11-NEXT:    v_lshrrev_b64 v[14:15], 1, v[14:15]
-; GFX11-NEXT:    v_or_b32_e32 v9, v9, v16
-; GFX11-NEXT:    v_sub_nc_u32_e32 v20, 64, v25
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v3, 64, v23
-; GFX11-NEXT:    v_lshlrev_b64 v[16:17], v23, v[4:5]
 ; GFX11-NEXT:    v_or_b32_e32 v12, v10, v12
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v10, 64, v25
-; GFX11-NEXT:    v_lshrrev_b64 v[18:19], v25, v[8:9]
-; GFX11-NEXT:    v_lshlrev_b64 v[20:21], v20, v[14:15]
-; GFX11-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 64, v23
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX11-NEXT:    v_lshlrev_b64 v[3:4], v3, v[4:5]
 ; GFX11-NEXT:    v_or_b32_e32 v5, v11, v13
-; GFX11-NEXT:    v_lshrrev_b64 v[10:11], v10, v[14:15]
-; GFX11-NEXT:    v_cmp_gt_u32_e64 s0, 64, v25
 ; GFX11-NEXT:    v_cndmask_b32_e32 v13, 0, v16, vcc_lo
+; GFX11-NEXT:    v_sub_nc_u32_e32 v20, 64, v22
+; GFX11-NEXT:    v_subrev_nc_u32_e32 v10, 64, v22
+; GFX11-NEXT:    v_lshrrev_b64 v[18:19], v22, v[8:9]
+; GFX11-NEXT:    v_cmp_gt_u32_e64 s0, 64, v22
+; GFX11-NEXT:    v_cndmask_b32_e32 v12, v3, v12, vcc_lo
+; GFX11-NEXT:    v_lshlrev_b64 v[20:21], v20, v[14:15]
+; GFX11-NEXT:    v_lshrrev_b64 v[10:11], v10, v[14:15]
+; GFX11-NEXT:    v_cndmask_b32_e32 v5, v4, v5, vcc_lo
+; GFX11-NEXT:    v_lshrrev_b64 v[3:4], v22, v[14:15]
+; GFX11-NEXT:    v_cmp_eq_u32_e64 s1, 0, v22
+; GFX11-NEXT:    v_cmp_eq_u32_e64 s2, 0, v24
 ; GFX11-NEXT:    v_or_b32_e32 v16, v18, v20
 ; GFX11-NEXT:    v_or_b32_e32 v18, v19, v21
-; GFX11-NEXT:    v_dual_cndmask_b32 v12, v3, v12 :: v_dual_cndmask_b32 v5, v4, v5
-; GFX11-NEXT:    v_lshrrev_b64 v[3:4], v25, v[14:15]
 ; GFX11-NEXT:    v_cndmask_b32_e32 v14, 0, v17, vcc_lo
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX11-NEXT:    v_cndmask_b32_e64 v6, v12, v6, s2
 ; GFX11-NEXT:    v_cndmask_b32_e64 v10, v10, v16, s0
-; GFX11-NEXT:    v_cmp_eq_u32_e64 s1, 0, v25
-; GFX11-NEXT:    v_cmp_eq_u32_e64 s2, 0, v23
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_cndmask_b32_e64 v11, v11, v18, s0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_4)
-; GFX11-NEXT:    v_cndmask_b32_e64 v6, v12, v6, s2
 ; GFX11-NEXT:    v_cndmask_b32_e64 v7, v5, v7, s2
 ; GFX11-NEXT:    v_cndmask_b32_e64 v5, v10, v8, s1
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX11-NEXT:    v_cndmask_b32_e64 v8, v11, v9, s1
 ; GFX11-NEXT:    v_cndmask_b32_e64 v9, 0, v3, s0
 ; GFX11-NEXT:    v_cndmask_b32_e64 v10, 0, v4, s0
-; GFX11-NEXT:    v_or_b32_e32 v3, v22, v24
+; GFX11-NEXT:    v_or_b32_e32 v3, v23, v25
 ; GFX11-NEXT:    v_or_b32_e32 v4, v13, v5
 ; GFX11-NEXT:    v_or_b32_e32 v5, v14, v8
 ; GFX11-NEXT:    v_or_b32_e32 v6, v6, v9
 ; GFX11-NEXT:    v_or_b32_e32 v7, v7, v10
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %result = call <2 x i128> @llvm.fshl.v2i128(<2 x i128> %lhs, <2 x i128> %rhs, <2 x i128> %amt)
   ret <2 x i128> %result
 }

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

 ; GFX8-NEXT:    v_or_b32_e32 v1, v2, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_i64_5:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX9-NEXT:    v_lshrrev_b64 v[0:1], 5, v[2:3]
-; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 27, v4
-; GFX9-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 27, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_fshr_i64_5:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX10-NEXT:    v_lshrrev_b64 v[0:1], 5, v[2:3]
-; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 27, v4
-; GFX10-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX10-NEXT:    v_lshl_or_b32 v1, v4, 27, v1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_fshr_i64_5:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_mov_b32_e32 v4, v0
 ; GFX11-NEXT:    v_lshrrev_b64 v[0:1], 5, v[2:3]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 27, v4
-; GFX11-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_lshl_or_b32 v1, v4, 27, v1
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %result = call i64 @llvm.fshr.i64(i64 %lhs, i64 %rhs, i64 5)
   ret i64 %result
 }
 
 define i64 @v_fshr_i64_32(i64 %lhs, i64 %rhs) {
 ; GFX6-LABEL: v_fshr_i64_32:
 ; GFX6:       ; %bb.0:

 ; GFX8-NEXT:    v_lshrrev_b64 v[0:1], 1, v[6:7]
 ; GFX8-NEXT:    v_or_b32_e32 v3, v5, v3
 ; GFX8-NEXT:    v_or_b32_e32 v1, v4, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_i128_65:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 31, v0
-; GFX9-NEXT:    v_lshlrev_b32_e32 v5, 31, v2
+; GFX9-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX9-NEXT:    v_lshrrev_b64 v[2:3], 1, v[0:1]
-; GFX9-NEXT:    v_lshrrev_b64 v[0:1], 1, v[6:7]
-; GFX9-NEXT:    v_or_b32_e32 v3, v5, v3
-; GFX9-NEXT:    v_or_b32_e32 v1, v4, v1
+; GFX9-NEXT:    v_lshrrev_b64 v[4:5], 1, v[6:7]
+; GFX9-NEXT:    v_lshl_or_b32 v3, v8, 31, v3
+; GFX9-NEXT:    v_lshl_or_b32 v1, v0, 31, v5
+; GFX9-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_fshr_i128_65:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX10-NEXT:    v_lshrrev_b64 v[4:5], 1, v[6:7]
 ; GFX10-NEXT:    v_lshrrev_b64 v[2:3], 1, v[0:1]
-; GFX10-NEXT:    v_lshlrev_b32_e32 v9, 31, v0
-; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 31, v8
-; GFX10-NEXT:    v_or_b32_e32 v1, v9, v5
-; GFX10-NEXT:    v_or_b32_e32 v3, v0, v3
+; GFX10-NEXT:    v_lshl_or_b32 v1, v0, 31, v5
+; GFX10-NEXT:    v_lshl_or_b32 v3, v8, 31, v3
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: v_fshr_i128_65:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    v_dual_mov_b32 v8, v2 :: v_dual_lshlrev_b32 v9, 31, v0
+; GFX11-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX11-NEXT:    v_lshrrev_b64 v[4:5], 1, v[6:7]
 ; GFX11-NEXT:    v_lshrrev_b64 v[2:3], 1, v[0:1]
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 31, v8
-; GFX11-NEXT:    v_or_b32_e32 v1, v9, v5
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_or_b32_e32 v3, v0, v3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    v_lshl_or_b32 v1, v0, 31, v5
+; GFX11-NEXT:    v_lshl_or_b32 v3, v8, 31, v3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
 ; GFX11-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   %result = call i128 @llvm.fshr.i128(i128 %lhs, i128 %rhs, i128 65)
   ret i128 %result
 }
 
 define amdgpu_ps <2 x i128> @s_fshr_v2i128(<2 x i128> inreg %lhs, <2 x i128> inreg %rhs, <2 x i128> inreg %amt) {
 ; GFX6-LABEL: s_fshr_v2i128:

llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i16.ll

 }
 
 define amdgpu_ps void @insertelement_s_v4i16_v_s(<4 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
 ; GFX9-LABEL: insertelement_s_v4i16_v_s:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
 ; GFX9-NEXT:    s_lshr_b32 s2, s4, 1
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX9-NEXT:    s_and_b32 s4, s4, 1
 ; GFX9-NEXT:    s_lshl_b32 s4, s4, 4
-; GFX9-NEXT:    s_lshl_b32 s5, 0xffff, s4
-; GFX9-NEXT:    s_andn2_b32 s3, s3, s5
-; GFX9-NEXT:    v_mov_b32_e32 v1, s3
-; GFX9-NEXT:    v_lshl_or_b32 v4, v0, s4, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX9-NEXT:    s_not_b32 s4, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NEXT:    v_and_or_b32 v4, s3, v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
 ; GFX9-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: insertelement_s_v4i16_v_s:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
 ; GFX8-NEXT:    s_lshr_b32 s2, s4, 1
 ; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX8-NEXT:    s_and_b32 s4, s4, 1
 ; GFX8-NEXT:    s_lshl_b32 s4, s4, 4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX8-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
-; GFX8-NEXT:    s_andn2_b32 s3, s3, s4
-; GFX8-NEXT:    v_or_b32_e32 v4, s3, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX8-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
 ; GFX8-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: insertelement_s_v4i16_v_s:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
 ; GFX7-NEXT:    s_lshr_b32 s2, s4, 1
 ; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
 ; GFX7-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX7-NEXT:    s_and_b32 s4, s4, 1
 ; GFX7-NEXT:    s_lshl_b32 s4, s4, 4
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
 ; GFX7-NEXT:    s_lshl_b32 s4, 0xffff, s4
-; GFX7-NEXT:    s_andn2_b32 s3, s3, s4
-; GFX7-NEXT:    v_or_b32_e32 v4, s3, v0
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX7-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX7-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v4i16_v_s:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
 ; GFX10-NEXT:    s_lshr_b32 s2, s4, 1
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xffff, v0
 ; GFX10-NEXT:    s_cmp_eq_u32 s2, 1
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX10-NEXT:    s_and_b32 s4, s4, 1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX10-NEXT:    v_lshlrev_b32_sdwa v2, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX10-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    s_not_b32 s4, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-NEXT:    s_lshl_b32 s5, 0xffff, s4
-; GFX10-NEXT:    s_andn2_b32 s3, s3, s5
-; GFX10-NEXT:    v_lshl_or_b32 v4, v2, s4, s3
+; GFX10-NEXT:    v_and_or_b32 v4, s3, s4, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc_lo
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 1
 ; GFX10-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc_lo
 ; GFX10-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: insertelement_s_v4i16_v_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[2:3], 0x0
 ; GFX11-NEXT:    s_lshr_b32 s2, s4, 1
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xffff, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-NEXT:    s_cmp_eq_u32 s2, 1
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX11-NEXT:    s_and_b32 s4, s4, 1
-; GFX11-NEXT:    v_mov_b32_e32 v0, s0
-; GFX11-NEXT:    s_lshl_b32 s4, s4, 4
-; GFX11-NEXT:    v_mov_b32_e32 v1, s1
-; GFX11-NEXT:    s_lshl_b32 s5, 0xffff, s4
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 s3, s3, s5
-; GFX11-NEXT:    v_lshl_or_b32 v4, v2, s4, s3
+; GFX11-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX11-NEXT:    v_lshlrev_b32_e32 v2, s4, v0
+; GFX11-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_not_b32 s4, s4
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_and_or_b32 v4, s3, s4, v2
+; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_4)
 ; GFX11-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_cndmask_b32 v0, v0, v4
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 1
 ; GFX11-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc_lo
 ; GFX11-NEXT:    global_store_b64 v[2:3], v[0:1], off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %vec = load <4 x i16>, <4 x i16> addrspace(4)* %ptr
   %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx

 }
 
 define amdgpu_ps void @insertelement_s_v8i16_v_s(<8 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
 ; GFX9-LABEL: insertelement_s_v8i16_v_s:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
 ; GFX9-NEXT:    s_lshr_b32 s5, s4, 1
 ; GFX9-NEXT:    s_cmp_eq_u32 s5, 1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_cselect_b32 s6, s1, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s5, 2
 ; GFX9-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX9-NEXT:    s_cmp_eq_u32 s5, 3
 ; GFX9-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX9-NEXT:    s_and_b32 s4, s4, 1
 ; GFX9-NEXT:    s_lshl_b32 s4, s4, 4
-; GFX9-NEXT:    s_lshl_b32 s7, 0xffff, s4
-; GFX9-NEXT:    s_andn2_b32 s6, s6, s7
-; GFX9-NEXT:    v_mov_b32_e32 v1, s6
-; GFX9-NEXT:    v_lshl_or_b32 v6, v0, s4, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX9-NEXT:    s_not_b32 s4, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NEXT:    v_and_or_b32 v6, s6, v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
 ; GFX9-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
 ; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: insertelement_s_v8i16_v_s:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
 ; GFX8-NEXT:    s_lshr_b32 s5, s4, 1
 ; GFX8-NEXT:    s_cmp_eq_u32 s5, 1
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_cselect_b32 s6, s1, s0
 ; GFX8-NEXT:    s_cmp_eq_u32 s5, 2
 ; GFX8-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX8-NEXT:    s_cmp_eq_u32 s5, 3
 ; GFX8-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX8-NEXT:    s_and_b32 s4, s4, 1
 ; GFX8-NEXT:    s_lshl_b32 s4, s4, 4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX8-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
-; GFX8-NEXT:    s_andn2_b32 s4, s6, s4
-; GFX8-NEXT:    v_or_b32_e32 v6, s4, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX8-NEXT:    v_or_b32_e32 v6, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
 ; GFX8-NEXT:    v_mov_b32_e32 v3, s3

 ; GFX7-NEXT:    s_cmp_eq_u32 s5, 2
 ; GFX7-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX7-NEXT:    s_cmp_eq_u32 s5, 3
 ; GFX7-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX7-NEXT:    s_and_b32 s4, s4, 1
 ; GFX7-NEXT:    s_lshl_b32 s4, s4, 4
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
 ; GFX7-NEXT:    s_lshl_b32 s4, 0xffff, s4
-; GFX7-NEXT:    s_andn2_b32 s4, s6, s4
-; GFX7-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
 ; GFX7-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
 ; GFX7-NEXT:    s_mov_b64 s[0:1], 0
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v8i16_v_s:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
 ; GFX10-NEXT:    s_lshr_b32 s5, s4, 1
-; GFX10-NEXT:    v_and_b32_e32 v4, 0xffff, v0
 ; GFX10-NEXT:    s_cmp_eq_u32 s5, 1
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_cselect_b32 s6, s1, s0
 ; GFX10-NEXT:    s_cmp_eq_u32 s5, 2
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX10-NEXT:    s_cmp_eq_u32 s5, 3
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX10-NEXT:    s_and_b32 s4, s4, 1
-; GFX10-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX10-NEXT:    v_lshlrev_b32_sdwa v4, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX10-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    s_not_b32 s4, s4
+; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_and_or_b32 v6, s6, s4, v4
+; GFX10-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s3
-; GFX10-NEXT:    s_lshl_b32 s7, 0xffff, s4
-; GFX10-NEXT:    s_andn2_b32 s6, s6, s7
-; GFX10-NEXT:    v_lshl_or_b32 v6, v4, s4, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 1
 ; GFX10-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc_lo
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 2
 ; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc_lo
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 3
 ; GFX10-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc_lo
 ; GFX10-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: insertelement_s_v8i16_v_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[2:3], 0x0
 ; GFX11-NEXT:    s_lshr_b32 s5, s4, 1
-; GFX11-NEXT:    v_and_b32_e32 v4, 0xffff, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-NEXT:    s_cmp_eq_u32 s5, 1
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_cselect_b32 s6, s1, s0
 ; GFX11-NEXT:    s_cmp_eq_u32 s5, 2
-; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s3
 ; GFX11-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX11-NEXT:    s_cmp_eq_u32 s5, 3
-; GFX11-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX11-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX11-NEXT:    s_and_b32 s4, s4, 1
-; GFX11-NEXT:    v_mov_b32_e32 v2, s2
-; GFX11-NEXT:    s_lshl_b32 s4, s4, 4
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_lshl_b32 s7, 0xffff, s4
-; GFX11-NEXT:    s_and_not1_b32 s6, s6, s7
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v6, v4, s4, s6
-; GFX11-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc_lo
+; GFX11-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX11-NEXT:    v_lshlrev_b32_e32 v4, s4, v0
+; GFX11-NEXT:    s_lshl_b32 s4, 0xffff, s4
+; GFX11-NEXT:    v_mov_b32_e32 v0, s0
+; GFX11-NEXT:    s_not_b32 s4, s4
+; GFX11-NEXT:    v_mov_b32_e32 v3, s3
+; GFX11-NEXT:    v_and_or_b32 v6, s6, s4, v4
+; GFX11-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX11-NEXT:    v_dual_mov_b32 v5, 0 :: v_dual_cndmask_b32 v0, v0, v6
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 1
-; GFX11-NEXT:    v_mov_b32_e32 v4, 0
-; GFX11-NEXT:    v_mov_b32_e32 v5, 0
-; GFX11-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc_lo
+; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_cndmask_b32 v1, v1, v6
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc_lo
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s5, 3
 ; GFX11-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc_lo
 ; GFX11-NEXT:    global_store_b128 v[4:5], v[0:3], off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %vec = load <8 x i16>, <8 x i16> addrspace(4)* %ptr
   %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx

 }
 
 define amdgpu_ps void @insertelement_s_v16i16_v_s(<16 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
 ; GFX9-LABEL: insertelement_s_v16i16_v_s:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
 ; GFX9-NEXT:    s_lshr_b32 s2, s4, 1
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v10, 16
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_cselect_b32 s0, s9, s8
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 2
 ; GFX9-NEXT:    s_cselect_b32 s0, s10, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 3
 ; GFX9-NEXT:    s_cselect_b32 s0, s11, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 4
 ; GFX9-NEXT:    s_cselect_b32 s0, s12, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 5
 ; GFX9-NEXT:    s_cselect_b32 s0, s13, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 6
 ; GFX9-NEXT:    s_cselect_b32 s0, s14, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s2, 7
 ; GFX9-NEXT:    s_cselect_b32 s0, s15, s0
 ; GFX9-NEXT:    s_and_b32 s1, s4, 1
 ; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
-; GFX9-NEXT:    s_lshl_b32 s3, 0xffff, s1
-; GFX9-NEXT:    s_andn2_b32 s0, s0, s3
-; GFX9-NEXT:    v_mov_b32_e32 v1, s0
-; GFX9-NEXT:    v_lshl_or_b32 v8, v0, s1, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_lshl_b32 s1, 0xffff, s1
+; GFX9-NEXT:    s_not_b32 s1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_and_or_b32 v8, s0, v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s11

 ; GFX9-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s2, 4
 ; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v8, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 7
 ; GFX9-NEXT:    v_cndmask_b32_e64 v4, v4, v8, s[0:1]
 ; GFX9-NEXT:    v_cndmask_b32_e32 v7, v7, v8, vcc
 ; GFX9-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9-NEXT:    v_mov_b32_e32 v10, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v11, 0
 ; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
 ; GFX9-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: insertelement_s_v16i16_v_s:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
 ; GFX8-NEXT:    s_and_b32 s1, s4, 1
-; GFX8-NEXT:    s_lshr_b32 m0, s4, 1
 ; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_lshl_b32 s1, 0xffff, s1
+; GFX8-NEXT:    s_lshr_b32 m0, s4, 1
+; GFX8-NEXT:    s_not_b32 s1, s1
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_movrels_b32 s0, s8
-; GFX8-NEXT:    s_lshl_b32 s1, 0xffff, s1
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
-; GFX8-NEXT:    s_andn2_b32 s0, s0, s1
-; GFX8-NEXT:    v_or_b32_e32 v8, s0, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX8-NEXT:    v_or_b32_e32 v8, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX8-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s12
 ; GFX8-NEXT:    v_mov_b32_e32 v5, s13
 ; GFX8-NEXT:    v_mov_b32_e32 v6, s14
 ; GFX8-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX8-NEXT:    v_movreld_b32_e32 v0, v8
 ; GFX8-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v9, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v10, 16
 ; GFX8-NEXT:    v_mov_b32_e32 v11, 0
 ; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
 ; GFX8-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: insertelement_s_v16i16_v_s:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
 ; GFX7-NEXT:    s_and_b32 s1, s4, 1
-; GFX7-NEXT:    s_lshr_b32 m0, s4, 1
 ; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
 ; GFX7-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-NEXT:    s_movrels_b32 s0, s8
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
 ; GFX7-NEXT:    s_lshl_b32 s1, 0xffff, s1
-; GFX7-NEXT:    s_andn2_b32 s0, s0, s1
-; GFX7-NEXT:    v_or_b32_e32 v8, s0, v0
+; GFX7-NEXT:    s_lshr_b32 m0, s4, 1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_movrels_b32 s0, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_or_b32_e32 v8, v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s12
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s13
 ; GFX7-NEXT:    v_mov_b32_e32 v6, s14
 ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX7-NEXT:    s_mov_b64 s[0:1], 0
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    v_movreld_b32_e32 v0, v8
 ; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX7-NEXT:    s_mov_b64 s[0:1], 16
 ; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v16i16_v_s:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
 ; GFX10-NEXT:    s_and_b32 s0, s4, 1
 ; GFX10-NEXT:    s_lshr_b32 m0, s4, 1
 ; GFX10-NEXT:    s_lshl_b32 s0, s0, 4
-; GFX10-NEXT:    v_and_b32_e32 v8, 0xffff, v0
-; GFX10-NEXT:    s_lshl_b32 s1, 0xffff, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v10, 16
+; GFX10-NEXT:    v_lshlrev_b32_sdwa v8, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX10-NEXT:    s_lshl_b32 s0, 0xffff, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v11, 0
+; GFX10-NEXT:    s_not_b32 s0, s0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_movrels_b32 s2, s8
+; GFX10-NEXT:    s_movrels_b32 s1, s8
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s8
-; GFX10-NEXT:    s_andn2_b32 s1, s2, s1
+; GFX10-NEXT:    v_and_or_b32 v12, s1, s0, v8
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s9
-; GFX10-NEXT:    v_lshl_or_b32 v12, v8, s0, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s12
 ; GFX10-NEXT:    v_mov_b32_e32 v5, s13
 ; GFX10-NEXT:    v_mov_b32_e32 v6, s14
 ; GFX10-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX10-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v9, 0
 ; GFX10-NEXT:    v_movreld_b32_e32 v0, v12
 ; GFX10-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
 ; GFX10-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: insertelement_s_v16i16_v_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b256 s[8:15], s[2:3], 0x0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX11-NEXT:    s_and_b32 s0, s4, 1
 ; GFX11-NEXT:    s_lshr_b32 m0, s4, 1
 ; GFX11-NEXT:    s_lshl_b32 s0, s0, 4
-; GFX11-NEXT:    v_and_b32_e32 v8, 0xffff, v0
-; GFX11-NEXT:    s_lshl_b32 s1, 0xffff, s0
 ; GFX11-NEXT:    v_mov_b32_e32 v10, 16
+; GFX11-NEXT:    v_lshlrev_b32_e32 v8, s0, v0
+; GFX11-NEXT:    s_lshl_b32 s1, 0xffff, s0
 ; GFX11-NEXT:    v_mov_b32_e32 v11, 0
+; GFX11-NEXT:    s_not_b32 s0, s1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_movrels_b32 s2, s8
-; GFX11-NEXT:    v_dual_mov_b32 v0, s8 :: v_dual_mov_b32 v3, s11
-; GFX11-NEXT:    s_and_not1_b32 s1, s2, s1
-; GFX11-NEXT:    v_mov_b32_e32 v1, s9
-; GFX11-NEXT:    v_lshl_or_b32 v12, v8, s0, s1
-; GFX11-NEXT:    v_dual_mov_b32 v2, s10 :: v_dual_mov_b32 v5, s13
-; GFX11-NEXT:    v_dual_mov_b32 v4, s12 :: v_dual_mov_b32 v7, s15
+; GFX11-NEXT:    s_movrels_b32 s1, s8
+; GFX11-NEXT:    v_dual_mov_b32 v0, s8 :: v_dual_mov_b32 v5, s13
+; GFX11-NEXT:    v_mov_b32_e32 v3, s11
+; GFX11-NEXT:    v_and_or_b32 v12, s1, s0, v8
+; GFX11-NEXT:    v_dual_mov_b32 v1, s9 :: v_dual_mov_b32 v2, s10
+; GFX11-NEXT:    v_dual_mov_b32 v7, s15 :: v_dual_mov_b32 v4, s12
 ; GFX11-NEXT:    v_mov_b32_e32 v6, s14
 ; GFX11-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX11-NEXT:    v_mov_b32_e32 v9, 0
 ; GFX11-NEXT:    v_movreld_b32_e32 v0, v12
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    global_store_b128 v[8:9], v[0:3], off
 ; GFX11-NEXT:    global_store_b128 v[10:11], v[4:7], off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)

llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i8.ll

; GFX11-NEXT:    s_endpgm
   store <4 x i8> %insert, <4 x i8> addrspace(1)* null
   ret void
 }
 
 define amdgpu_ps void @insertelement_s_v4i8_v_s(<4 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
 ; GFX9-LABEL: insertelement_s_v4i8_v_s:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX9-NEXT:    s_mov_b32 s1, 8
 ; GFX9-NEXT:    s_mov_b32 s2, 16
 ; GFX9-NEXT:    s_movk_i32 s5, 0xff
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s7, s0, 0x80008
 ; GFX9-NEXT:    s_lshr_b32 s3, s0, 24
-; GFX9-NEXT:    s_and_b32 s6, s0, 0xff
 ; GFX9-NEXT:    s_lshl_b32 s7, s7, 8
+; GFX9-NEXT:    s_and_b32 s6, s0, 0xff
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX9-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX9-NEXT:    s_or_b32 s6, s6, s7
-; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX9-NEXT:    s_or_b32 s0, s6, s0
 ; GFX9-NEXT:    s_lshl_b32 s3, s3, 24
-; GFX9-NEXT:    s_or_b32 s0, s0, s3
-; GFX9-NEXT:    s_and_b32 s3, s4, 3
-; GFX9-NEXT:    s_lshl_b32 s3, s3, 3
-; GFX9-NEXT:    s_lshl_b32 s4, 0xff, s3
-; GFX9-NEXT:    s_andn2_b32 s0, s0, s4
-; GFX9-NEXT:    v_mov_b32_e32 v1, s0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, s3, v1
+; GFX9-NEXT:    v_or_b32_e32 v1, s6, v1
+; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
+; GFX9-NEXT:    v_mov_b32_e32 v2, s3
+; GFX9-NEXT:    v_or3_b32 v1, v1, s0, v2
+; GFX9-NEXT:    s_and_b32 s0, s4, 3
+; GFX9-NEXT:    s_lshl_b32 s0, s0, 3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_lshl_b32 s0, 0xff, s0
+; GFX9-NEXT:    s_not_b32 s0, s0
+; GFX9-NEXT:    v_and_or_b32 v0, v1, s0, v0
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX9-NEXT:    v_and_or_b32 v2, v0, s5, v2
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, s2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 24, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    v_or3_b32 v2, v2, v3, v4
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: insertelement_s_v4i8_v_s:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
-; GFX8-NEXT:    v_mov_b32_e32 v2, 8
 ; GFX8-NEXT:    v_mov_b32_e32 v3, 16
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_bfe_u32 s3, s0, 0x80008
+; GFX8-NEXT:    s_lshl_b32 s3, s3, 8
 ; GFX8-NEXT:    s_lshr_b32 s1, s0, 24
 ; GFX8-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX8-NEXT:    s_lshl_b32 s3, s3, 8
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX8-NEXT:    s_or_b32 s2, s2, s3
+; GFX8-NEXT:    v_or_b32_e32 v1, s2, v1
 ; GFX8-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX8-NEXT:    s_or_b32 s0, s2, s0
-; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX8-NEXT:    s_or_b32 s0, s0, s1
-; GFX8-NEXT:    s_and_b32 s1, s4, 3
-; GFX8-NEXT:    s_lshl_b32 s1, s1, 3
-; GFX8-NEXT:    v_mov_b32_e32 v1, s1
-; GFX8-NEXT:    s_lshl_b32 s1, 0xff, s1
-; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX8-NEXT:    s_andn2_b32 s0, s0, s1
-; GFX8-NEXT:    v_or_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX8-NEXT:    s_lshl_b32 s0, s1, 24
+; GFX8-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX8-NEXT:    s_and_b32 s0, s4, 3
+; GFX8-NEXT:    s_lshl_b32 s0, s0, 3
+; GFX8-NEXT:    v_mov_b32_e32 v2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, 0xff, s0
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX8-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX8-NEXT:    v_or_b32_e32 v2, v2, v0
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 24, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: insertelement_s_v4i8_v_s:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX7-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_bfe_u32 s3, s0, 0x80008
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
 ; GFX7-NEXT:    s_lshr_b32 s1, s0, 24
 ; GFX7-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX7-NEXT:    s_or_b32 s2, s2, s3
+; GFX7-NEXT:    v_or_b32_e32 v1, s2, v1
 ; GFX7-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX7-NEXT:    s_or_b32 s0, s2, s0
-; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX7-NEXT:    s_or_b32 s0, s0, s1
-; GFX7-NEXT:    s_and_b32 s1, s4, 3
-; GFX7-NEXT:    s_lshl_b32 s1, s1, 3
-; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
-; GFX7-NEXT:    s_lshl_b32 s1, 0xff, s1
-; GFX7-NEXT:    s_andn2_b32 s0, s0, s1
-; GFX7-NEXT:    v_or_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX7-NEXT:    s_lshl_b32 s0, s1, 24
+; GFX7-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX7-NEXT:    s_and_b32 s0, s4, 3
+; GFX7-NEXT:    s_lshl_b32 s0, s0, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s0, v0
+; GFX7-NEXT:    s_lshl_b32 s0, 0xff, s0
+; GFX7-NEXT:    s_not_b32 s0, s0
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
 ; GFX7-NEXT:    v_bfe_u32 v3, v0, 8, 8
 ; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX7-NEXT:    v_and_b32_e32 v2, 0xff, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
 ; GFX7-NEXT:    v_bfe_u32 v0, v0, 16, 8
 ; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX7-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX7-NEXT:    s_mov_b64 s[0:1], 0
 ; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v4i8_v_s:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX10-NEXT:    s_and_b32 s1, s4, 3
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX10-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX10-NEXT:    v_lshlrev_b32_sdwa v0, s1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_bfe_u32 s4, s0, 0x80008
-; GFX10-NEXT:    s_lshr_b32 s2, s0, 24
+; GFX10-NEXT:    s_bfe_u32 s2, s0, 0x80008
 ; GFX10-NEXT:    s_and_b32 s3, s0, 0xff
-; GFX10-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX10-NEXT:    s_lshl_b32 s4, s4, 8
-; GFX10-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX10-NEXT:    s_or_b32 s3, s3, s4
-; GFX10-NEXT:    s_lshl_b32 s2, s2, 24
-; GFX10-NEXT:    s_or_b32 s0, s3, s0
-; GFX10-NEXT:    s_lshl_b32 s3, 0xff, s1
-; GFX10-NEXT:    s_or_b32 s0, s0, s2
-; GFX10-NEXT:    s_andn2_b32 s0, s0, s3
-; GFX10-NEXT:    v_lshl_or_b32 v0, v0, s1, s0
+; GFX10-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX10-NEXT:    s_bfe_u32 s4, s0, 0x80010
+; GFX10-NEXT:    v_or_b32_e64 v1, s3, s2
+; GFX10-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX10-NEXT:    s_lshl_b32 s2, s4, 16
+; GFX10-NEXT:    s_lshl_b32 s0, s0, 24
+; GFX10-NEXT:    v_or3_b32 v1, v1, s2, s0
+; GFX10-NEXT:    s_lshl_b32 s0, 0xff, s1
+; GFX10-NEXT:    s_not_b32 s0, s0
+; GFX10-NEXT:    v_and_or_b32 v0, v1, s0, v0
 ; GFX10-NEXT:    s_mov_b32 s0, 8
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v1, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v2, 24, v0
 ; GFX10-NEXT:    s_mov_b32 s0, 16
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v3, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX10-NEXT:    v_and_or_b32 v4, v0, 0xff, v1
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    v_or3_b32 v2, v4, v3, v2
 ; GFX10-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: insertelement_s_v4i8_v_s:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b32 s0, s[2:3], 0x0
-; GFX11-NEXT:    s_and_b32 s1, s4, 3
 ; GFX11-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX11-NEXT:    s_and_b32 s1, s4, 3
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_bfe_u32 s4, s0, 0x80008
-; GFX11-NEXT:    s_lshr_b32 s2, s0, 24
+; GFX11-NEXT:    s_bfe_u32 s2, s0, 0x80008
 ; GFX11-NEXT:    s_and_b32 s3, s0, 0xff
-; GFX11-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX11-NEXT:    s_lshl_b32 s4, s4, 8
-; GFX11-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX11-NEXT:    s_or_b32 s3, s3, s4
-; GFX11-NEXT:    s_lshl_b32 s2, s2, 24
-; GFX11-NEXT:    s_or_b32 s0, s3, s0
-; GFX11-NEXT:    s_lshl_b32 s3, 0xff, s1
-; GFX11-NEXT:    s_or_b32 s0, s0, s2
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 s0, s0, s3
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, s1, s0
+; GFX11-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX11-NEXT:    s_bfe_u32 s4, s0, 0x80010
+; GFX11-NEXT:    v_or_b32_e64 v1, s3, s2
+; GFX11-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX11-NEXT:    s_lshl_b32 s2, s4, 16
+; GFX11-NEXT:    s_lshl_b32 s0, s0, 24
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_or3_b32 v1, v1, s2, s0
+; GFX11-NEXT:    s_lshl_b32 s0, 0xff, s1
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_not_b32 s0, s0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_and_or_b32 v0, v1, s0, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_bfe_u32 v1, v0, 8, 8
 ; GFX11-NEXT:    v_bfe_u32 v2, v0, 16, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v3, 24, v3

 ; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX9-NEXT:    v_and_b32_e32 v0, 3, v0
 ; GFX9-NEXT:    s_movk_i32 s5, 0xff
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; GFX9-NEXT:    s_mov_b32 s1, 8
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s7, s0, 0x80008
 ; GFX9-NEXT:    s_lshr_b32 s3, s0, 24
-; GFX9-NEXT:    s_and_b32 s6, s0, 0xff
 ; GFX9-NEXT:    s_lshl_b32 s7, s7, 8
+; GFX9-NEXT:    s_and_b32 s6, s0, 0xff
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX9-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX9-NEXT:    s_or_b32 s6, s6, s7
-; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX9-NEXT:    s_or_b32 s0, s6, s0
 ; GFX9-NEXT:    s_lshl_b32 s3, s3, 24
-; GFX9-NEXT:    s_or_b32 s0, s0, s3
-; GFX9-NEXT:    s_and_b32 s3, s4, 0xff
-; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v0, s3
+; GFX9-NEXT:    v_or_b32_e32 v1, s6, v1
+; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
+; GFX9-NEXT:    v_mov_b32_e32 v2, s3
+; GFX9-NEXT:    v_or3_b32 v1, v1, s0, v2
+; GFX9-NEXT:    s_and_b32 s0, s4, 0xff
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v0, s0
 ; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s5
 ; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
-; GFX9-NEXT:    v_and_or_b32 v0, s0, v0, v1
+; GFX9-NEXT:    v_and_or_b32 v0, v1, v0, v2
 ; GFX9-NEXT:    s_mov_b32 s2, 16
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX9-NEXT:    v_and_or_b32 v2, v0, s5, v2
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, s2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 24, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    v_or3_b32 v2, v2, v3, v4
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: insertelement_s_v4i8_s_v:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX8-NEXT:    v_and_b32_e32 v0, 3, v0
 ; GFX8-NEXT:    s_movk_i32 s1, 0xff
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    v_mov_b32_e32 v3, 16
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_bfe_u32 s5, s0, 0x80008
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
 ; GFX8-NEXT:    s_lshr_b32 s2, s0, 24
 ; GFX8-NEXT:    s_and_b32 s3, s0, 0xff
-; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX8-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX8-NEXT:    s_or_b32 s3, s3, s5
+; GFX8-NEXT:    v_or_b32_e32 v1, s3, v1
 ; GFX8-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX8-NEXT:    s_or_b32 s0, s3, s0
-; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
-; GFX8-NEXT:    s_or_b32 s0, s0, s2
-; GFX8-NEXT:    s_and_b32 s2, s4, 0xff
-; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v0, s2
+; GFX8-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX8-NEXT:    s_lshl_b32 s0, s2, 24
+; GFX8-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX8-NEXT:    s_and_b32 s0, s4, 0xff
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v0, s0
 ; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s1
 ; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
-; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX8-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
-; GFX8-NEXT:    v_mov_b32_e32 v3, 16
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX8-NEXT:    v_or_b32_e32 v2, v2, v0
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 24, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: insertelement_s_v4i8_s_v:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX7-NEXT:    v_and_b32_e32 v0, 3, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_bfe_u32 s3, s0, 0x80008
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
 ; GFX7-NEXT:    s_lshr_b32 s1, s0, 24
 ; GFX7-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX7-NEXT:    s_or_b32 s2, s2, s3
+; GFX7-NEXT:    v_or_b32_e32 v1, s2, v1
 ; GFX7-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX7-NEXT:    s_or_b32 s0, s2, s0
-; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX7-NEXT:    s_or_b32 s0, s0, s1
-; GFX7-NEXT:    s_and_b32 s1, s4, 0xff
-; GFX7-NEXT:    v_lshl_b32_e32 v1, s1, v0
+; GFX7-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX7-NEXT:    s_lshl_b32 s0, s1, 24
+; GFX7-NEXT:    v_or_b32_e32 v1, s0, v1
+; GFX7-NEXT:    s_and_b32 s0, s4, 0xff
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s0, v0
 ; GFX7-NEXT:    v_lshl_b32_e32 v0, 0xff, v0
 ; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
-; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX7-NEXT:    v_bfe_u32 v3, v0, 8, 8
 ; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX7-NEXT:    v_and_b32_e32 v2, 0xff, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
 ; GFX7-NEXT:    v_bfe_u32 v0, v0, 16, 8
 ; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX7-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX7-NEXT:    s_mov_b64 s[0:1], 0
 ; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v4i8_s_v:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v0, 3, v0
 ; GFX10-NEXT:    s_and_b32 s1, s4, 0xff
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX10-NEXT:    v_lshlrev_b32_e64 v1, v0, 0xff
-; GFX10-NEXT:    v_lshlrev_b32_e64 v0, v0, s1
-; GFX10-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX10-NEXT:    v_lshlrev_b32_e64 v1, v0, s1
+; GFX10-NEXT:    v_lshlrev_b32_e64 v0, v0, 0xff
+; GFX10-NEXT:    v_xor_b32_e32 v0, -1, v0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_bfe_u32 s3, s0, 0x80008
-; GFX10-NEXT:    s_lshr_b32 s1, s0, 24
+; GFX10-NEXT:    s_bfe_u32 s1, s0, 0x80008
 ; GFX10-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX10-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX10-NEXT:    s_lshl_b32 s3, s3, 8
-; GFX10-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX10-NEXT:    s_or_b32 s2, s2, s3
-; GFX10-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX10-NEXT:    s_or_b32 s0, s2, s0
-; GFX10-NEXT:    s_or_b32 s0, s0, s1
-; GFX10-NEXT:    v_and_or_b32 v0, s0, v1, v0
+; GFX10-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX10-NEXT:    s_bfe_u32 s3, s0, 0x80010
+; GFX10-NEXT:    v_or_b32_e64 v2, s2, s1
+; GFX10-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX10-NEXT:    s_lshl_b32 s1, s3, 16
+; GFX10-NEXT:    s_lshl_b32 s0, s0, 24
+; GFX10-NEXT:    v_or3_b32 v2, v2, s1, s0
 ; GFX10-NEXT:    s_mov_b32 s0, 8
+; GFX10-NEXT:    v_and_or_b32 v0, v2, v0, v1
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v1, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v2, 24, v0
 ; GFX10-NEXT:    s_mov_b32 s0, 16
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v3, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX10-NEXT:    v_and_or_b32 v4, v0, 0xff, v1
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    v_or3_b32 v2, v4, v3, v2
 ; GFX10-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: insertelement_s_v4i8_s_v:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b32 s0, s[2:3], 0x0
 ; GFX11-NEXT:    v_and_b32_e32 v0, 3, v0
 ; GFX11-NEXT:    s_and_b32 s1, s4, 0xff
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX11-NEXT:    v_lshlrev_b32_e64 v1, v0, 0xff
-; GFX11-NEXT:    v_lshlrev_b32_e64 v0, v0, s1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX11-NEXT:    v_lshlrev_b32_e64 v1, v0, s1
+; GFX11-NEXT:    v_lshlrev_b32_e64 v0, v0, 0xff
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_xor_b32_e32 v0, -1, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_bfe_u32 s3, s0, 0x80008
-; GFX11-NEXT:    s_lshr_b32 s1, s0, 24
+; GFX11-NEXT:    s_bfe_u32 s1, s0, 0x80008
 ; GFX11-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX11-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX11-NEXT:    s_lshl_b32 s3, s3, 8
-; GFX11-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX11-NEXT:    s_or_b32 s2, s2, s3
-; GFX11-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX11-NEXT:    s_or_b32 s0, s2, s0
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_or_b32 s0, s0, s1
-; GFX11-NEXT:    v_and_or_b32 v0, s0, v1, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3)
+; GFX11-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX11-NEXT:    s_bfe_u32 s3, s0, 0x80010
+; GFX11-NEXT:    v_or_b32_e64 v2, s2, s1
+; GFX11-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX11-NEXT:    s_lshl_b32 s1, s3, 16
+; GFX11-NEXT:    s_lshl_b32 s0, s0, 24
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_or3_b32 v2, v2, s1, s0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-NEXT:    v_and_or_b32 v0, v2, v0, v1
 ; GFX11-NEXT:    v_bfe_u32 v1, v0, 8, 8
 ; GFX11-NEXT:    v_bfe_u32 v2, v0, 16, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
-; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_and_or_b32 v4, v0, 0xff, v1
 ; GFX11-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX11-NEXT:    v_or3_b32 v2, v4, v2, v3
 ; GFX11-NEXT:    global_store_b32 v[0:1], v2, off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %vec = load <4 x i8>, <4 x i8> addrspace(4)* %ptr
   %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
   store <4 x i8> %insert, <4 x i8> addrspace(1)* null
   ret void
 }
 
 define amdgpu_ps void @insertelement_s_v4i8_v_v(<4 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 %idx) {
 ; GFX9-LABEL: insertelement_s_v4i8_v_v:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX9-NEXT:    v_and_b32_e32 v1, 3, v1
 ; GFX9-NEXT:    s_movk_i32 s4, 0xff
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s6, s0, 0x80008
 ; GFX9-NEXT:    s_lshr_b32 s3, s0, 24
-; GFX9-NEXT:    s_and_b32 s5, s0, 0xff
 ; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_and_b32 s5, s0, 0xff
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX9-NEXT:    s_or_b32 s5, s5, s6
-; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX9-NEXT:    s_or_b32 s0, s5, s0
 ; GFX9-NEXT:    s_lshl_b32 s3, s3, 24
+; GFX9-NEXT:    v_or_b32_e32 v2, s5, v2
+; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s4
-; GFX9-NEXT:    s_or_b32 s0, s0, s3
+; GFX9-NEXT:    v_or3_b32 v2, v2, s0, v3
 ; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
 ; GFX9-NEXT:    s_mov_b32 s1, 8
-; GFX9-NEXT:    v_and_or_b32 v0, s0, v1, v0
+; GFX9-NEXT:    v_and_or_b32 v0, v2, v1, v0
 ; GFX9-NEXT:    s_mov_b32 s2, 16
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX9-NEXT:    v_and_or_b32 v2, v0, s4, v2
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, s2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 24, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    v_or3_b32 v2, v2, v3, v4
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: insertelement_s_v4i8_v_v:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX8-NEXT:    v_and_b32_e32 v1, 3, v1
 ; GFX8-NEXT:    s_movk_i32 s1, 0xff
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_bfe_u32 s4, s0, 0x80008
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
 ; GFX8-NEXT:    s_lshr_b32 s2, s0, 24
 ; GFX8-NEXT:    s_and_b32 s3, s0, 0xff
-; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX8-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX8-NEXT:    s_or_b32 s3, s3, s4
+; GFX8-NEXT:    v_or_b32_e32 v2, s3, v2
 ; GFX8-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX8-NEXT:    s_or_b32 s0, s3, s0
-; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    v_or_b32_e32 v2, s0, v2
+; GFX8-NEXT:    s_lshl_b32 s0, s2, 24
 ; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s1
-; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    v_or_b32_e32 v2, s0, v2
 ; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
-; GFX8-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, v2, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v2, 8
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX8-NEXT:    v_mov_b32_e32 v3, 16
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX8-NEXT:    v_or_b32_e32 v2, v2, v0
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 24, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: insertelement_s_v4i8_v_v:
 ; GFX7:       ; %bb.0:
 ; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX7-NEXT:    v_and_b32_e32 v1, 3, v1
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
 ; GFX7-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_bfe_u32 s3, s0, 0x80008
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
 ; GFX7-NEXT:    s_lshr_b32 s1, s0, 24
 ; GFX7-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
+; GFX7-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX7-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX7-NEXT:    s_or_b32 s2, s2, s3
+; GFX7-NEXT:    v_or_b32_e32 v2, s2, v2
 ; GFX7-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX7-NEXT:    s_or_b32 s0, s2, s0
-; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    v_or_b32_e32 v2, s0, v2
+; GFX7-NEXT:    s_lshl_b32 s0, s1, 24
 ; GFX7-NEXT:    v_lshl_b32_e32 v1, 0xff, v1
-; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    v_or_b32_e32 v2, s0, v2
 ; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
-; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
 ; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
 ; GFX7-NEXT:    v_bfe_u32 v3, v0, 8, 8
 ; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 24, v0
 ; GFX7-NEXT:    v_and_b32_e32 v2, 0xff, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
 ; GFX7-NEXT:    v_bfe_u32 v0, v0, 16, 8
 ; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX7-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX7-NEXT:    s_mov_b64 s[0:1], 0
 ; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v4i8_v_v:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dword s0, s[2:3], 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v1, 3, v1
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
-; GFX10-NEXT:    v_lshlrev_b32_e64 v2, v1, 0xff
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX10-NEXT:    v_xor_b32_e32 v1, -1, v2
+; GFX10-NEXT:    v_lshlrev_b32_e64 v1, v1, 0xff
+; GFX10-NEXT:    v_xor_b32_e32 v1, -1, v1
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_bfe_u32 s3, s0, 0x80008
-; GFX10-NEXT:    s_lshr_b32 s1, s0, 24
+; GFX10-NEXT:    s_bfe_u32 s1, s0, 0x80008
 ; GFX10-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX10-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX10-NEXT:    s_lshl_b32 s3, s3, 8
-; GFX10-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX10-NEXT:    s_or_b32 s2, s2, s3
-; GFX10-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX10-NEXT:    s_or_b32 s0, s2, s0
-; GFX10-NEXT:    s_or_b32 s0, s0, s1
-; GFX10-NEXT:    v_and_or_b32 v0, s0, v1, v0
+; GFX10-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX10-NEXT:    s_bfe_u32 s3, s0, 0x80010
+; GFX10-NEXT:    v_or_b32_e64 v2, s2, s1
+; GFX10-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX10-NEXT:    s_lshl_b32 s1, s3, 16
+; GFX10-NEXT:    s_lshl_b32 s0, s0, 24
+; GFX10-NEXT:    v_or3_b32 v2, v2, s1, s0
 ; GFX10-NEXT:    s_mov_b32 s0, 8
+; GFX10-NEXT:    v_and_or_b32 v0, v2, v1, v0
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v1, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v2, 24, v0
 ; GFX10-NEXT:    s_mov_b32 s0, 16
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v3, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX10-NEXT:    v_and_or_b32 v4, v0, 0xff, v1
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    v_or3_b32 v2, v4, v3, v2
 ; GFX10-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: insertelement_s_v4i8_v_v:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b32 s0, s[2:3], 0x0
 ; GFX11-NEXT:    v_and_b32_e32 v1, 3, v1
 ; GFX11-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
-; GFX11-NEXT:    v_lshlrev_b32_e64 v2, v1, 0xff
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
-; GFX11-NEXT:    v_xor_b32_e32 v1, -1, v2
+; GFX11-NEXT:    v_lshlrev_b32_e64 v1, v1, 0xff
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_xor_b32_e32 v1, -1, v1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_bfe_u32 s3, s0, 0x80008
-; GFX11-NEXT:    s_lshr_b32 s1, s0, 24
+; GFX11-NEXT:    s_bfe_u32 s1, s0, 0x80008
 ; GFX11-NEXT:    s_and_b32 s2, s0, 0xff
-; GFX11-NEXT:    s_bfe_u32 s0, s0, 0x80010
-; GFX11-NEXT:    s_lshl_b32 s3, s3, 8
-; GFX11-NEXT:    s_lshl_b32 s0, s0, 16
-; GFX11-NEXT:    s_or_b32 s2, s2, s3
-; GFX11-NEXT:    s_lshl_b32 s1, s1, 24
-; GFX11-NEXT:    s_or_b32 s0, s2, s0
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_or_b32 s0, s0, s1
-; GFX11-NEXT:    v_and_or_b32 v0, s0, v1, v0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3)
+; GFX11-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX11-NEXT:    s_bfe_u32 s3, s0, 0x80010
+; GFX11-NEXT:    v_or_b32_e64 v2, s2, s1
+; GFX11-NEXT:    s_lshr_b32 s0, s0, 24
+; GFX11-NEXT:    s_lshl_b32 s1, s3, 16
+; GFX11-NEXT:    s_lshl_b32 s0, s0, 24
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_or3_b32 v2, v2, s1, s0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-NEXT:    v_and_or_b32 v0, v2, v1, v0
 ; GFX11-NEXT:    v_bfe_u32 v1, v0, 8, 8
 ; GFX11-NEXT:    v_bfe_u32 v2, v0, 16, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
-; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_and_or_b32 v4, v0, 0xff, v1
 ; GFX11-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
 ; GFX11-NEXT:    v_or3_b32 v2, v4, v2, v3
 ; GFX11-NEXT:    global_store_b32 v[0:1], v2, off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
   %vec = load <4 x i8>, <4 x i8> addrspace(4)* %ptr
   %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
   store <4 x i8> %insert, <4 x i8> addrspace(1)* null
   ret void
; GFX11-NEXT:    s_endpgm
   store <8 x i8> %insert, <8 x i8> addrspace(1)* null
   ret void
 }
 
 define amdgpu_ps void @insertelement_s_v8i8_v_s(<8 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
 ; GFX9-LABEL: insertelement_s_v8i8_v_s:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX9-NEXT:    s_mov_b32 s2, 8
 ; GFX9-NEXT:    s_mov_b32 s3, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0xff
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s8, s0, 0x80008
 ; GFX9-NEXT:    s_lshr_b32 s5, s0, 24
 ; GFX9-NEXT:    s_and_b32 s7, s0, 0xff
 ; GFX9-NEXT:    s_lshl_b32 s8, s8, 8

 ; GFX9-NEXT:    s_or_b32 s1, s5, s1
 ; GFX9-NEXT:    s_lshl_b32 s5, s6, 24
 ; GFX9-NEXT:    s_or_b32 s1, s1, s5
 ; GFX9-NEXT:    s_lshr_b32 s5, s4, 2
 ; GFX9-NEXT:    s_cmp_eq_u32 s5, 1
 ; GFX9-NEXT:    s_cselect_b32 s6, s1, s0
 ; GFX9-NEXT:    s_and_b32 s4, s4, 3
 ; GFX9-NEXT:    s_lshl_b32 s4, s4, 3
-; GFX9-NEXT:    s_lshl_b32 s7, 0xff, s4
-; GFX9-NEXT:    s_andn2_b32 s6, s6, s7
-; GFX9-NEXT:    v_mov_b32_e32 v1, s6
-; GFX9-NEXT:    v_lshl_or_b32 v3, v0, s4, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX9-NEXT:    s_not_b32 s4, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NEXT:    v_and_or_b32 v3, s6, v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
 ; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, s2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1

 ; GFX8-NEXT:    s_or_b32 s1, s1, s2
 ; GFX8-NEXT:    s_lshr_b32 s2, s4, 2
 ; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
 ; GFX8-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX8-NEXT:    s_and_b32 s4, s4, 3
 ; GFX8-NEXT:    s_lshl_b32 s4, s4, 3
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX8-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX8-NEXT:    s_andn2_b32 s3, s3, s4
-; GFX8-NEXT:    v_or_b32_e32 v2, s3, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX8-NEXT:    v_or_b32_e32 v2, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
 ; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 24, v0

 ; GFX7-NEXT:    s_or_b32 s1, s1, s2
 ; GFX7-NEXT:    s_lshr_b32 s2, s4, 2
 ; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
 ; GFX7-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX7-NEXT:    s_and_b32 s4, s4, 3
 ; GFX7-NEXT:    s_lshl_b32 s4, s4, 3
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
 ; GFX7-NEXT:    s_lshl_b32 s4, 0xff, s4
-; GFX7-NEXT:    s_andn2_b32 s3, s3, s4
-; GFX7-NEXT:    v_or_b32_e32 v2, s3, v0
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
 ; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
 ; GFX7-NEXT:    v_bfe_u32 v5, v0, 8, 8
 ; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
 ; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 24, v0

 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v8i8_v_s:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
 ; GFX10-NEXT:    s_lshr_b32 s2, s4, 2
-; GFX10-NEXT:    v_and_b32_e32 v2, 0xff, v0
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_bfe_u32 s7, s0, 0x80008
 ; GFX10-NEXT:    s_bfe_u32 s9, s1, 0x80008
 ; GFX10-NEXT:    s_lshr_b32 s3, s0, 24
 ; GFX10-NEXT:    s_lshr_b32 s5, s1, 24
 ; GFX10-NEXT:    s_and_b32 s6, s0, 0xff
 ; GFX10-NEXT:    s_bfe_u32 s0, s0, 0x80010
 ; GFX10-NEXT:    s_and_b32 s8, s1, 0xff
 ; GFX10-NEXT:    s_bfe_u32 s1, s1, 0x80010
 ; GFX10-NEXT:    s_lshl_b32 s7, s7, 8
 ; GFX10-NEXT:    s_lshl_b32 s9, s9, 8
 ; GFX10-NEXT:    s_lshl_b32 s0, s0, 16
 ; GFX10-NEXT:    s_lshl_b32 s1, s1, 16
 ; GFX10-NEXT:    s_or_b32 s6, s6, s7
 ; GFX10-NEXT:    s_or_b32 s7, s8, s9
 ; GFX10-NEXT:    s_lshl_b32 s3, s3, 24
 ; GFX10-NEXT:    s_lshl_b32 s5, s5, 24
 ; GFX10-NEXT:    s_or_b32 s0, s6, s0
 ; GFX10-NEXT:    s_or_b32 s1, s7, s1
 ; GFX10-NEXT:    s_or_b32 s0, s0, s3
 ; GFX10-NEXT:    s_or_b32 s1, s1, s5
 ; GFX10-NEXT:    s_cmp_eq_u32 s2, 1
-; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX10-NEXT:    s_and_b32 s4, s4, 3
-; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX10-NEXT:    v_lshlrev_b32_sdwa v2, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX10-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    s_not_b32 s4, s4
+; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    v_and_or_b32 v2, s3, s4, v2
 ; GFX10-NEXT:    s_mov_b32 s0, 8
-; GFX10-NEXT:    s_lshl_b32 s5, 0xff, s4
 ; GFX10-NEXT:    s_mov_b32 s1, 16
-; GFX10-NEXT:    s_andn2_b32 s3, s3, s5
-; GFX10-NEXT:    v_lshl_or_b32 v2, v2, s4, s3
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc_lo
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 1
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v3, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; GFX10-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc_lo
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v2, 24, v0
 ; GFX10-NEXT:    v_lshlrev_b32_sdwa v6, s1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2
 ; GFX10-NEXT:    v_and_or_b32 v3, 0xff, v0, v3
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v4, 24, v1

 ; GFX11-NEXT:    s_or_b32 s7, s8, s9
 ; GFX11-NEXT:    s_lshl_b32 s3, s3, 24
 ; GFX11-NEXT:    s_lshl_b32 s5, s5, 24
 ; GFX11-NEXT:    s_or_b32 s0, s6, s0
 ; GFX11-NEXT:    s_or_b32 s1, s7, s1
 ; GFX11-NEXT:    s_or_b32 s0, s0, s3
 ; GFX11-NEXT:    s_or_b32 s1, s1, s5
 ; GFX11-NEXT:    s_cmp_eq_u32 s2, 1
-; GFX11-NEXT:    v_and_b32_e32 v2, 0xff, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX11-NEXT:    s_cselect_b32 s3, s1, s0
 ; GFX11-NEXT:    s_and_b32 s4, s4, 3
-; GFX11-NEXT:    v_mov_b32_e32 v0, s0
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_lshl_b32 s4, s4, 3
-; GFX11-NEXT:    v_mov_b32_e32 v1, s1
-; GFX11-NEXT:    s_lshl_b32 s5, 0xff, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 s3, s3, s5
-; GFX11-NEXT:    v_lshl_or_b32 v2, v2, s4, s3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_lshlrev_b32_e32 v2, s4, v0
+; GFX11-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_not_b32 s4, s4
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_and_or_b32 v2, s3, s4, v2
+; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc_lo
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s2, 1
 ; GFX11-NEXT:    v_bfe_u32 v3, v0, 8, 8
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX11-NEXT:    v_bfe_u32 v5, v1, 8, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 24, v0
 ; GFX11-NEXT:    v_bfe_u32 v4, v0, 16, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
 ; GFX11-NEXT:    v_bfe_u32 v7, v1, 16, 8
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
; GFX11-NEXT:    s_endpgm
   store <16 x i8> %insert, <16 x i8> addrspace(1)* null
   ret void
 }
 
 define amdgpu_ps void @insertelement_s_v16i8_v_s(<16 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
 ; GFX9-LABEL: insertelement_s_v16i8_v_s:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX9-NEXT:    s_mov_b32 s5, 8
 ; GFX9-NEXT:    s_mov_b32 s6, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0xff
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s12, s0, 0x80008
 ; GFX9-NEXT:    s_lshr_b32 s7, s0, 24
 ; GFX9-NEXT:    s_and_b32 s11, s0, 0xff
 ; GFX9-NEXT:    s_lshl_b32 s12, s12, 8

 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 1
 ; GFX9-NEXT:    s_cselect_b32 s8, s1, s0
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
 ; GFX9-NEXT:    s_cselect_b32 s8, s2, s8
 ; GFX9-NEXT:    s_cmp_eq_u32 s7, 3
 ; GFX9-NEXT:    s_cselect_b32 s8, s3, s8
 ; GFX9-NEXT:    s_and_b32 s4, s4, 3
 ; GFX9-NEXT:    s_lshl_b32 s4, s4, 3
-; GFX9-NEXT:    s_lshl_b32 s9, 0xff, s4
-; GFX9-NEXT:    s_andn2_b32 s8, s8, s9
-; GFX9-NEXT:    v_mov_b32_e32 v1, s8
-; GFX9-NEXT:    v_lshl_or_b32 v5, v0, s4, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX9-NEXT:    s_not_b32 s4, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NEXT:    v_and_or_b32 v5, s8, v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s7, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s7, 1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
 ; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s7, 2

 ; GFX8-NEXT:    s_cmp_eq_u32 s5, 2
 ; GFX8-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX8-NEXT:    s_cmp_eq_u32 s5, 3
 ; GFX8-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX8-NEXT:    s_and_b32 s4, s4, 3
 ; GFX8-NEXT:    s_lshl_b32 s4, s4, 3
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX8-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
 ; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX8-NEXT:    s_andn2_b32 s4, s6, s4
-; GFX8-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX8-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2

 ; GFX7-NEXT:    s_cmp_eq_u32 s5, 2
 ; GFX7-NEXT:    s_cselect_b32 s6, s2, s6
 ; GFX7-NEXT:    s_cmp_eq_u32 s5, 3
 ; GFX7-NEXT:    s_cselect_b32 s6, s3, s6
 ; GFX7-NEXT:    s_and_b32 s4, s4, 3
 ; GFX7-NEXT:    s_lshl_b32 s4, s4, 3
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
 ; GFX7-NEXT:    s_lshl_b32 s4, 0xff, s4
-; GFX7-NEXT:    s_andn2_b32 s4, s6, s4
-; GFX7-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2

 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: insertelement_s_v16i8_v_s:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v10, 8
 ; GFX10-NEXT:    v_mov_b32_e32 v12, 16
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_bfe_u32 s10, s0, 0x80008
 ; GFX10-NEXT:    s_bfe_u32 s12, s1, 0x80008
 ; GFX10-NEXT:    s_lshr_b32 s6, s1, 24
 ; GFX10-NEXT:    s_and_b32 s9, s0, 0xff
 ; GFX10-NEXT:    s_and_b32 s11, s1, 0xff

 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s6, 0
 ; GFX10-NEXT:    s_cselect_b32 s5, s1, s0
 ; GFX10-NEXT:    s_cmp_eq_u32 s6, 2
 ; GFX10-NEXT:    s_cselect_b32 s5, s2, s5
 ; GFX10-NEXT:    s_cmp_eq_u32 s6, 3
 ; GFX10-NEXT:    s_cselect_b32 s5, s3, s5
 ; GFX10-NEXT:    s_and_b32 s4, s4, 3
 ; GFX10-NEXT:    s_lshl_b32 s4, s4, 3
-; GFX10-NEXT:    s_lshl_b32 s7, 0xff, s4
-; GFX10-NEXT:    s_andn2_b32 s5, s5, s7
-; GFX10-NEXT:    v_lshl_or_b32 v4, v0, s4, s5
+; GFX10-NEXT:    v_lshlrev_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX10-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX10-NEXT:    s_not_b32 s4, s4
+; GFX10-NEXT:    v_and_or_b32 v4, s5, s4, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-NEXT:    s_mov_b32 s0, 8
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc_lo
 ; GFX10-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s6, 1
 ; GFX10-NEXT:    s_mov_b32 s1, 16

 ; GFX11-NEXT:    s_cmp_eq_u32 s6, 2
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s6, 0
 ; GFX11-NEXT:    s_cselect_b32 s5, s2, s5
 ; GFX11-NEXT:    s_cmp_eq_u32 s6, 3
 ; GFX11-NEXT:    s_cselect_b32 s5, s3, s5
 ; GFX11-NEXT:    s_and_b32 s4, s4, 3
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_lshl_b32 s4, s4, 3
-; GFX11-NEXT:    s_lshl_b32 s7, 0xff, s4
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_not1_b32 s5, s5, s7
-; GFX11-NEXT:    v_lshl_or_b32 v4, v0, s4, s5
-; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3)
-; GFX11-NEXT:    v_dual_mov_b32 v1, s1 :: v_dual_cndmask_b32 v0, v0, v4
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
+; GFX11-NEXT:    s_lshl_b32 s4, 0xff, s4
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_not_b32 s4, s4
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    v_and_or_b32 v4, s5, s4, v0
+; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX11-NEXT:    v_dual_mov_b32 v3, s3 :: v_dual_mov_b32 v2, s2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc_lo
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s6, 1
-; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    v_bfe_u32 v5, v0, 8, 8
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc_lo
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s6, 2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_bfe_u32 v7, v1, 8, 8
 ; GFX11-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
 ; GFX11-NEXT:    v_cmp_eq_u32_e64 vcc_lo, s6, 3
-; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
 ; GFX11-NEXT:    v_bfe_u32 v8, v0, 16, 8
-; GFX11-NEXT:    v_dual_cndmask_b32 v3, v3, v4 :: v_dual_lshlrev_b32 v6, 24, v6
+; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
+; GFX11-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc_lo
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
 ; GFX11-NEXT:    v_and_or_b32 v0, 0xff, v0, v5
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v5, 8, v7
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v7, 16, v8
 ; GFX11-NEXT:    v_bfe_u32 v8, v1, 16, 8
 ; GFX11-NEXT:    v_bfe_u32 v11, v2, 8, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 24, v2
 ; GFX11-NEXT:    v_and_or_b32 v1, 0xff, v1, v5
 ; GFX11-NEXT:    v_bfe_u32 v5, v3, 8, 8
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v10, 24, v3
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v11, 8, v11
 ; GFX11-NEXT:    v_bfe_u32 v12, v2, 16, 8
 ; GFX11-NEXT:    v_bfe_u32 v13, v3, 16, 8
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v8, 16, v8
+; GFX11-NEXT:    v_lshlrev_b32_e32 v6, 24, v6
 ; GFX11-NEXT:    v_and_or_b32 v2, 0xff, v2, v11
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v11, 16, v12
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v9, 24, v9
 ; GFX11-NEXT:    v_and_or_b32 v3, 0xff, v3, v5
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v12, 16, v13
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v10, 24, v10
 ; GFX11-NEXT:    v_or3_b32 v0, v0, v7, v4
 ; GFX11-NEXT:    v_mov_b32_e32 v4, 0

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankcombiner-clamp-minmax-const.mir

bb.1 :
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 2.000000e+00
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[FMUL:%[0-9]+]]:vgpr(s32) = G_FMUL [[COPY]], [[COPY1]]
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 0.000000e+00
     ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:vgpr(s32) = G_FCANONICALIZE [[FMUL]]
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 1.000000e+00
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[C2]](s32)
     ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_FMED3 [[FCANONICALIZE]], [[COPY2]], [[COPY3]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_FMED3_]](s32)
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_FCONSTANT float 2.000000e+00
     %10:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_FMUL %0, %10
     %4:sgpr(s32) = G_FCONSTANT float 0.000000e+00

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankcombiner-fmed3-minmax-const.mir

body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_max_ValK0_K1_f32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 2.000000e+00
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 4.000000e+00
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = nnan G_AMDGPU_FMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_FMED3_]](s32)
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_FCONSTANT float 2.000000e+00
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = nnan G_FMAXNUM_IEEE %0, %7
     %4:sgpr(s32) = G_FCONSTANT float 4.000000e+00
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_max_K0Val_K1_f32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 2.000000e+00
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 4.000000e+00
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = nnan G_AMDGPU_FMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_FMED3_]](s32)
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_FCONSTANT float 2.000000e+00
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = nnan G_FMAXNUM %7, %0
     %4:sgpr(s32) = G_FCONSTANT float 4.000000e+00
bb.1:
 
     ; CHECK-LABEL: name: test_min_K1max_ValK0_f16
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4000
     ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:vgpr(s16) = G_FCANONICALIZE [[TRUNC]]
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4400
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[C1]](s16)
     ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s16) = G_AMDGPU_FMED3 [[FCANONICALIZE]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:vgpr(s32) = G_ANYEXT [[AMDGPU_FMED3_]](s16)
     ; CHECK-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
     %2:vgpr(s32) = COPY $vgpr0
     %0:vgpr(s16) = G_TRUNC %2(s32)
     %3:sgpr(s16) = G_FCONSTANT half 0xH4000
     %9:vgpr(s16) = G_FCANONICALIZE %0
bb.1:
     liveins: $vgpr0, $sgpr30_sgpr31
 
     ; CHECK-LABEL: name: test_min_K1max_K0Val_f16
     ; CHECK: liveins: $vgpr0, $sgpr30_sgpr31
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4000
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4400
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[C1]](s16)
     ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s16) = nnan G_AMDGPU_FMED3 [[TRUNC]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:vgpr(s32) = G_ANYEXT [[AMDGPU_FMED3_]](s16)
     ; CHECK-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
     %2:vgpr(s32) = COPY $vgpr0
     %0:vgpr(s16) = G_TRUNC %2(s32)
     %3:sgpr(s16) = G_FCONSTANT half 0xH4000
     %9:vgpr(s16) = COPY %3(s16)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_min_ValK1_K0_f32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 4.000000e+00
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 2.000000e+00
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = nnan G_AMDGPU_FMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = nnan G_AMDGPU_FMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_FMED3_]](s32)
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_FCONSTANT float 4.000000e+00
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = nnan G_FMINNUM_IEEE %0, %7
     %4:sgpr(s32) = G_FCONSTANT float 2.000000e+00
     %8:vgpr(s32) = COPY %4(s32)
     %5:vgpr(s32) = nnan G_FMAXNUM_IEEE %3, %8
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_min_K1Val_K0_f32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 4.000000e+00
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_FCONSTANT float 2.000000e+00
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = nnan G_AMDGPU_FMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s32) = nnan G_AMDGPU_FMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_FMED3_]](s32)
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_FCONSTANT float 4.000000e+00
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = nnan G_FMINNUM %7, %0
     %4:sgpr(s32) = G_FCONSTANT float 2.000000e+00
     %8:vgpr(s32) = COPY %4(s32)
     %5:vgpr(s32) = nnan G_FMAXNUM %3, %8
bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_K0min_ValK1_f16
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4400
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4000
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[C1]](s16)
-    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s16) = nnan G_AMDGPU_FMED3 [[TRUNC]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C1]](s16)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
+    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s16) = nnan G_AMDGPU_FMED3 [[TRUNC]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:vgpr(s32) = G_ANYEXT [[AMDGPU_FMED3_]](s16)
     ; CHECK-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
     %2:vgpr(s32) = COPY $vgpr0
     %0:vgpr(s16) = G_TRUNC %2(s32)
     %3:sgpr(s16) = G_FCONSTANT half 0xH4400
     %9:vgpr(s16) = COPY %3(s16)
     %4:vgpr(s16) = nnan G_FMINNUM_IEEE %0, %9
     %5:sgpr(s16) = G_FCONSTANT half 0xH4000
bb.1:
     liveins: $vgpr0, $sgpr30_sgpr31
 
     ; CHECK-LABEL: name: test_max_K0min_K1Val_f16
     ; CHECK: liveins: $vgpr0, $sgpr30_sgpr31
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4400
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s16) = G_FCONSTANT half 0xH4000
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[C1]](s16)
-    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s16) = nnan G_AMDGPU_FMED3 [[TRUNC]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s16) = COPY [[C1]](s16)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s16) = COPY [[C]](s16)
+    ; CHECK-NEXT: [[AMDGPU_FMED3_:%[0-9]+]]:vgpr(s16) = nnan G_AMDGPU_FMED3 [[TRUNC]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:vgpr(s32) = G_ANYEXT [[AMDGPU_FMED3_]](s16)
     ; CHECK-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
     %2:vgpr(s32) = COPY $vgpr0
     %0:vgpr(s16) = G_TRUNC %2(s32)
     %3:sgpr(s16) = G_FCONSTANT half 0xH4400
     %9:vgpr(s16) = COPY %3(s16)
     %4:vgpr(s16) = nnan G_FMINNUM %9, %0
     %5:sgpr(s16) = G_FCONSTANT half 0xH4000

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankcombiner-smed3.mir

body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_max_ValK0_K1_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 -12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMAX %0, %7
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: min_max_ValK0_K1_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 -12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMAX %7, %0
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_K1max_ValK0__i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 -12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMAX %0, %7
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_K1max_K0Val__i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 -12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMAX %7, %0
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_min_ValK1_K0_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMIN %0, %7
     %4:sgpr(s32) = G_CONSTANT i32 -12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_min_K1Val_K0_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMIN %7, %0
     %4:sgpr(s32) = G_CONSTANT i32 -12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_K0min_ValK1__i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMIN %0, %7
     %4:sgpr(s32) = G_CONSTANT i32 -12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_K0min_K1Val__i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMIN %7, %0
     %4:sgpr(s32) = G_CONSTANT i32 -12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_non_inline_constant_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 -12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_SMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_SMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_SMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 -12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_SMAX %0, %7
     %4:sgpr(s32) = G_CONSTANT i32 65
     %8:vgpr(s32) = COPY %4(s32)
     %5:vgpr(s32) = G_SMIN %3, %8
     $vgpr0 = COPY %5(s32)
     S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
 ...

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankcombiner-umed3.mir

body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_max_ValK0_K1_u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMAX %0, %7
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: min_max_ValK0_K1_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMAX %7, %0
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_K1max_ValK0__u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMAX %0, %7
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_min_K1max_K0Val__u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMAX %7, %0
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_min_ValK1_K0_u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMIN %0, %7
     %4:sgpr(s32) = G_CONSTANT i32 12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_min_K1Val_K0_u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMIN %7, %0
     %4:sgpr(s32) = G_CONSTANT i32 12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_K0min_ValK1__u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMIN %0, %7
     %4:sgpr(s32) = G_CONSTANT i32 12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_max_K0min_K1Val__u32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 17
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
-    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY2]], [[COPY1]]
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 17
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMIN %7, %0
     %4:sgpr(s32) = G_CONSTANT i32 12
     %8:vgpr(s32) = COPY %4(s32)
body: |
   bb.1:
     liveins: $vgpr0
 
     ; CHECK-LABEL: name: test_non_inline_constant_i32
     ; CHECK: liveins: $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 12
-    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[C1]](s32)
     ; CHECK-NEXT: [[AMDGPU_UMED3_:%[0-9]+]]:vgpr(s32) = G_AMDGPU_UMED3 [[COPY]], [[COPY1]], [[COPY2]]
     ; CHECK-NEXT: $vgpr0 = COPY [[AMDGPU_UMED3_]](s32)
     ; CHECK-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
     %0:vgpr(s32) = COPY $vgpr0
     %2:sgpr(s32) = G_CONSTANT i32 12
     %7:vgpr(s32) = COPY %2(s32)
     %3:vgpr(s32) = G_UMAX %0, %7
     %4:sgpr(s32) = G_CONSTANT i32 65
     %8:vgpr(s32) = COPY %4(s32)
     %5:vgpr(s32) = G_UMIN %3, %8
     $vgpr0 = COPY %5(s32)
     S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0
 ...

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-and.mir

bb.0:
     ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[COPY]], [[COPY2]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[COPY1]], [[COPY3]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND]](s32), [[AND1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $vgpr1
     %2:_(s32) = COPY $vgpr2
     %3:_(s32) = COPY $vgpr3
     %4:_(s64) = G_MERGE_VALUES %0, %1
     %5:_(s64) = G_MERGE_VALUES %2, %3
bb.0:
     ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2, $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[COPY3]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[COPY2]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND]](s32), [[AND1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s32) = COPY $sgpr2
     %2:_(s32) = COPY $vgpr0
     %3:_(s64) = G_MERGE_VALUES %1, %2
     %4:_(s64) = G_AND %0, %3
     S_NOP 0, implicit %4
bb.0:
     ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2, $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[COPY2]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[COPY3]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND]](s32), [[AND1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s32) = COPY $sgpr2
     %2:_(s32) = COPY $vgpr0
     %3:_(s64) = G_MERGE_VALUES %2, %1
     %4:_(s64) = G_AND %0, %3
     S_NOP 0, implicit %4
bb.0:
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY5]](s32), [[COPY3]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[COPY4]], [[COPY5]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[COPY2]], [[COPY3]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND]](s32), [[AND1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = COPY $vgpr0
     %3:_(s32) = COPY $vgpr1
     %4:_(s64) = G_MERGE_VALUES %0, %2
     %5:_(s64) = G_MERGE_VALUES %1, %3
bb.0:
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY5]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[COPY4]], [[COPY3]]
+    ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[COPY2]], [[COPY5]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND]](s32), [[AND1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = COPY $vgpr0
     %3:_(s32) = COPY $vgpr1
     %4:_(s64) = G_MERGE_VALUES %0, %2
     %5:_(s64) = G_MERGE_VALUES %3, %1
bb.0:
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr2_sgpr3
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
     ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY2]](s64)
     ; CHECK-NEXT: [[AND:%[0-9]+]]:vgpr(s32) = G_AND [[UV]], [[UV2]]
     ; CHECK-NEXT: [[AND1:%[0-9]+]]:vgpr(s32) = G_AND [[UV1]], [[UV3]]
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND]](s32), [[AND1]](s32)
     ; CHECK-NEXT: [[UV4:%[0-9]+]]:sgpr(s32), [[UV5:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
-    ; CHECK-NEXT: [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[AND2:%[0-9]+]]:vgpr(s32) = G_AND [[UV4]], [[UV6]]
-    ; CHECK-NEXT: [[AND3:%[0-9]+]]:vgpr(s32) = G_AND [[UV5]], [[UV7]]
+    ; CHECK-NEXT: [[AND2:%[0-9]+]]:vgpr(s32) = G_AND [[UV4]], [[AND]]
+    ; CHECK-NEXT: [[AND3:%[0-9]+]]:vgpr(s32) = G_AND [[UV5]], [[AND1]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[AND2]](s32), [[AND3]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s64) = COPY $sgpr2_sgpr3
     %2:_(s64) = COPY $vgpr0_vgpr1
     %3:_(s64) = G_AND %0, %2
     %4:_(s64) = G_AND %1, %3
     S_NOP 0, implicit %4

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-or.mir

bb.0:
     ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[COPY]], [[COPY2]]
+    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[COPY1]], [[COPY3]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $vgpr1
     %2:_(s32) = COPY $vgpr2
     %3:_(s32) = COPY $vgpr3
     %4:_(s64) = G_MERGE_VALUES %0, %1
     %5:_(s64) = G_MERGE_VALUES %2, %3
bb.0:
     ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2, $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[COPY3]]
+    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[COPY2]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s32) = COPY $sgpr2
     %2:_(s32) = COPY $vgpr0
     %3:_(s64) = G_MERGE_VALUES %1, %2
     %4:_(s64) = G_OR %0, %3
     S_NOP 0, implicit %4
bb.0:
     ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2, $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[COPY2]]
+    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[COPY3]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s32) = COPY $sgpr2
     %2:_(s32) = COPY $vgpr0
     %3:_(s64) = G_MERGE_VALUES %2, %1
     %4:_(s64) = G_OR %0, %3
     S_NOP 0, implicit %4
bb.0:
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY5]](s32), [[COPY3]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[COPY4]], [[COPY5]]
+    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[COPY2]], [[COPY3]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = COPY $vgpr0
     %3:_(s32) = COPY $vgpr1
     %4:_(s64) = G_MERGE_VALUES %0, %2
     %5:_(s64) = G_MERGE_VALUES %1, %3
bb.0:
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY5]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[COPY4]], [[COPY3]]
+    ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[COPY2]], [[COPY5]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = COPY $vgpr0
     %3:_(s32) = COPY $vgpr1
     %4:_(s64) = G_MERGE_VALUES %0, %2
     %5:_(s64) = G_MERGE_VALUES %3, %1
bb.0:
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr2_sgpr3
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
     ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY2]](s64)
     ; CHECK-NEXT: [[OR:%[0-9]+]]:vgpr(s32) = G_OR [[UV]], [[UV2]]
     ; CHECK-NEXT: [[OR1:%[0-9]+]]:vgpr(s32) = G_OR [[UV1]], [[UV3]]
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR]](s32), [[OR1]](s32)
     ; CHECK-NEXT: [[UV4:%[0-9]+]]:sgpr(s32), [[UV5:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
-    ; CHECK-NEXT: [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[OR2:%[0-9]+]]:vgpr(s32) = G_OR [[UV4]], [[UV6]]
-    ; CHECK-NEXT: [[OR3:%[0-9]+]]:vgpr(s32) = G_OR [[UV5]], [[UV7]]
+    ; CHECK-NEXT: [[OR2:%[0-9]+]]:vgpr(s32) = G_OR [[UV4]], [[OR]]
+    ; CHECK-NEXT: [[OR3:%[0-9]+]]:vgpr(s32) = G_OR [[UV5]], [[OR1]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[OR2]](s32), [[OR3]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s64) = COPY $sgpr2_sgpr3
     %2:_(s64) = COPY $vgpr0_vgpr1
     %3:_(s64) = G_OR %0, %2
     %4:_(s64) = G_OR %1, %3
     S_NOP 0, implicit %4

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-xor.mir

bb.0:
     ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[COPY]], [[COPY2]]
+    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[COPY1]], [[COPY3]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR]](s32), [[XOR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $vgpr1
     %2:_(s32) = COPY $vgpr2
     %3:_(s32) = COPY $vgpr3
     %4:_(s64) = G_MERGE_VALUES %0, %1
     %5:_(s64) = G_MERGE_VALUES %2, %3
bb.0:
     ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2, $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[COPY3]]
+    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[COPY2]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR]](s32), [[XOR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s32) = COPY $sgpr2
     %2:_(s32) = COPY $vgpr0
     %3:_(s64) = G_MERGE_VALUES %1, %2
     %4:_(s64) = G_XOR %0, %3
     S_NOP 0, implicit %4
bb.0:
     ; CHECK: liveins: $sgpr0_sgpr1, $sgpr2, $vgpr0
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s64) = COPY $sgpr0_sgpr1
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY2]](s32), [[COPY3]](s32)
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[COPY2]]
+    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[COPY3]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR]](s32), [[XOR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s32) = COPY $sgpr2
     %2:_(s32) = COPY $vgpr0
     %3:_(s64) = G_MERGE_VALUES %2, %1
     %4:_(s64) = G_XOR %0, %3
     S_NOP 0, implicit %4
bb.0:
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY5]](s32), [[COPY3]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[COPY4]], [[COPY5]]
+    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[COPY2]], [[COPY3]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR]](s32), [[XOR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = COPY $vgpr0
     %3:_(s32) = COPY $vgpr1
     %4:_(s64) = G_MERGE_VALUES %0, %2
     %5:_(s64) = G_MERGE_VALUES %1, %3
bb.0:
     ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
     ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
     ; CHECK-NEXT: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY2]](s32)
     ; CHECK-NEXT: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[COPY3]](s32), [[COPY5]](s32)
-    ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr(s32), [[UV1:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV1]](s64)
-    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[UV2]]
-    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[UV3]]
+    ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[COPY4]], [[COPY3]]
+    ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[COPY2]], [[COPY5]]
     ; CHECK-NEXT: [[MV2:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR]](s32), [[XOR1]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV2]](s64)
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = COPY $vgpr0
     %3:_(s32) = COPY $vgpr1
     %4:_(s64) = G_MERGE_VALUES %0, %2
     %5:_(s64) = G_MERGE_VALUES %3, %1
bb.0:
     ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(s64) = COPY $sgpr2_sgpr3
     ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1
     ; CHECK-NEXT: [[UV:%[0-9]+]]:sgpr(s32), [[UV1:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY]](s64)
     ; CHECK-NEXT: [[UV2:%[0-9]+]]:vgpr(s32), [[UV3:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[COPY2]](s64)
     ; CHECK-NEXT: [[XOR:%[0-9]+]]:vgpr(s32) = G_XOR [[UV]], [[UV2]]
     ; CHECK-NEXT: [[XOR1:%[0-9]+]]:vgpr(s32) = G_XOR [[UV1]], [[UV3]]
     ; CHECK-NEXT: [[MV:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR]](s32), [[XOR1]](s32)
     ; CHECK-NEXT: [[UV4:%[0-9]+]]:sgpr(s32), [[UV5:%[0-9]+]]:sgpr(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
-    ; CHECK-NEXT: [[UV6:%[0-9]+]]:vgpr(s32), [[UV7:%[0-9]+]]:vgpr(s32) = G_UNMERGE_VALUES [[MV]](s64)
-    ; CHECK-NEXT: [[XOR2:%[0-9]+]]:vgpr(s32) = G_XOR [[UV4]], [[UV6]]
-    ; CHECK-NEXT: [[XOR3:%[0-9]+]]:vgpr(s32) = G_XOR [[UV5]], [[UV7]]
+    ; CHECK-NEXT: [[XOR2:%[0-9]+]]:vgpr(s32) = G_XOR [[UV4]], [[XOR]]
+    ; CHECK-NEXT: [[XOR3:%[0-9]+]]:vgpr(s32) = G_XOR [[UV5]], [[XOR1]]
     ; CHECK-NEXT: [[MV1:%[0-9]+]]:vgpr(s64) = G_MERGE_VALUES [[XOR2]](s32), [[XOR3]](s32)
     ; CHECK-NEXT: S_NOP 0, implicit [[MV1]](s64)
     %0:_(s64) = COPY $sgpr0_sgpr1
     %1:_(s64) = COPY $sgpr2_sgpr3
     %2:_(s64) = COPY $vgpr0_vgpr1
     %3:_(s64) = G_XOR %0, %2
     %4:_(s64) = G_XOR %1, %3
     S_NOP 0, implicit %4

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

 ; GCN-NEXT:    v_xor_b32_e32 v0, -1, v0
 ; GCN-NEXT:    v_xor_b32_e32 v1, -1, v1
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: vector_xnor_i64_one_use:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    v_xor_b32_e32 v0, v0, v2
-; GFX10-NEXT:    v_xor_b32_e32 v1, v1, v3
-; GFX10-NEXT:    v_xor_b32_e32 v0, -1, v0
-; GFX10-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX10-NEXT:    v_xor3_b32 v0, v0, v2, -1
+; GFX10-NEXT:    v_xor3_b32 v1, v1, v3, -1
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %xor = xor i64 %a, %b
   %r = xor i64 %xor, -1
   ret i64 %r
 }
 
 define amdgpu_ps float @xnor_s_v_i32_one_use(i32 inreg %s, i32 %v) {

 ; GFX906-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GFX906-NEXT:    v_xor_b32_e32 v0, -1, v0
 ; GFX906-NEXT:    v_xor_b32_e32 v1, -1, v1
 ; GFX906-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: xnor_i64_s_v_one_use:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    v_lshlrev_b64 v[0:1], 29, v[0:1]
-; GFX10-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GFX10-NEXT:    v_xor_b32_e32 v1, s1, v1
-; GFX10-NEXT:    v_xor_b32_e32 v0, -1, v0
-; GFX10-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX10-NEXT:    v_xor3_b32 v0, s0, v0, -1
+; GFX10-NEXT:    v_xor3_b32 v1, s1, v1, -1
 ; GFX10-NEXT:    ; return to shader part epilog
 entry:
   %b = shl i64 %b64, 29
   %xor = xor i64 %a, %b
   %r0.val = xor i64 %xor, -1
   %cast = bitcast i64 %r0.val to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX906-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GFX906-NEXT:    v_xor_b32_e32 v0, -1, v0
 ; GFX906-NEXT:    v_xor_b32_e32 v1, -1, v1
 ; GFX906-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: xnor_i64_v_s_one_use:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    v_lshlrev_b64 v[0:1], 29, v[0:1]
-; GFX10-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GFX10-NEXT:    v_xor_b32_e32 v1, s1, v1
-; GFX10-NEXT:    v_xor_b32_e32 v0, -1, v0
-; GFX10-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX10-NEXT:    v_xor3_b32 v0, v0, s0, -1
+; GFX10-NEXT:    v_xor3_b32 v1, v1, s1, -1
 ; GFX10-NEXT:    ; return to shader part epilog
   %b = shl i64 %b64, 29
   %xor = xor i64 %b, %a
   %r0.val = xor i64 %xor, -1
   %cast = bitcast i64 %r0.val to <2 x float>
   ret <2 x float> %cast
 }
 

llvm/test/CodeGen/AMDGPU/bfi_int.ll

entry:
   %1 = and i32 %x, %0
   %2 = xor i32 %z, %1
   ret i32 %2
 }
 
 define amdgpu_ps float @v_s_s_bfi_sha256_ch(i32 %x, i32 inreg %y, i32 inreg %z) {
 ; GFX7-LABEL: v_s_s_bfi_sha256_ch:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    v_mov_b32_e32 v1, s1
-; GFX7-NEXT:    v_bfi_b32 v0, v0, s0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    v_bfi_b32 v0, v0, v1, s1
 ; GFX7-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-LABEL: v_s_s_bfi_sha256_ch:
 ; GFX8:       ; %bb.0: ; %entry
-; GFX8-NEXT:    v_mov_b32_e32 v1, s1
-; GFX8-NEXT:    v_bfi_b32 v0, v0, s0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v1, s0
+; GFX8-NEXT:    v_bfi_b32 v0, v0, v1, s1
 ; GFX8-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: v_s_s_bfi_sha256_ch:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    v_bfi_b32 v0, v0, s0, s1
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: v_s_s_bfi_sha256_ch:
 ; GFX8-GISEL:       ; %bb.0: ; %entry
-; GFX8-GISEL-NEXT:    s_xor_b32 s0, s0, s1
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, s1, v0
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v0, s0, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_s_s_bfi_sha256_ch:
 ; GFX10-GISEL:       ; %bb.0: ; %entry
-; GFX10-GISEL-NEXT:    s_xor_b32 s0, s0, s1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, s1, v0
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v0, s0, s1
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
 entry:
   %xor0 = xor i32 %y, %z
   %and = and i32 %x, %xor0
   %xor1 = xor i32 %z, %and
   %cast = bitcast i32 %xor1 to float
   ret float %cast
 }
 
 define amdgpu_ps float @s_v_s_bfi_sha256_ch(i32 inreg %x, i32 %y, i32 inreg %z) {
 ; GFX7-LABEL: s_v_s_bfi_sha256_ch:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    v_mov_b32_e32 v1, s1
-; GFX7-NEXT:    v_bfi_b32 v0, s0, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    v_bfi_b32 v0, v1, v0, s1
 ; GFX7-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-LABEL: s_v_s_bfi_sha256_ch:
 ; GFX8:       ; %bb.0: ; %entry
-; GFX8-NEXT:    v_mov_b32_e32 v1, s1
-; GFX8-NEXT:    v_bfi_b32 v0, s0, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v1, s0
+; GFX8-NEXT:    v_bfi_b32 v0, v1, v0, s1
 ; GFX8-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: s_v_s_bfi_sha256_ch:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    v_bfi_b32 v0, s0, v0, s1
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: s_v_s_bfi_sha256_ch:
entry:
   %xor1 = xor i32 %z, %and
   %cast = bitcast i32 %xor1 to float
   ret float %cast
 }
 
 define amdgpu_ps float @s_s_v_bfi_sha256_ch(i32 inreg %x, i32 inreg %y, i32 %z) {
 ; GFX7-LABEL: s_s_v_bfi_sha256_ch:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    v_mov_b32_e32 v1, s1
-; GFX7-NEXT:    v_bfi_b32 v0, s0, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    v_bfi_b32 v0, v1, s1, v0
 ; GFX7-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-LABEL: s_s_v_bfi_sha256_ch:
 ; GFX8:       ; %bb.0: ; %entry
-; GFX8-NEXT:    v_mov_b32_e32 v1, s1
-; GFX8-NEXT:    v_bfi_b32 v0, s0, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s0
+; GFX8-NEXT:    v_bfi_b32 v0, v1, s1, v0
 ; GFX8-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-LABEL: s_s_v_bfi_sha256_ch:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    v_bfi_b32 v0, s0, s1, v0
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: s_s_v_bfi_sha256_ch:

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_bfi_b32 v0, v2, v0, v4
 ; GFX10-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-GISEL-LABEL: v_bitselect_v2i32_pat1:
 ; GFX8-GISEL:       ; %bb.0:
 ; GFX8-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, v4
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX8-GISEL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-GISEL-LABEL: v_bitselect_v2i32_pat1:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, v4
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
   %xor.0 = xor <2 x i32> %a, %mask
   %and = and <2 x i32> %xor.0, %b
   %bitselect = xor <2 x i32> %and, %mask
   ret <2 x i32> %bitselect
 }
 
 define i64 @v_bitselect_i64_pat_0(i64 %a, i64 %b, i64 %mask) {

 ; GFX8-GISEL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-GISEL-LABEL: v_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v6, -1, v0
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v7, -1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, v6, v4
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, v7, v5
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_and_b32_e32 v4, v6, v4
+; GFX10-GISEL-NEXT:    v_and_b32_e32 v5, v7, v5
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, v0, v2, v4
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, v1, v3, v5
 ; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   ret i64 %bitselect
 }
 

 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v3, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_s_s_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v2, -1, v0
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v3, -1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s2, v2
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s3, v3
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, v0, s0, v2
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, v1, s1, v3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
 ; GFX8-GISEL-NEXT:    s_andn2_b64 s[0:1], s[2:3], s[0:1]
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, s0, v0
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, s1, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: s_v_s_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
-; GFX10-GISEL-NEXT:    s_andn2_b64 s[0:1], s[2:3], s[0:1]
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, s1, v1
+; GFX10-GISEL-NEXT:    s_andn2_b64 s[2:3], s[2:3], s[0:1]
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, s0, v0, s2
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, s1, v1, s3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
 ; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, s2, v0
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, s3, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: s_s_v_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_not_b64 s[4:5], s[0:1]
-; GFX10-GISEL-NEXT:    s_and_b64 s[0:1], s[0:1], s[2:3]
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s4, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s5, v1
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, s1, v1
+; GFX10-GISEL-NEXT:    s_and_b64 s[2:3], s[0:1], s[2:3]
+; GFX10-GISEL-NEXT:    s_not_b64 s[0:1], s[0:1]
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, s0, v0, s2
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, s1, v1, s3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v3, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_v_s_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v4, -1, v0
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v5, -1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s0, v4
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s1, v5
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_and_b32_e32 v4, s0, v4
+; GFX10-GISEL-NEXT:    v_and_b32_e32 v5, s1, v5
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, v0, v2, v4
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, v1, v3, v5
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v4, v0
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v5, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_s_v_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v4, -1, v0
 ; GFX10-GISEL-NEXT:    v_xor_b32_e32 v5, -1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, v4, v2
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, v5, v3
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, v0, s0, v2
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, v1, s1, v3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX8-GISEL-NEXT:    v_and_b32_e32 v3, s1, v3
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: s_v_v_bitselect_i64_pat_0:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    s_not_b64 s[2:3], s[0:1]
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s2, v2
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s3, v3
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, s0, v0, v2
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, s1, v1, v3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %and0 = and i64 %a, %b
   %not.a = xor i64 %a, -1
   %and1 = and i64 %not.a, %mask
   %bitselect = or i64 %and0, %and1
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_bfi_b32 v0, v2, v0, v4
 ; GFX10-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-GISEL-LABEL: v_bitselect_i64_pat_1:
 ; GFX8-GISEL:       ; %bb.0:
 ; GFX8-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, v4
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX8-GISEL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-GISEL-LABEL: v_bitselect_i64_pat_1:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, v4
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
   %xor.0 = xor i64 %a, %mask
   %and = and i64 %xor.0, %b
   %bitselect = xor i64 %and, %mask
   ret i64 %bitselect
 }
 
 define amdgpu_ps <2 x float> @v_s_s_bitselect_i64_pat_1(i64 %a, i64 inreg %b, i64 inreg %mask) {

 ; GFX10-LABEL: v_s_s_bitselect_i64_pat_1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    v_bfi_b32 v0, s0, v0, s2
 ; GFX10-NEXT:    v_bfi_b32 v1, s1, v1, s3
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: v_s_s_bitselect_i64_pat_1:
 ; GFX8-GISEL:       ; %bb.0:
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, s3, v1
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, s3, v1
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s0
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, s2
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s1
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v2, v1, s3
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_s_s_bitselect_i64_pat_1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, s3, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, s3, v1
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, s0, v0, s2
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, s1, v1, s3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %xor.0 = xor i64 %a, %mask
   %and = and i64 %xor.0, %b
   %bitselect = xor i64 %and, %mask
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }
 

 ; GFX10-LABEL: s_s_v_bitselect_i64_pat_1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    v_bfi_b32 v0, s2, s0, v0
 ; GFX10-NEXT:    v_bfi_b32 v1, s3, s1, v1
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: s_s_v_bitselect_i64_pat_1:
 ; GFX8-GISEL:       ; %bb.0:
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v2, s0, v0
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v3, s1, v1
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v2, s2, v2
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v3, s3, v3
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v2, v0
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v3, v1
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v2, s0, v0
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s3
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v2, s1, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: s_s_v_bitselect_i64_pat_1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v2, s0, v0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v3, s1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s2, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s3, v3
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v2, v0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v3, v1
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, s2, s0, v0
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, s3, s1, v1
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
   %xor.0 = xor i64 %a, %mask
   %and = and i64 %xor.0, %b
   %bitselect = xor i64 %and, %mask
   %cast = bitcast i64 %bitselect to <2 x float>
   ret <2 x float> %cast
 }
 

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_bfi_b32 v0, v2, v0, v4
 ; GFX10-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-GISEL-LABEL: v_bitselect_i64_pat_2:
 ; GFX8-GISEL:       ; %bb.0:
 ; GFX8-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, v4
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX8-GISEL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-GISEL-LABEL: v_bitselect_i64_pat_2:
 ; GFX10-GISEL:       ; %bb.0:
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v5
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, v4
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, v5
 ; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
   %xor.0 = xor i64 %a, %mask
   %and = and i64 %xor.0, %b
   %bitselect = xor i64 %and, %mask
   ret i64 %bitselect
 }
 
 define i64 @v_bfi_sha256_ma_i64(i64 %x, i64 %y, i64 %z) {

 ; GFX10-NEXT:    v_xor_b32_e32 v1, v1, v3
 ; GFX10-NEXT:    v_bfi_b32 v0, v0, v4, v2
 ; GFX10-NEXT:    v_bfi_b32 v1, v1, v5, v3
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-GISEL-LABEL: v_bfi_sha256_ma_i64:
 ; GFX8-GISEL:       ; %bb.0: ; %entry
 ; GFX8-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v6, v0, v4
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v7, v1, v5
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v0, v4
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v1, v5
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, v2, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, v3, v1
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v6, v0
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v7, v1
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v2
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v3
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v0, v4, v2
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v1, v5, v3
 ; GFX8-GISEL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-GISEL-LABEL: v_bfi_sha256_ma_i64:
 ; GFX10-GISEL:       ; %bb.0: ; %entry
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v6, v0, v4
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v7, v1, v5
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v4
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v5
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, v2, v6
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, v3, v7
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, v0, v2
+; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v0, v4, v2
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v1, v5, v3
 ; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %and0 = and i64 %x, %z
   %or0 = or i64 %x, %z
   %and1 = and i64 %y, %or0
   %or1 = or i64 %and0, %and1
   ret i64 %or1
 }

 ; GFX10-NEXT:    v_xor_b32_e32 v0, s0, v0
 ; GFX10-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GFX10-NEXT:    v_bfi_b32 v0, v0, s2, s0
 ; GFX10-NEXT:    v_bfi_b32 v1, v1, s3, s1
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: v_s_s_bfi_sha256_ma_i64:
 ; GFX8-GISEL:       ; %bb.0: ; %entry
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v2, s2, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v3, s3, v1
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, s2, v0
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, s3, v1
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v2, v0
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v3, v1
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v0, v2, s0
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s3
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, s1, v1
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v1, v2, s1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_s_s_bfi_sha256_ma_i64:
 ; GFX10-GISEL:       ; %bb.0: ; %entry
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v2, s2, v0
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v3, s3, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s2, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s3, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s1, v3
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, s1, v1
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v0, s2, s0
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v1, s3, s1
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
 entry:
   %and0 = and i64 %x, %z
   %or0 = or i64 %x, %z
   %and1 = and i64 %y, %or0
   %or1 = or i64 %and0, %and1
   %cast = bitcast i64 %or1 to <2 x float>
   ret <2 x float> %cast

 ; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
 ; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, s4, v0
 ; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, s5, v1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: s_v_s_bfi_sha256_ma_i64:
 ; GFX10-GISEL:       ; %bb.0: ; %entry
-; GFX10-GISEL-NEXT:    s_or_b64 s[4:5], s[0:1], s[2:3]
-; GFX10-GISEL-NEXT:    s_and_b64 s[0:1], s[0:1], s[2:3]
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s4, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s5, v1
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, s1, v1
+; GFX10-GISEL-NEXT:    s_and_b64 s[4:5], s[0:1], s[2:3]
+; GFX10-GISEL-NEXT:    s_or_b64 s[0:1], s[0:1], s[2:3]
+; GFX10-GISEL-NEXT:    v_and_or_b32 v0, v0, s0, s4
+; GFX10-GISEL-NEXT:    v_and_or_b32 v1, v1, s1, s5
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
 entry:
   %and0 = and i64 %x, %z
   %or0 = or i64 %x, %z
   %and1 = and i64 %y, %or0
   %or1 = or i64 %and0, %and1
   %cast = bitcast i64 %or1 to <2 x float>
   ret <2 x float> %cast

 ; GFX10-NEXT:    v_xor_b32_e64 v2, s0, s2
 ; GFX10-NEXT:    v_xor_b32_e64 v3, s1, s3
 ; GFX10-NEXT:    v_bfi_b32 v0, v2, v0, s2
 ; GFX10-NEXT:    v_bfi_b32 v1, v3, v1, s3
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: s_s_v_bfi_sha256_ma_i64:
 ; GFX8-GISEL:       ; %bb.0: ; %entry
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v2, s0, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v3, s1, v1
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, s0, v0
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, s1, v1
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s2, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s3, v1
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v2, v0
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v3, v1
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s0
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v2, s2, v2
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, s2
+; GFX8-GISEL-NEXT:    v_mov_b32_e32 v2, s1
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v2, s3, v2
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v2, v1, s3
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: s_s_v_bfi_sha256_ma_i64:
 ; GFX10-GISEL:       ; %bb.0: ; %entry
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v2, s0, v0
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v3, s1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s2, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s3, v3
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_xor_b32_e64 v2, s0, s2
+; GFX10-GISEL-NEXT:    v_xor_b32_e64 v3, s1, s3
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v2, v0, s2
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v3, v1, s3
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
 entry:
   %and0 = and i64 %x, %z
   %or0 = or i64 %x, %z
   %and1 = and i64 %y, %or0
   %or1 = or i64 %and0, %and1
   %cast = bitcast i64 %or1 to <2 x float>
   ret <2 x float> %cast

 ; GFX10-NEXT:    v_xor_b32_e32 v0, s0, v0
 ; GFX10-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GFX10-NEXT:    v_bfi_b32 v0, v0, v2, s0
 ; GFX10-NEXT:    v_bfi_b32 v1, v1, v3, s1
 ; GFX10-NEXT:    ; return to shader part epilog
 ;
 ; GFX8-GISEL-LABEL: v_s_v_bfi_sha256_ma_i64:
 ; GFX8-GISEL:       ; %bb.0: ; %entry
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v4, v0, v2
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v5, v1, v3
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v0, s0, v0
-; GFX8-GISEL-NEXT:    v_and_b32_e32 v1, s1, v1
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v0, v4, v0
-; GFX8-GISEL-NEXT:    v_or_b32_e32 v1, v5, v1
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX8-GISEL-NEXT:    v_xor_b32_e32 v1, s1, v1
+; GFX8-GISEL-NEXT:    v_bfi_b32 v0, v0, v2, s0
+; GFX8-GISEL-NEXT:    v_bfi_b32 v1, v1, v3, s1
 ; GFX8-GISEL-NEXT:    ; return to shader part epilog
 ;
 ; GFX10-GISEL-LABEL: v_s_v_bfi_sha256_ma_i64:
 ; GFX10-GISEL:       ; %bb.0: ; %entry
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v4, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v5, v1, v3
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v1, v1, v3
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s0, v4
-; GFX10-GISEL-NEXT:    v_and_b32_e32 v3, s1, v5
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX10-GISEL-NEXT:    v_xor_b32_e32 v1, s1, v1
+; GFX10-GISEL-NEXT:    v_bfi_b32 v0, v0, v2, s0
+; GFX10-GISEL-NEXT:    v_bfi_b32 v1, v1, v3, s1
 ; GFX10-GISEL-NEXT:    ; return to shader part epilog
 entry:
   %and0 = and i64 %x, %z
   %or0 = or i64 %x, %z
   %and1 = and i64 %y, %or0
   %or1 = or i64 %and0, %and1
   %cast = bitcast i64 %or1 to <2 x float>
   ret <2 x float> %cast

llvm/test/CodeGen/AMDGPU/cttz_zero_undef.ll

 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v0, v1, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v1, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    global_load_ubyte v3, v1, s[2:3] offset:2
 ; GFX9-GISEL-NEXT:    global_load_ubyte v4, v1, s[2:3] offset:3
 ; GFX9-GISEL-NEXT:    global_load_ubyte v5, v1, s[2:3] offset:4
 ; GFX9-GISEL-NEXT:    global_load_ubyte v6, v1, s[2:3] offset:5
-; GFX9-GISEL-NEXT:    global_load_ubyte v7, v1, s[2:3] offset:7
-; GFX9-GISEL-NEXT:    global_load_ubyte v8, v1, s[2:3] offset:6
+; GFX9-GISEL-NEXT:    global_load_ubyte v7, v1, s[2:3] offset:6
+; GFX9-GISEL-NEXT:    global_load_ubyte v8, v1, s[2:3] offset:7
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(6)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v0, v2, 8, v0
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(5)
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(4)
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 24, v4
 ; GFX9-GISEL-NEXT:    v_or3_b32 v2, v2, v3, v0
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(2)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v4, v6, 8, v5
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v5, 24, v7
+; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v5, 16, v7
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v6, 16, v8
-; GFX9-GISEL-NEXT:    v_or3_b32 v3, v5, v6, v4
+; GFX9-GISEL-NEXT:    v_lshl_or_b32 v0, v8, 24, v5
+; GFX9-GISEL-NEXT:    v_or3_b32 v3, v0, v4, 0
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v4, v3
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v0, v2
 ; GFX9-GISEL-NEXT:    v_add_u32_e32 v4, 32, v4
 ; GFX9-GISEL-NEXT:    v_cmp_ne_u64_e32 vcc, 0, v[2:3]
 ; GFX9-GISEL-NEXT:    v_min_u32_e32 v0, v0, v4
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v0, 64, v0, vcc
 ; GFX9-GISEL-NEXT:    global_store_dwordx2 v1, v[0:1], s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm