[AMDGPU] Implement whole wave register spill

[cdevadas]

Summary

To reduce the register pressure during allocation,
when the allocator spills a virtual register that
corresponds to a whole wave mode operation, the
spill loads and restores should be activated for
all lanes by temporarily flipping all bits in exec
register to one just before the spills. It is not
implemented in the compiler as of today and this
patch enables the necessary support.

This is a pre-patch before the SGPR spill to virtual
VGPR lanes that would eventually causes the whole
wave register spills during allocation.

[yassingh]

• Rebase over new TARGET::PRED_COPY approach

[yassingh]

Rebase

[arsenm]

llvm/lib/Target/AMDGPU/SIDefines.h

934

I assumed the value would be 1. Change this to be the pre-shifted values rather than using 1 << flag in the getter?

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

13430–13436

Move this up with the other reserved register logic

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

549

Capitalization is weird, Capitalize clone

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h

651

Why are the inbounds checks needed? The analagous MRI functions do not tolerate invalid register values

656–657

This should assert (like setFlag already does)

659

Ditto, why range check?

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2151–2152

Braces and indent under if

2160

Indent off

[cdevadas]

llvm/lib/Target/AMDGPU/SIDefines.h

934

Will do.

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

549

Will do.

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h

651

VRegFlags will be grown each time (if the size exceeds the initial size set) when a new virtual register is created/cloned. The MRI's 'VRegToType' has a similar bound check.

656–657

I guess it's already been discussed in the downstream review. There are instances when Physical registers are taken to spill - fastAlloc, CSR spills during PEI. To handle such instances, an early return was inserted intentionally for the PhysRegs instead of an assertion.

[arsenm]

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h

651

VRegToType isn't really analagous. Untyped and typed registers can coexist, I'd expect it to be unnecessary here

[arsenm]

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

13430–13436

Actually, why is this not in getReservedRegs?

[yassingh]

Addressed Matt's comments:

• Updated flag
• Formatting

[yassingh]

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h

651

Getting rid of inbounds check causes multiple test failures. (In spill192.mir, spill224.mir, spill288.mir, spill320.mir, etc).

[yassingh]

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2151–2152

Missed indenting, will fix it.

[cdevadas]

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2151–2152

Update the patch after clang-format.

[arsenm]

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

647

D149775 is adding another reserved register, but I'm pretty sure they cannot contextually overlap. After both patches are in they should be consolidated to reuse the same reserved register

[arsenm]

llvm/lib/Target/AMDGPU/SIFrameLowering.cpp

1467

ST.getWaveMaskRegClass

1469

This misses VCC as a possibility, which is fixed by using getWaveMaskRegClass

llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp

338

Same, use getWaveMaskRegClass

[yassingh]

• Review comments (use getWaveMaskRegClass(), getExec() functions)

[yassingh]

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

647

Will update once both lands. Added FIXME for now.

2151–2152

Oh right, clang-format reverts the indentation. Going to leave it as it is, else clang-format will keep trying.

[arsenm]

lgtm besides a few nits

llvm/include/llvm/CodeGen/MachineRegisterInfo.h

60

Didn't realize this already existed

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

4985

ST is already a member

4986

This is just this already, you don't need to query it

4987–4988

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2218–2219

Indentation's off

2226

Indentation's off

[yassingh]

• review comments

[yassingh]

Reuploading diff

[yassingh]

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2218–2219

clang format as well arc undos this indentation.

2226

same

[arsenm]

lgtm with nits

llvm/lib/Target/AMDGPU/SIFrameLowering.cpp

70

Documentation comment for what IncludeScratchCopy is

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2218–2219

Well it's clearly wrong, so just manually fix it

2226

Well it's clearly wrong, so just manually fix it

[yassingh]

Review comments and fix formatting

[yassingh]

Rebase

[cdevadas]

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

2151

No need for braces in case of a single instruction.

2159

Wrong indentation.

2217

Avoid braces.

[yassingh]

rebase and formatting

[arsenm]

llvm/lib/Target/AMDGPU/SIInstructions.td

936

Should be marked convergent

[yassingh]

Mark SI_SPILL_WWM_V32 convergent, rebase.

[yassingh]

llvm/test/CodeGen/MIR/AMDGPU/long-branch-reg-all-sgpr-used.ll

42 ↗

(On Diff #536981)

We are using sgpr100, sgpr101 in our code below using inline asm calls why not sure what do they mean here?
cc @cdevadas

307 ↗

(On Diff #536981)

@crobeck I believe this test is now not serving the purpose you added it for. I tried freeing up s98 and s99(see below changes in the same file) expecting them to be used for 'longBranchReservedReg', but didn't work. Any pointers on why it might be?

[yassingh]

Rebase before merge

llvm/include/llvm/CodeGen/MachineRegisterInfo.h

 public:
   class Delegate {
     virtual void anchor();
 
   public:
     virtual ~Delegate() = default;
 
     virtual void MRI_NoteNewVirtualRegister(Register Reg) = 0;
-    virtual void MRI_NotecloneVirtualRegister(Register NewReg,
+    virtual void MRI_NoteCloneVirtualRegister(Register NewReg,
                                               Register SrcReg) {
       MRI_NoteNewVirtualRegister(NewReg);
     }
   };
 
 private:
   MachineFunction *MF;
   SmallPtrSet<Delegate *, 1> TheDelegates;
public:
 
   void noteNewVirtualRegister(Register Reg) {
     for (auto *TheDelegate : TheDelegates)
       TheDelegate->MRI_NoteNewVirtualRegister(Reg);
   }
 
   void noteCloneVirtualRegister(Register NewReg, Register SrcReg) {
     for (auto *TheDelegate : TheDelegates)
-      TheDelegate->MRI_NotecloneVirtualRegister(NewReg, SrcReg);
+      TheDelegate->MRI_NoteCloneVirtualRegister(NewReg, SrcReg);
   }
 
   //===--------------------------------------------------------------------===//
   // Function State
   //===--------------------------------------------------------------------===//
 
   // isSSA - Returns true when the machine function is in SSA form. Early
   // passes require the machine function to be in SSA form where every virtual

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h

public:
   const TargetSubtargetInfo *getSubtargetImpl(const Function &) const override;
 
   TargetTransformInfo getTargetTransformInfo(const Function &F) const override;
 
   bool useIPRA() const override {
     return true;
   }
 
+  void registerMachineRegisterInfoCallback(MachineFunction &MF) const override;
+
   MachineFunctionInfo *
   createMachineFunctionInfo(BumpPtrAllocator &Allocator, const Function &F,
                             const TargetSubtargetInfo *STI) const override;
 
   yaml::MachineFunctionInfo *createDefaultFuncInfoYAML() const override;
   yaml::MachineFunctionInfo *
   convertFuncInfoToYAML(const MachineFunction &MF) const override;
   bool parseMachineFunctionInfo(const yaml::MachineFunctionInfo &,

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

void GCNPassConfig::addPreEmitPass() {
 
   addPass(&BranchRelaxationPassID);
 }
 
 TargetPassConfig *GCNTargetMachine::createPassConfig(PassManagerBase &PM) {
   return new GCNPassConfig(*this, PM);
 }
 
+void GCNTargetMachine::registerMachineRegisterInfoCallback(
+    MachineFunction &MF) const {
+  SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
+  MF.getRegInfo().addDelegate(MFI);
+}
+
 MachineFunctionInfo *GCNTargetMachine::createMachineFunctionInfo(
     BumpPtrAllocator &Allocator, const Function &F,
     const TargetSubtargetInfo *STI) const {
   return SIMachineFunctionInfo::create<SIMachineFunctionInfo>(
       Allocator, F, static_cast<const GCNSubtarget *>(STI));
 }
 
 yaml::MachineFunctionInfo *GCNTargetMachine::createDefaultFuncInfoYAML() const {
bool GCNTargetMachine::parseMachineFunctionInfo(
   auto parseOptionalRegister = [&](const yaml::StringValue &RegName,
                                    Register &RegVal) {
     return !RegName.Value.empty() && parseRegister(RegName, RegVal);
   };
 
   if (parseOptionalRegister(YamlMFI.VGPRForAGPRCopy, MFI->VGPRForAGPRCopy))
     return true;
 
+  if (parseOptionalRegister(YamlMFI.SGPRForEXECCopy, MFI->SGPRForEXECCopy))
+    return true;
+
   auto diagnoseRegisterClass = [&](const yaml::StringValue &RegName) {
     // Create a diagnostic for a the register string literal.
     const MemoryBuffer &Buffer =
         *PFS.SM->getMemoryBuffer(PFS.SM->getMainFileID());
     Error = SMDiagnostic(*PFS.SM, SMLoc(), Buffer.getBufferIdentifier(), 1,
                          RegName.Value.size(), SourceMgr::DK_Error,
                          "incorrect register class for field", RegName.Value,
                          std::nullopt, std::nullopt);

llvm/lib/Target/AMDGPU/SIDefines.h

enum Offset_COV5 : unsigned {
   COMPLETION_ACTION_OFFSET = 112,
 
   PRIVATE_BASE_OFFSET = 192,
   SHARED_BASE_OFFSET = 196,
   QUEUE_PTR_OFFSET = 200,
 };
 
 } // namespace ImplicitArg
+
+namespace VirtRegFlag {
+// Virtual register flags used for various target specific handlings during
+// codegen.
+enum Register_Flag : uint8_t {
+  // Register operand in a whole-wave mode operation.
+  WWM_REG = 1 << 0,
+};
+
+} // namespace VirtRegFlag
+
 } // namespace AMDGPU
 
 #define R_00B028_SPI_SHADER_PGM_RSRC1_PS                                0x00B028
 #define   S_00B028_VGPRS(x)                                           (((x) & 0x3F) << 0)
 #define   S_00B028_SGPRS(x)                                           (((x) & 0x0F) << 6)
 #define   S_00B028_MEM_ORDERED(x)                                     (((x) & 0x1) << 25)
 #define   G_00B028_MEM_ORDERED(x)                                     (((x) >> 25) & 0x1)
 #define   C_00B028_MEM_ORDERED                                        0xFDFFFFFF

llvm/lib/Target/AMDGPU/SIFrameLowering.h

void emitEpilogue(MachineFunction &MF,
                     MachineBasicBlock &MBB) const override;
   StackOffset getFrameIndexReference(const MachineFunction &MF, int FI,
                                      Register &FrameReg) const override;
 
   void determineCalleeSaves(MachineFunction &MF, BitVector &SavedRegs,
                             RegScavenger *RS = nullptr) const override;
   void determineCalleeSavesSGPR(MachineFunction &MF, BitVector &SavedRegs,
                                 RegScavenger *RS = nullptr) const;
-  void determinePrologEpilogSGPRSaves(MachineFunction &MF,
-                                      BitVector &SavedRegs) const;
+  void determinePrologEpilogSGPRSaves(MachineFunction &MF, BitVector &SavedRegs,
+                                      bool NeedExecCopyReservedReg) const;
   void emitCSRSpillStores(MachineFunction &MF, MachineBasicBlock &MBB,
                           MachineBasicBlock::iterator MBBI, DebugLoc &DL,
                           LivePhysRegs &LiveRegs, Register FrameReg,
                           Register FramePtrRegScratchCopy) const;
   void emitCSRSpillRestores(MachineFunction &MF, MachineBasicBlock &MBB,
                             MachineBasicBlock::iterator MBBI, DebugLoc &DL,
                             LivePhysRegs &LiveRegs, Register FrameReg,
                             Register FramePtrRegScratchCopy) const;

llvm/lib/Target/AMDGPU/SIFrameLowering.cpp

if (LiveRegs.available(MRI, Reg))
       return Reg;
   }
 
   return MCRegister();
 }
 
 static void getVGPRSpillLaneOrTempRegister(
     MachineFunction &MF, LivePhysRegs &LiveRegs, Register SGPR,
-    const TargetRegisterClass &RC = AMDGPU::SReg_32_XM0_XEXECRegClass) {
+    const TargetRegisterClass &RC = AMDGPU::SReg_32_XM0_XEXECRegClass,
+    bool IncludeScratchCopy = true) {
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   MachineFrameInfo &FrameInfo = MF.getFrameInfo();
 
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   const SIRegisterInfo *TRI = ST.getRegisterInfo();
   unsigned Size = TRI->getSpillSize(RC);
   Align Alignment = TRI->getSpillAlign(RC);
 
   // We need to save and restore the given SGPR.
 
+  Register ScratchSGPR;
   // 1: Try to save the given register into an unused scratch SGPR. The LiveRegs
-  // should have all the callee saved registers marked as used.
-  Register ScratchSGPR = findUnusedRegister(MF.getRegInfo(), LiveRegs, RC);
+  // should have all the callee saved registers marked as used. For certain
+  // cases we skip copy to scratch SGPR.
+  if (IncludeScratchCopy)
+    ScratchSGPR = findUnusedRegister(MF.getRegInfo(), LiveRegs, RC);
 
   if (!ScratchSGPR) {
     int FI = FrameInfo.CreateStackObject(Size, Alignment, true, nullptr,
                                          TargetStackID::SGPRSpill);
 
     if (TRI->spillSGPRToVGPR() &&
         MFI->allocateSGPRSpillToVGPRLane(MF, FI, /* IsPrologEpilog */ true)) {
       // 2: There's no free lane to spill, and no free register to save the
auto StoreWWMRegisters =
                            VGPR, FI, FrameReg);
         }
       };
 
   StoreWWMRegisters(WWMScratchRegs);
   if (!WWMCalleeSavedRegs.empty()) {
     if (ScratchExecCopy) {
       unsigned MovOpc = ST.isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
-      MCRegister Exec = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
-      BuildMI(MBB, MBBI, DL, TII->get(MovOpc), Exec).addImm(-1);
+      BuildMI(MBB, MBBI, DL, TII->get(MovOpc), TRI.getExec()).addImm(-1);
     } else {
       ScratchExecCopy = buildScratchExecCopy(LiveRegs, MF, MBB, MBBI, DL,
                                              /*IsProlog*/ true,
                                              /*EnableInactiveLanes*/ false);
     }
   }
 
   StoreWWMRegisters(WWMCalleeSavedRegs);
   if (ScratchExecCopy) {
     // FIXME: Split block and make terminator.
     unsigned ExecMov = ST.isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
-    MCRegister Exec = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
-    BuildMI(MBB, MBBI, DL, TII->get(ExecMov), Exec)
+    BuildMI(MBB, MBBI, DL, TII->get(ExecMov), TRI.getExec())
         .addReg(ScratchExecCopy, RegState::Kill);
     LiveRegs.addReg(ScratchExecCopy);
   }
 
   Register FramePtrReg = FuncInfo->getFrameOffsetReg();
 
   for (const auto &Spill : FuncInfo->getPrologEpilogSGPRSpills()) {
     // Special handle FP spill:
auto RestoreWWMRegisters =
                              VGPR, FI, FrameReg);
         }
       };
 
   RestoreWWMRegisters(WWMScratchRegs);
   if (!WWMCalleeSavedRegs.empty()) {
     if (ScratchExecCopy) {
       unsigned MovOpc = ST.isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
-      MCRegister Exec = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
-      BuildMI(MBB, MBBI, DL, TII->get(MovOpc), Exec).addImm(-1);
+      BuildMI(MBB, MBBI, DL, TII->get(MovOpc), TRI.getExec()).addImm(-1);
     } else {
       ScratchExecCopy = buildScratchExecCopy(LiveRegs, MF, MBB, MBBI, DL,
                                              /*IsProlog*/ false,
                                              /*EnableInactiveLanes*/ false);
     }
   }
 
   RestoreWWMRegisters(WWMCalleeSavedRegs);
   if (ScratchExecCopy) {
     // FIXME: Split block and make terminator.
     unsigned ExecMov = ST.isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
-    MCRegister Exec = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
-    BuildMI(MBB, MBBI, DL, TII->get(ExecMov), Exec)
+    BuildMI(MBB, MBBI, DL, TII->get(ExecMov), TRI.getExec())
         .addReg(ScratchExecCopy, RegState::Kill);
   }
 }
 
 void SIFrameLowering::emitPrologue(MachineFunction &MF,
                                    MachineBasicBlock &MBB) const {
   SIMachineFunctionInfo *FuncInfo = MF.getInfo<SIMachineFunctionInfo>();
   if (FuncInfo->isEntryFunction()) {
if (UnusedLowVGPR && (TRI->getHWRegIndex(UnusedLowVGPR) <
       MRI.freezeReservedRegs(MF);
     }
   }
 }
 
 // The special SGPR spills like the one needed for FP, BP or any reserved
 // registers delayed until frame lowering.
 void SIFrameLowering::determinePrologEpilogSGPRSaves(
-    MachineFunction &MF, BitVector &SavedVGPRs) const {
+    MachineFunction &MF, BitVector &SavedVGPRs,
+    bool NeedExecCopyReservedReg) const {
   MachineFrameInfo &FrameInfo = MF.getFrameInfo();
+  MachineRegisterInfo &MRI = MF.getRegInfo();
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   const SIRegisterInfo *TRI = ST.getRegisterInfo();
   LivePhysRegs LiveRegs;
   LiveRegs.init(*TRI);
   // Initially mark callee saved registers as used so we will not choose them
   // while looking for scratch SGPRs.
   const MCPhysReg *CSRegs = MF.getRegInfo().getCalleeSavedRegs();
   for (unsigned I = 0; CSRegs[I]; ++I)
     LiveRegs.addReg(CSRegs[I]);
 
+  const TargetRegisterClass &RC = *TRI->getWaveMaskRegClass();
+
+  if (NeedExecCopyReservedReg) {
+    Register ReservedReg = MFI->getSGPRForEXECCopy();
+    assert(ReservedReg && "Should have reserved an SGPR for EXEC copy.");
+    Register UnusedScratchReg = findUnusedRegister(MRI, LiveRegs, RC);
+    if (UnusedScratchReg) {
+      // If found any unused scratch SGPR, reserve the register itself for Exec
+      // copy and there is no need for any spills in that case.
+      MFI->setSGPRForEXECCopy(UnusedScratchReg);
+      LiveRegs.addReg(UnusedScratchReg);
+    } else {
+      // Needs spill.
+      assert(!MFI->hasPrologEpilogSGPRSpillEntry(ReservedReg) &&
+             "Re-reserving spill slot for EXEC copy register");
+      getVGPRSpillLaneOrTempRegister(MF, LiveRegs, ReservedReg, RC,
+                                     /*IncludeScratchCopy=*/false);
+    }
+  }
+
   // hasFP only knows about stack objects that already exist. We're now
   // determining the stack slots that will be created, so we have to predict
   // them. Stack objects force FP usage with calls.
   //
   // Note a new VGPR CSR may be introduced if one is used for the spill, but we
   // don't want to report it here.
   //
   // FIXME: Is this really hasReservedCallFrame?
void SIFrameLowering::determineCalleeSaves(MachineFunction &MF,
                                            RegScavenger *RS) const {
   TargetFrameLowering::determineCalleeSaves(MF, SavedVGPRs, RS);
   SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
   if (MFI->isEntryFunction())
     return;
 
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   const SIRegisterInfo *TRI = ST.getRegisterInfo();
+  const SIInstrInfo *TII = ST.getInstrInfo();
+  bool NeedExecCopyReservedReg = false;
 
   MachineInstr *ReturnMI = nullptr;
   for (MachineBasicBlock &MBB : MF) {
     for (MachineInstr &MI : MBB) {
       // WRITELANE instructions used for SGPR spills can overwrite the inactive
       // lanes of VGPRs and callee must spill and restore them even if they are
       // marked Caller-saved.
 
       // TODO: Handle this elsewhere at an early point. Walking through all MBBs
       // here would be a bad heuristic. A better way should be by calling
       // allocateWWMSpill during the regalloc pipeline whenever a physical
       // register is allocated for the intended virtual registers. That will
       // also help excluding the general use of WRITELANE/READLANE intrinsics
       // that won't really need any such special handling.
       if (MI.getOpcode() == AMDGPU::V_WRITELANE_B32)
         MFI->allocateWWMSpill(MF, MI.getOperand(0).getReg());
       else if (MI.getOpcode() == AMDGPU::V_READLANE_B32)
         MFI->allocateWWMSpill(MF, MI.getOperand(1).getReg());
+      else if (TII->isWWMRegSpillOpcode(MI.getOpcode()))
+        NeedExecCopyReservedReg = true;
       else if (MI.getOpcode() == AMDGPU::SI_RETURN ||
                MI.getOpcode() == AMDGPU::SI_RETURN_TO_EPILOG) {
         // We expect all return to be the same size.
         assert(!ReturnMI ||
                (count_if(MI.operands(), [](auto Op) { return Op.isReg(); }) ==
                 count_if(ReturnMI->operands(), [](auto Op) { return Op.isReg(); })));
         ReturnMI = &MI;
       }
void SIFrameLowering::determineCalleeSaves(MachineFunction &MF,
   SavedVGPRs.clearBitsNotInMask(TRI->getAllVectorRegMask());
 
   // Do not save AGPRs prior to GFX90A because there was no easy way to do so.
   // In gfx908 there was do AGPR loads and stores and thus spilling also
   // require a temporary VGPR.
   if (!ST.hasGFX90AInsts())
     SavedVGPRs.clearBitsInMask(TRI->getAllAGPRRegMask());
 
-  determinePrologEpilogSGPRSaves(MF, SavedVGPRs);
+  determinePrologEpilogSGPRSaves(MF, SavedVGPRs, NeedExecCopyReservedReg);
 
   // The Whole-Wave VGPRs need to be specially inserted in the prolog, so don't
   // allow the default insertion to handle them.
   for (auto &Reg : MFI->getWWMSpills())
     SavedVGPRs.reset(Reg.first);
 
   // Mark all lane VGPRs as BB LiveIns.
   for (MachineBasicBlock &MBB : MF) {

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

void SITargetLowering::finalizeLowering(MachineFunction &MF) const {
   const SIRegisterInfo *TRI = Subtarget->getRegisterInfo();
   const SIInstrInfo *TII = ST.getInstrInfo();
 
   if (Info->isEntryFunction()) {
     // Callable functions have fixed registers used for stack access.
     reservePrivateMemoryRegs(getTargetMachine(), MF, *TRI, *Info);
   }
 
+  // TODO: Move this logic to getReservedRegs()
+  // Reserve the SGPR(s) to save/restore EXEC for WWM spill/copy handling.
+  unsigned MaxNumSGPRs = ST.getMaxNumSGPRs(MF);
+  Register SReg = ST.isWave32()
+                      ? AMDGPU::SGPR_32RegClass.getRegister(MaxNumSGPRs - 1)
+                      : TRI->getAlignedHighSGPRForRC(MF, /*Align=*/2,
+                                                     &AMDGPU::SGPR_64RegClass);
+  Info->setSGPRForEXECCopy(SReg);
+
   assert(!TRI->isSubRegister(Info->getScratchRSrcReg(),
                              Info->getStackPtrOffsetReg()));
   if (Info->getStackPtrOffsetReg() != AMDGPU::SP_REG)
     MRI.replaceRegWith(AMDGPU::SP_REG, Info->getStackPtrOffsetReg());
 
   // We need to worry about replacing the default register with itself in case
   // of MIR testcases missing the MFI.
   if (Info->getScratchRSrcReg() != AMDGPU::PRIVATE_RSRC_REG)
for (unsigned I = 0, E = MRI.getNumVirtRegs(); I != E; ++I) {
       if (!RC)
         continue;
       int NewClassID = getAlignedAGPRClassID(RC->getID());
       if (NewClassID != -1)
         MRI.setRegClass(Reg, TRI->getRegClass(NewClassID));
     }
   }
 
   TargetLoweringBase::finalizeLowering(MF);
 }
 
 void SITargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
                                                      KnownBits &Known,
                                                      const APInt &DemandedElts,
                                                      const SelectionDAG &DAG,
                                                      unsigned Depth) const {
   Known.resetAll();
   unsigned Opc = Op.getOpcode();
   switch (Opc) {
   case ISD::INTRINSIC_WO_CHAIN: {
     unsigned IID = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
     switch (IID) {
     case Intrinsic::amdgcn_mbcnt_lo:

llvm/lib/Target/AMDGPU/SIInstrInfo.h

public:
   static bool isSGPRSpill(const MachineInstr &MI) {
     return MI.getDesc().TSFlags & SIInstrFlags::SGPRSpill;
   }
 
   bool isSGPRSpill(uint16_t Opcode) const {
     return get(Opcode).TSFlags & SIInstrFlags::SGPRSpill;
   }
 
+  static bool isWWMRegSpillOpcode(uint16_t Opcode) {
+    return Opcode == AMDGPU::SI_SPILL_WWM_V32_SAVE ||
+           Opcode == AMDGPU::SI_SPILL_WWM_V32_RESTORE;
+  }
+
   static bool isDPP(const MachineInstr &MI) {
     return MI.getDesc().TSFlags & SIInstrFlags::DPP;
   }
 
   bool isDPP(uint16_t Opcode) const {
     return get(Opcode).TSFlags & SIInstrFlags::DPP;
   }
 
public:
   MachineInstr *buildShrunkInst(MachineInstr &MI,
                                 unsigned NewOpcode) const;
 
   bool verifyInstruction(const MachineInstr &MI,
                          StringRef &ErrInfo) const override;
 
   unsigned getVALUOp(const MachineInstr &MI) const;
 
+  void insertScratchExecCopy(MachineFunction &MF, MachineBasicBlock &MBB,
+                             MachineBasicBlock::iterator MBBI,
+                             const DebugLoc &DL, Register Reg,
+                             bool IsSCCLive) const;
+
+  void restoreExec(MachineFunction &MF, MachineBasicBlock &MBB,
+                   MachineBasicBlock::iterator MBBI, const DebugLoc &DL,
+                   Register Reg) const;
+
   /// Return the correct register class for \p OpNo.  For target-specific
   /// instructions, this will return the register class that has been defined
   /// in tablegen.  For generic instructions, like REG_SEQUENCE it will return
   /// the register class of its machine operand.
   /// to infer the correct register class base on the other operands.
   const TargetRegisterClass *getOpRegClass(const MachineInstr &MI,
                                            unsigned OpNo) const;
 

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

case 64:
     return AMDGPU::SI_SPILL_AV512_SAVE;
   case 128:
     return AMDGPU::SI_SPILL_AV1024_SAVE;
   default:
     llvm_unreachable("unknown register size");
   }
 }
 
+static unsigned getWWMRegSpillSaveOpcode(unsigned Size) {
+  // Currently, there is only 32-bit WWM register spills needed.
+  if (Size != 4)
+    llvm_unreachable("unknown wwm register spill size");
+
+  return AMDGPU::SI_SPILL_WWM_V32_SAVE;
+}
+
+static unsigned getVectorRegSpillSaveOpcode(Register Reg,
+                                            const TargetRegisterClass *RC,
+                                            unsigned Size,
+                                            const SIRegisterInfo &TRI,
+                                            const SIMachineFunctionInfo &MFI) {
+  // Choose the right opcode if spilling a WWM register.
+  if (MFI.checkFlag(Reg, AMDGPU::VirtRegFlag::WWM_REG))
+    return getWWMRegSpillSaveOpcode(Size);
+
+  if (TRI.isVectorSuperClass(RC))
+    return getAVSpillSaveOpcode(Size);
+
+  return TRI.isAGPRClass(RC) ? getAGPRSpillSaveOpcode(Size)
+                             : getVGPRSpillSaveOpcode(Size);
+}
+
 void SIInstrInfo::storeRegToStackSlot(
     MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, Register SrcReg,
     bool isKill, int FrameIndex, const TargetRegisterClass *RC,
     const TargetRegisterInfo *TRI, Register VReg) const {
   MachineFunction *MF = MBB.getParent();
   SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
   MachineFrameInfo &FrameInfo = MF->getFrameInfo();
   const DebugLoc &DL = MBB.findDebugLoc(MI);
BuildMI(MBB, MI, DL, OpDesc)
       .addMemOperand(MMO)
       .addReg(MFI->getStackPtrOffsetReg(), RegState::Implicit);
 
     if (RI.spillSGPRToVGPR())
       FrameInfo.setStackID(FrameIndex, TargetStackID::SGPRSpill);
     return;
   }
 
-  unsigned Opcode = RI.isVectorSuperClass(RC)
-                        ? getAVSpillSaveOpcode(SpillSize)
-                        : RI.isAGPRClass(RC)
-                              ? getAGPRSpillSaveOpcode(SpillSize)
-                              : getVGPRSpillSaveOpcode(SpillSize);
+  unsigned Opcode = getVectorRegSpillSaveOpcode(VReg ? VReg : SrcReg, RC,
+                                                SpillSize, RI, *MFI);
   MFI->setHasSpilledVGPRs();
 
   BuildMI(MBB, MI, DL, get(Opcode))
     .addReg(SrcReg, getKillRegState(isKill)) // data
     .addFrameIndex(FrameIndex)               // addr
     .addReg(MFI->getStackPtrOffsetReg())     // scratch_offset
     .addImm(0)                               // offset
     .addMemOperand(MMO);
case 64:
     return AMDGPU::SI_SPILL_AV512_RESTORE;
   case 128:
     return AMDGPU::SI_SPILL_AV1024_RESTORE;
   default:
     llvm_unreachable("unknown register size");
   }
 }
 
+static unsigned getWWMRegSpillRestoreOpcode(unsigned Size) {
+  // Currently, there is only 32-bit WWM register spills needed.
+  if (Size != 4)
+    llvm_unreachable("unknown wwm register spill size");
+
+  return AMDGPU::SI_SPILL_WWM_V32_RESTORE;
+}
+
+static unsigned
+getVectorRegSpillRestoreOpcode(Register Reg, const TargetRegisterClass *RC,
+                               unsigned Size, const SIRegisterInfo &TRI,
+                               const SIMachineFunctionInfo &MFI) {
+  // Choose the right opcode if restoring a WWM register.
+  if (MFI.checkFlag(Reg, AMDGPU::VirtRegFlag::WWM_REG))
+    return getWWMRegSpillRestoreOpcode(Size);
+
+  if (TRI.isVectorSuperClass(RC))
+    return getAVSpillRestoreOpcode(Size);
+
+  return TRI.isAGPRClass(RC) ? getAGPRSpillRestoreOpcode(Size)
+                             : getVGPRSpillRestoreOpcode(Size);
+}
+
 void SIInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
                                        MachineBasicBlock::iterator MI,
                                        Register DestReg, int FrameIndex,
                                        const TargetRegisterClass *RC,
                                        const TargetRegisterInfo *TRI,
                                        Register VReg) const {
   MachineFunction *MF = MBB.getParent();
   SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
if (RI.isSGPRClass(RC)) {
     BuildMI(MBB, MI, DL, OpDesc, DestReg)
       .addFrameIndex(FrameIndex) // addr
       .addMemOperand(MMO)
       .addReg(MFI->getStackPtrOffsetReg(), RegState::Implicit);
 
     return;
   }
 
-  unsigned Opcode = RI.isVectorSuperClass(RC)
-                        ? getAVSpillRestoreOpcode(SpillSize)
-                        : RI.isAGPRClass(RC)
-                              ? getAGPRSpillRestoreOpcode(SpillSize)
-                              : getVGPRSpillRestoreOpcode(SpillSize);
+  unsigned Opcode = getVectorRegSpillRestoreOpcode(VReg ? VReg : DestReg, RC,
+                                                   SpillSize, RI, *MFI);
   BuildMI(MBB, MI, DL, get(Opcode), DestReg)
       .addFrameIndex(FrameIndex)           // vaddr
       .addReg(MFI->getStackPtrOffsetReg()) // scratch_offset
       .addImm(0)                           // offset
       .addMemOperand(MMO);
 }
 
 void SIInstrInfo::insertNoop(MachineBasicBlock &MBB,
unsigned SIInstrInfo::getVALUOp(const MachineInstr &MI) const {
   case AMDGPU::S_FLBIT_I32: return AMDGPU::V_FFBH_I32_e64;
   case AMDGPU::S_CBRANCH_SCC0: return AMDGPU::S_CBRANCH_VCCZ;
   case AMDGPU::S_CBRANCH_SCC1: return AMDGPU::S_CBRANCH_VCCNZ;
   }
   llvm_unreachable(
       "Unexpected scalar opcode without corresponding vector one!");
 }
 
+void SIInstrInfo::insertScratchExecCopy(MachineFunction &MF,
+                                        MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator MBBI,
+                                        const DebugLoc &DL, Register Reg,
+                                        bool IsSCCLive) const {
+  const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
+  const SIInstrInfo *TII = ST.getInstrInfo();
+  bool IsWave32 = ST.isWave32();
+  if (IsSCCLive) {
+    // Insert two move instructions, one to save the original value of EXEC and
+    // the other to turn on all bits in EXEC. This is required as we can't use
+    // the single instruction S_OR_SAVEEXEC that clobbers SCC.
+    unsigned MovOpc = IsWave32 ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
+    MCRegister Exec = IsWave32 ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
+    BuildMI(MBB, MBBI, DL, TII->get(MovOpc), Reg).addReg(Exec, RegState::Kill);
+    BuildMI(MBB, MBBI, DL, TII->get(MovOpc), Exec).addImm(-1);
+  } else {
+    const unsigned OrSaveExec =
+        IsWave32 ? AMDGPU::S_OR_SAVEEXEC_B32 : AMDGPU::S_OR_SAVEEXEC_B64;
+    auto SaveExec =
+        BuildMI(MBB, MBBI, DL, TII->get(OrSaveExec), Reg).addImm(-1);
+    SaveExec->getOperand(3).setIsDead(); // Mark SCC as dead.
+  }
+}
+
+void SIInstrInfo::restoreExec(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator MBBI,
+                              const DebugLoc &DL, Register Reg) const {
+  unsigned ExecMov = isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
+  MCRegister Exec = isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
+  BuildMI(MBB, MBBI, DL, get(ExecMov), Exec).addReg(Reg, RegState::Kill);
+}
+
 static const TargetRegisterClass *
 adjustAllocatableRegClass(const GCNSubtarget &ST, const SIRegisterInfo &RI,
                           const MachineRegisterInfo &MRI,
                           const MCInstrDesc &TID, unsigned RCID,
                           bool IsAllocatable) {
   if ((IsAllocatable || !ST.hasGFX90AInsts() || !MRI.reservedRegsFrozen()) &&
       (((TID.mayLoad() || TID.mayStore()) &&
         !(TID.TSFlags & SIInstrFlags::VGPRSpill)) ||

llvm/lib/Target/AMDGPU/SIInstructions.td

 defm SI_SPILL_AV256 : SI_SPILL_VGPR <AV_256, 1>;
 defm SI_SPILL_AV288 : SI_SPILL_VGPR <AV_288, 1>;
 defm SI_SPILL_AV320 : SI_SPILL_VGPR <AV_320, 1>;
 defm SI_SPILL_AV352 : SI_SPILL_VGPR <AV_352, 1>;
 defm SI_SPILL_AV384 : SI_SPILL_VGPR <AV_384, 1>;
 defm SI_SPILL_AV512 : SI_SPILL_VGPR <AV_512, 1>;
 defm SI_SPILL_AV1024 : SI_SPILL_VGPR <AV_1024, 1>;
 
+defm SI_SPILL_WWM_V32 : SI_SPILL_VGPR <VGPR_32>;
+
 def SI_PC_ADD_REL_OFFSET : SPseudoInstSI <
   (outs SReg_64:$dst),
   (ins si_ga:$ptr_lo, si_ga:$ptr_hi),
   [(set SReg_64:$dst,
       (i64 (SIpc_add_rel_offset tglobaladdr:$ptr_lo, tglobaladdr:$ptr_hi)))]> {
   let Defs = [SCC];
 }
 

llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp

if (HasSGPRSpillToVGPR) {
 
     // All those frame indices which are dead by now should be removed from the
     // function frame. Otherwise, there is a side effect such as re-mapping of
     // free frame index ids by the later pass(es) like "stack slot coloring"
     // which in turn could mess-up with the book keeping of "frame index to VGPR
     // lane".
     FuncInfo->removeDeadFrameIndices(MFI, /*ResetSGPRSpillStackIDs*/ false);
 
+    const TargetRegisterClass *RC = TRI->getWaveMaskRegClass();
+    // Shift back the reserved SGPR for EXEC copy into the lowest range.
+    // This SGPR is reserved to handle the whole-wave spill/copy operations
+    // that might get inserted during vgpr regalloc.
+    Register UnusedLowSGPR = TRI->findUnusedRegister(MRI, RC, MF);
+    if (UnusedLowSGPR && TRI->getHWRegIndex(UnusedLowSGPR) <
+                             TRI->getHWRegIndex(FuncInfo->getSGPRForEXECCopy()))
+      FuncInfo->setSGPRForEXECCopy(UnusedLowSGPR);
+
     MadeChange = true;
+  } else {
+    // No SGPR spills and hence there won't be any WWM spills/copies. Reset the
+    // SGPR reserved for EXEC copy.
+    FuncInfo->setSGPRForEXECCopy(AMDGPU::NoRegister);
   }
 
   SaveBlocks.clear();
   RestoreBlocks.clear();
 
   // Updated the reserved registers with any VGPRs added for SGPR spills.
   if (NewReservedRegs)
     MRI.freezeReservedRegs(MF);
 
   return MadeChange;
 }

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h

struct SIMachineFunctionInfo final : public yaml::MachineFunctionInfo {
   std::optional<SIArgumentInfo> ArgInfo;
 
   unsigned PSInputAddr = 0;
   unsigned PSInputEnable = 0;
 
   SIMode Mode;
   std::optional<FrameIndex> ScavengeFI;
   StringValue VGPRForAGPRCopy;
+  StringValue SGPRForEXECCopy;
 
   SIMachineFunctionInfo() = default;
   SIMachineFunctionInfo(const llvm::SIMachineFunctionInfo &,
                         const TargetRegisterInfo &TRI,
                         const llvm::MachineFunction &MF);
 
   void mappingImpl(yaml::IO &YamlIO) override;
   ~SIMachineFunctionInfo() = default;
static void mapping(IO &YamlIO, SIMachineFunctionInfo &MFI) {
     YamlIO.mapOptional("mode", MFI.Mode, SIMode());
     YamlIO.mapOptional("highBitsOf32BitAddress",
                        MFI.HighBitsOf32BitAddress, 0u);
     YamlIO.mapOptional("occupancy", MFI.Occupancy, 0);
     YamlIO.mapOptional("wwmReservedRegs", MFI.WWMReservedRegs);
     YamlIO.mapOptional("scavengeFI", MFI.ScavengeFI);
     YamlIO.mapOptional("vgprForAGPRCopy", MFI.VGPRForAGPRCopy,
                        StringValue()); // Don't print out when it's empty.
+    YamlIO.mapOptional("sgprForEXECCopy", MFI.SGPRForEXECCopy,
+                       StringValue()); // Don't print out when it's empty.
   }
 };
 
 } // end namespace yaml
 
 // A CSR SGPR value can be preserved inside a callee using one of the following
 // methods.
 //   1. Copy to an unused scratch SGPR.
PrologEpilogSGPRSaveRestoreInfo(SGPRSaveKind K, Register R)
       : Kind(K), Reg(R) {}
   Register getReg() const { return Reg; }
   int getIndex() const { return Index; }
   SGPRSaveKind getKind() const { return Kind; }
 };
 
 /// This class keeps track of the SPI_SP_INPUT_ADDR config register, which
 /// tells the hardware which interpolation parameters to load.
-class SIMachineFunctionInfo final : public AMDGPUMachineFunction {
+class SIMachineFunctionInfo final : public AMDGPUMachineFunction,
+                                    private MachineRegisterInfo::Delegate {
   friend class GCNTargetMachine;
 
   // State of MODE register, assumed FP mode.
   SIModeRegisterDefaults Mode;
 
   // Registers that may be reserved for spilling purposes. These may be the same
   // as the input registers.
   Register ScratchRSrcReg = AMDGPU::PRIVATE_RSRC_REG;
private:
 
   // The hard-wired high half of the address of the global information table
   // for AMDPAL OS type. 0xffffffff represents no hard-wired high half, since
   // current hardware only allows a 16 bit value.
   unsigned GITPtrHigh;
 
   unsigned HighBitsOf32BitAddress;
 
+  // Flags associated with the virtual registers.
+  IndexedMap<uint8_t, VirtReg2IndexFunctor> VRegFlags;
+
   // Current recorded maximum possible occupancy.
   unsigned Occupancy;
 
   mutable std::optional<bool> UsesAGPRs;
 
   MCPhysReg getNextUserSGPR() const;
 
   MCPhysReg getNextSystemSGPR() const;
 
+  // MachineRegisterInfo callback functions to notify events.
+  void MRI_NoteNewVirtualRegister(Register Reg) override;
+  void MRI_NoteCloneVirtualRegister(Register NewReg, Register SrcReg) override;
+
 public:
   struct VGPRSpillToAGPR {
     SmallVector<MCPhysReg, 32> Lanes;
     bool FullyAllocated = false;
     bool IsDead = false;
   };
 
 private:
private:
   using PrologEpilogSGPRSpillsMap =
       DenseMap<Register, PrologEpilogSGPRSaveRestoreInfo>;
   // To track the SGPR spill method used for a CSR SGPR register during
   // frame lowering. Even though the SGPR spills are handled during
   // SILowerSGPRSpills pass, some special handling needed later during the
   // PrologEpilogInserter.
   PrologEpilogSGPRSpillsMap PrologEpilogSGPRSpills;
 
+  // To save/restore EXEC MASK around WWM spills and copies.
+  Register SGPRForEXECCopy;
+
   DenseMap<int, VGPRSpillToAGPR> VGPRToAGPRSpills;
 
   // AGPRs used for VGPR spills.
   SmallVector<MCPhysReg, 32> SpillAGPR;
 
   // VGPRs used for AGPR spills.
   SmallVector<MCPhysReg, 32> SpillVGPR;
 
public:
   ArrayRef<SIRegisterInfo::SpilledReg>
   getPrologEpilogSGPRSpillToVGPRLanes(int FrameIndex) const {
     auto I = PrologEpilogSGPRSpillToVGPRLanes.find(FrameIndex);
     return (I == PrologEpilogSGPRSpillToVGPRLanes.end())
                ? ArrayRef<SIRegisterInfo::SpilledReg>()
                : ArrayRef(I->second);
   }
 
+  void setFlag(Register Reg, uint8_t Flag) {
+    assert(Reg.isVirtual());
+    if (VRegFlags.inBounds(Reg))
+      VRegFlags[Reg] |= Flag;
+  }
+
+  bool checkFlag(Register Reg, uint8_t Flag) const {
+    if (Reg.isPhysical())
+      return false;
+
+    return VRegFlags.inBounds(Reg) && VRegFlags[Reg] & Flag;
+  }
+
   void allocateWWMSpill(MachineFunction &MF, Register VGPR, uint64_t Size = 4,
                         Align Alignment = Align(4));
 
   void splitWWMSpillRegisters(
       MachineFunction &MF,
       SmallVectorImpl<std::pair<Register, int>> &CalleeSavedRegs,
       SmallVectorImpl<std::pair<Register, int>> &ScratchRegs) const;
 
   ArrayRef<MCPhysReg> getAGPRSpillVGPRs() const {
     return SpillAGPR;
   }
 
+  Register getSGPRForEXECCopy() const { return SGPRForEXECCopy; }
+
+  void setSGPRForEXECCopy(Register Reg) { SGPRForEXECCopy = Reg; }
+
   ArrayRef<MCPhysReg> getVGPRSpillAGPRs() const {
     return SpillVGPR;
   }
 
   MCPhysReg getVGPRToAGPRSpill(int FrameIndex, unsigned Lane) const {
     auto I = VGPRToAGPRSpills.find(FrameIndex);
     return (I == VGPRToAGPRSpills.end()) ? (MCPhysReg)AMDGPU::NoRegister
                                          : I->second.Lanes[Lane];

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp

: AMDGPUMachineFunction(F, *STI),
     HighBitsOf32BitAddress(0) {
   const GCNSubtarget &ST = *static_cast<const GCNSubtarget *>(STI);
   FlatWorkGroupSizes = ST.getFlatWorkGroupSizes(F);
   WavesPerEU = ST.getWavesPerEU(F);
 
   Occupancy = ST.computeOccupancy(F, getLDSSize());
   CallingConv::ID CC = F.getCallingConv();
 
+  VRegFlags.reserve(1024);
+
   // FIXME: Should have analysis or something rather than attribute to detect
   // calls.
   const bool HasCalls = F.hasFnAttribute("amdgpu-calls");
 
   const bool IsKernel = CC == CallingConv::AMDGPU_KERNEL ||
                         CC == CallingConv::SPIR_KERNEL;
 
   if (IsKernel) {
MCPhysReg SIMachineFunctionInfo::getNextUserSGPR() const {
   assert(NumSystemSGPRs == 0 && "System SGPRs must be added after user SGPRs");
   return AMDGPU::SGPR0 + NumUserSGPRs;
 }
 
 MCPhysReg SIMachineFunctionInfo::getNextSystemSGPR() const {
   return AMDGPU::SGPR0 + NumUserSGPRs + NumSystemSGPRs;
 }
 
+void SIMachineFunctionInfo::MRI_NoteNewVirtualRegister(Register Reg) {
+  VRegFlags.grow(Reg);
+}
+
+void SIMachineFunctionInfo::MRI_NoteCloneVirtualRegister(Register NewReg,
+                                                         Register SrcReg) {
+  VRegFlags.grow(NewReg);
+  VRegFlags[NewReg] = VRegFlags[SrcReg];
+}
+
 Register
 SIMachineFunctionInfo::getGITPtrLoReg(const MachineFunction &MF) const {
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   if (!ST.isAmdPalOS())
     return Register();
   Register GitPtrLo = AMDGPU::SGPR0; // Low GIT address passed in
   if (ST.hasMergedShaders()) {
     switch (MF.getFunction().getCallingConv()) {
: ExplicitKernArgSize(MFI.getExplicitKernArgSize()),
       PSInputAddr(MFI.getPSInputAddr()),
       PSInputEnable(MFI.getPSInputEnable()),
       Mode(MFI.getMode()) {
   for (Register Reg : MFI.getWWMReservedRegs())
     WWMReservedRegs.push_back(regToString(Reg, TRI));
 
   if (MFI.getVGPRForAGPRCopy())
     VGPRForAGPRCopy = regToString(MFI.getVGPRForAGPRCopy(), TRI);
+
+  if (MFI.getSGPRForEXECCopy())
+    SGPRForEXECCopy = regToString(MFI.getSGPRForEXECCopy(), TRI);
+
   auto SFI = MFI.getOptionalScavengeFI();
   if (SFI)
     ScavengeFI = yaml::FrameIndex(*SFI, MF.getFrameInfo());
 }
 
 void yaml::SIMachineFunctionInfo::mappingImpl(yaml::IO &YamlIO) {
   MappingTraits<SIMachineFunctionInfo>::mapping(YamlIO, *this);
 }

llvm/lib/Target/AMDGPU/SIRegisterInfo.h

public:
   /// \returns the sub reg enum value for the given \p Channel
   /// (e.g. getSubRegFromChannel(0) -> AMDGPU::sub0)
   static unsigned getSubRegFromChannel(unsigned Channel, unsigned NumRegs = 1);
 
   bool spillSGPRToVGPR() const {
     return SpillSGPRToVGPR;
   }
 
+  /// Return the largest available SGPR aligned to \p Align for the register
+  /// class \p RC.
+  MCRegister getAlignedHighSGPRForRC(const MachineFunction &MF,
+                                     const unsigned Align,
+                                     const TargetRegisterClass *RC) const;
+
   /// Return the end register initially reserved for the scratch buffer in case
   /// spilling is needed.
   MCRegister reservedPrivateSegmentBufferReg(const MachineFunction &MF) const;
 
   BitVector getReservedRegs(const MachineFunction &MF) const override;
   bool isAsmClobberable(const MachineFunction &MF,
                         MCRegister PhysReg) const override;
 

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

unsigned SIRegisterInfo::getSubRegFromChannel(unsigned Channel,
                                               unsigned NumRegs) {
   assert(NumRegs < SubRegFromChannelTableWidthMap.size());
   unsigned NumRegIndex = SubRegFromChannelTableWidthMap[NumRegs];
   assert(NumRegIndex && "Not implemented");
   assert(Channel < SubRegFromChannelTable[NumRegIndex - 1].size());
   return SubRegFromChannelTable[NumRegIndex - 1][Channel];
 }
 
+MCRegister
+SIRegisterInfo::getAlignedHighSGPRForRC(const MachineFunction &MF,
+                                        const unsigned Align,
+                                        const TargetRegisterClass *RC) const {
+  unsigned BaseIdx = alignDown(ST.getMaxNumSGPRs(MF), Align) - Align;
+  MCRegister BaseReg(AMDGPU::SGPR_32RegClass.getRegister(BaseIdx));
+  return getMatchingSuperReg(BaseReg, AMDGPU::sub0, RC);
+}
+
 MCRegister SIRegisterInfo::reservedPrivateSegmentBufferReg(
   const MachineFunction &MF) const {
-  unsigned BaseIdx = alignDown(ST.getMaxNumSGPRs(MF), 4) - 4;
-  MCRegister BaseReg(AMDGPU::SGPR_32RegClass.getRegister(BaseIdx));
-  return getMatchingSuperReg(BaseReg, AMDGPU::sub0, &AMDGPU::SGPR_128RegClass);
+  return getAlignedHighSGPRForRC(MF, /*Align=*/4, &AMDGPU::SGPR_128RegClass);
 }
 
 BitVector SIRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
   Reserved.set(AMDGPU::MODE);
 
   const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
 
BitVector SIRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   }
 
   if (hasBasePointer(MF)) {
     MCRegister BasePtrReg = getBaseRegister();
     reserveRegisterTuples(Reserved, BasePtrReg);
     assert(!isSubRegister(ScratchRSrcReg, BasePtrReg));
   }
 
+  // FIXME: Use same reserved register introduced in D149775
+  // SGPR used to preserve EXEC MASK around WWM spill/copy instructions.
+  Register ExecCopyReg = MFI->getSGPRForEXECCopy();
+  if (ExecCopyReg)
+    reserveRegisterTuples(Reserved, ExecCopyReg);
+
   // Reserve VGPRs/AGPRs.
   //
   unsigned MaxNumVGPRs = ST.getMaxNumVGPRs(MF);
   unsigned MaxNumAGPRs = MaxNumVGPRs;
   unsigned TotalNumVGPRs = AMDGPU::VGPR_32RegClass.getNumRegs();
 
   // On GFX90A, the number of VGPRs and AGPRs need not be equal. Theoretically,
   // a wave may have up to 512 total vector registers combining together both
static unsigned getNumSubRegsForSpillOp(unsigned Op) {
   case AMDGPU::SI_SPILL_S32_SAVE:
   case AMDGPU::SI_SPILL_S32_RESTORE:
   case AMDGPU::SI_SPILL_V32_SAVE:
   case AMDGPU::SI_SPILL_V32_RESTORE:
   case AMDGPU::SI_SPILL_A32_SAVE:
   case AMDGPU::SI_SPILL_A32_RESTORE:
   case AMDGPU::SI_SPILL_AV32_SAVE:
   case AMDGPU::SI_SPILL_AV32_RESTORE:
+  case AMDGPU::SI_SPILL_WWM_V32_SAVE:
+  case AMDGPU::SI_SPILL_WWM_V32_RESTORE:
     return 1;
   default: llvm_unreachable("Invalid spill opcode");
   }
 }
 
 static int getOffsetMUBUFStore(unsigned Opc) {
   switch (Opc) {
   case AMDGPU::BUFFER_STORE_DWORD_OFFEN:
switch (MI->getOpcode()) {
     case AMDGPU::SI_SPILL_AV288_SAVE:
     case AMDGPU::SI_SPILL_AV256_SAVE:
     case AMDGPU::SI_SPILL_AV224_SAVE:
     case AMDGPU::SI_SPILL_AV192_SAVE:
     case AMDGPU::SI_SPILL_AV160_SAVE:
     case AMDGPU::SI_SPILL_AV128_SAVE:
     case AMDGPU::SI_SPILL_AV96_SAVE:
     case AMDGPU::SI_SPILL_AV64_SAVE:
-    case AMDGPU::SI_SPILL_AV32_SAVE: {
+    case AMDGPU::SI_SPILL_AV32_SAVE:
+    case AMDGPU::SI_SPILL_WWM_V32_SAVE: {
       const MachineOperand *VData = TII->getNamedOperand(*MI,
                                                          AMDGPU::OpName::vdata);
       assert(TII->getNamedOperand(*MI, AMDGPU::OpName::soffset)->getReg() ==
              MFI->getStackPtrOffsetReg());
 
       unsigned Opc = ST.enableFlatScratch() ? AMDGPU::SCRATCH_STORE_DWORD_SADDR
                                             : AMDGPU::BUFFER_STORE_DWORD_OFFSET;
       auto *MBB = MI->getParent();
+      bool IsWWMRegSpill = TII->isWWMRegSpillOpcode(MI->getOpcode());
+      if (IsWWMRegSpill) {
+      TII->insertScratchExecCopy(*MF, *MBB, MI, DL, MFI->getSGPRForEXECCopy(),
+                                 RS->isRegUsed(AMDGPU::SCC));
+      }
       buildSpillLoadStore(
           *MBB, MI, DL, Opc, Index, VData->getReg(), VData->isKill(), FrameReg,
           TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(),
           *MI->memoperands_begin(), RS);
       MFI->addToSpilledVGPRs(getNumSubRegsForSpillOp(MI->getOpcode()));
+      if (IsWWMRegSpill)
+      TII->restoreExec(*MF, *MBB, MI, DL, MFI->getSGPRForEXECCopy());
+
       MI->eraseFromParent();
       return true;
     }
     case AMDGPU::SI_SPILL_V32_RESTORE:
     case AMDGPU::SI_SPILL_V64_RESTORE:
     case AMDGPU::SI_SPILL_V96_RESTORE:
     case AMDGPU::SI_SPILL_V128_RESTORE:
     case AMDGPU::SI_SPILL_V160_RESTORE:
switch (MI->getOpcode()) {
     case AMDGPU::SI_SPILL_AV192_RESTORE:
     case AMDGPU::SI_SPILL_AV224_RESTORE:
     case AMDGPU::SI_SPILL_AV256_RESTORE:
     case AMDGPU::SI_SPILL_AV288_RESTORE:
     case AMDGPU::SI_SPILL_AV320_RESTORE:
     case AMDGPU::SI_SPILL_AV352_RESTORE:
     case AMDGPU::SI_SPILL_AV384_RESTORE:
     case AMDGPU::SI_SPILL_AV512_RESTORE:
-    case AMDGPU::SI_SPILL_AV1024_RESTORE: {
+    case AMDGPU::SI_SPILL_AV1024_RESTORE:
+    case AMDGPU::SI_SPILL_WWM_V32_RESTORE: {
       const MachineOperand *VData = TII->getNamedOperand(*MI,
                                                          AMDGPU::OpName::vdata);
       assert(TII->getNamedOperand(*MI, AMDGPU::OpName::soffset)->getReg() ==
              MFI->getStackPtrOffsetReg());
 
       unsigned Opc = ST.enableFlatScratch() ? AMDGPU::SCRATCH_LOAD_DWORD_SADDR
                                             : AMDGPU::BUFFER_LOAD_DWORD_OFFSET;
       auto *MBB = MI->getParent();
+      bool IsWWMRegSpill = TII->isWWMRegSpillOpcode(MI->getOpcode());
+      if (IsWWMRegSpill) {
+      TII->insertScratchExecCopy(*MF, *MBB, MI, DL, MFI->getSGPRForEXECCopy(),
+                                 RS->isRegUsed(AMDGPU::SCC));
+      }
       buildSpillLoadStore(
           *MBB, MI, DL, Opc, Index, VData->getReg(), VData->isKill(), FrameReg,
           TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(),
           *MI->memoperands_begin(), RS);
+      if (IsWWMRegSpill) {
+      TII->restoreExec(*MF, *MBB, MI, DL, MFI->getSGPRForEXECCopy());
+      }
       MI->eraseFromParent();
       return true;
     }
 
     default: {
       // Other access to frame index
       const DebugLoc &DL = MI->getDebugLoc();
 

llvm/test/CodeGen/AMDGPU/bf16.ll

 entry:
   ret <16 x bfloat> %in
 }
 
 define void @test_call(bfloat %in, ptr addrspace(5) %out) {
 ; GCN-LABEL: test_call:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s33
+; GCN-NEXT:    s_mov_b32 s10, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NEXT:    v_writelane_b32 v2, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v2, s31, 1
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, test_arg_store@gotpcrel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, test_arg_store@gotpcrel32@hi+12
 ; GCN-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GCN-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GCN-NEXT:    buffer_store_short v0, v1, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s31, v2, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v2, 0
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s8
+; GCN-NEXT:    s_mov_b32 s33, s10
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-LABEL: test_call:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    s_mov_b32 s8, s33
+; GFX7-NEXT:    s_mov_b32 s10, s33
 ; GFX7-NEXT:    s_mov_b32 s33, s32
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0x400
 ; GFX7-NEXT:    s_getpc_b64 s[4:5]
 ; GFX7-NEXT:    s_add_u32 s4, s4, test_arg_store@gotpcrel32@lo+4
 ; GFX7-NEXT:    s_addc_u32 s5, s5, test_arg_store@gotpcrel32@hi+12
 ; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX7-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX7-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX7-NEXT:    buffer_store_short v0, v1, s[0:3], 0 offen
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX7-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX7-NEXT:    s_mov_b32 s33, s8
+; GFX7-NEXT:    s_mov_b32 s33, s10
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: test_call:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT:    s_mov_b32 s6, s33
+; GFX8-NEXT:    s_mov_b32 s8, s33
 ; GFX8-NEXT:    s_mov_b32 s33, s32
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0x400
 ; GFX8-NEXT:    s_getpc_b64 s[4:5]
 ; GFX8-NEXT:    s_add_u32 s4, s4, test_arg_store@gotpcrel32@lo+4
 ; GFX8-NEXT:    s_addc_u32 s5, s5, test_arg_store@gotpcrel32@hi+12
 ; GFX8-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX8-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX8-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX8-NEXT:    buffer_store_short v0, v1, s[0:3], 0 offen
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX8-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX8-NEXT:    s_mov_b32 s33, s6
+; GFX8-NEXT:    s_mov_b32 s33, s8
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: test_call:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0x400
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, test_arg_store@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, test_arg_store@gotpcrel32@hi+12
 ; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX9-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX9-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX9-NEXT:    buffer_store_short_d16_hi v0, v1, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: test_call:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0x200
 ; GFX10-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-NEXT:    s_add_u32 s4, s4, test_arg_store@gotpcrel32@lo+4
 ; GFX10-NEXT:    s_addc_u32 s5, s5, test_arg_store@gotpcrel32@hi+12
 ; GFX10-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX10-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX10-NEXT:    buffer_store_short_d16_hi v0, v1, s[0:3], 0 offen
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfe00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %result = call bfloat @test_arg_store(bfloat %in)
   store volatile bfloat %result, ptr addrspace(5) %out
   ret void
 }
 
 define void @test_call_v2bf16(<2 x bfloat> %in, ptr addrspace(5) %out) {
 ; GCN-LABEL: test_call_v2bf16:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s33
+; GCN-NEXT:    s_mov_b32 s10, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NEXT:    v_writelane_b32 v3, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v3, s31, 1

 ; GCN-NEXT:    buffer_store_short v0, v2, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s31, v3, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v3, 0
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s8
+; GCN-NEXT:    s_mov_b32 s33, s10
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-LABEL: test_call_v2bf16:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    s_mov_b32 s8, s33
+; GFX7-NEXT:    s_mov_b32 s10, s33
 ; GFX7-NEXT:    s_mov_b32 s33, s32
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0x400
 ; GFX7-NEXT:    s_getpc_b64 s[4:5]
 ; GFX7-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX7-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX7-NEXT:    buffer_store_short v0, v2, s[0:3], 0 offen
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX7-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX7-NEXT:    s_mov_b32 s33, s8
+; GFX7-NEXT:    s_mov_b32 s33, s10
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: test_call_v2bf16:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT:    s_mov_b32 s6, s33
+; GFX8-NEXT:    s_mov_b32 s8, s33
 ; GFX8-NEXT:    s_mov_b32 s33, s32
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0x400
 ; GFX8-NEXT:    s_getpc_b64 s[4:5]
 ; GFX8-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX8-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12
 ; GFX8-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX8-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX8-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX8-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX8-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX8-NEXT:    s_mov_b32 s33, s6
+; GFX8-NEXT:    s_mov_b32 s33, s8
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: test_call_v2bf16:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0x400
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12
 ; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX9-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX9-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX9-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: test_call_v2bf16:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0x200
 ; GFX10-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX10-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12
 ; GFX10-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX10-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
 ; GFX10-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfe00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %result = call <2 x bfloat> @test_arg_store_v2bf16(<2 x bfloat> %in)
   store volatile <2 x bfloat> %result, ptr addrspace(5) %out
   ret void
 }
 
 define void @test_call_v3bf16(<3 x bfloat> %in, ptr addrspace(5) %out) {
 ; GCN-LABEL: test_call_v3bf16:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s33
+; GCN-NEXT:    s_mov_b32 s10, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v4, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NEXT:    v_writelane_b32 v4, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v4, s31, 1

 ; GCN-NEXT:    buffer_store_dword v0, v3, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s31, v4, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v4, 0
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v4, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s8
+; GCN-NEXT:    s_mov_b32 s33, s10
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-LABEL: test_call_v3bf16:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    s_mov_b32 s8, s33
+; GFX7-NEXT:    s_mov_b32 s10, s33
 ; GFX7-NEXT:    s_mov_b32 s33, s32
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_store_dword v4, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0x400
 ; GFX7-NEXT:    s_getpc_b64 s[4:5]
 ; GFX7-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX7-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX7-NEXT:    buffer_store_dword v0, v3, s[0:3], 0 offen
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_readlane_b32 s31, v4, 1
 ; GFX7-NEXT:    v_readlane_b32 s30, v4, 0
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_load_dword v4, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX7-NEXT:    s_mov_b32 s33, s8
+; GFX7-NEXT:    s_mov_b32 s33, s10
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: test_call_v3bf16:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT:    s_mov_b32 s6, s33
+; GFX8-NEXT:    s_mov_b32 s8, s33
 ; GFX8-NEXT:    s_mov_b32 s33, s32
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0x400
 ; GFX8-NEXT:    s_getpc_b64 s[4:5]
 ; GFX8-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX8-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX8-NEXT:    buffer_store_dword v0, v2, s[0:3], 0 offen
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX8-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX8-NEXT:    s_mov_b32 s33, s6
+; GFX8-NEXT:    s_mov_b32 s33, s8
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: test_call_v3bf16:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0x400
 ; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff0000, v0
 ; GFX9-NEXT:    s_mov_b32 s4, 0xffff
 ; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v4

 ; GFX9-NEXT:    buffer_store_dword v0, v2, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: test_call_v3bf16:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0x200
 ; GFX10-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfe00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %result = call <3 x bfloat> @test_arg_store_v2bf16(<3 x bfloat> %in)
   store volatile <3 x bfloat> %result, ptr addrspace(5) %out
   ret void
 }
 
 define void @test_call_v4bf16(<4 x bfloat> %in, ptr addrspace(5) %out) {
 ; GCN-LABEL: test_call_v4bf16:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s33
+; GCN-NEXT:    s_mov_b32 s10, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v5, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NEXT:    v_writelane_b32 v5, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v5, s31, 1

 ; GCN-NEXT:    buffer_store_short v0, v4, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s31, v5, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v5, 0
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v5, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s8
+; GCN-NEXT:    s_mov_b32 s33, s10
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-LABEL: test_call_v4bf16:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    s_mov_b32 s8, s33
+; GFX7-NEXT:    s_mov_b32 s10, s33
 ; GFX7-NEXT:    s_mov_b32 s33, s32
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_store_dword v5, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0x400
 ; GFX7-NEXT:    s_getpc_b64 s[4:5]
 ; GFX7-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX7-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX7-NEXT:    buffer_store_short v0, v4, s[0:3], 0 offen
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_readlane_b32 s31, v5, 1
 ; GFX7-NEXT:    v_readlane_b32 s30, v5, 0
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_load_dword v5, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX7-NEXT:    s_mov_b32 s33, s8
+; GFX7-NEXT:    s_mov_b32 s33, s10
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: test_call_v4bf16:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT:    s_mov_b32 s6, s33
+; GFX8-NEXT:    s_mov_b32 s8, s33
 ; GFX8-NEXT:    s_mov_b32 s33, s32
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0x400
 ; GFX8-NEXT:    s_getpc_b64 s[4:5]
 ; GFX8-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX8-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX8-NEXT:    buffer_store_short v4, v0, s[0:3], 0 offen
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX8-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX8-NEXT:    s_mov_b32 s33, s6
+; GFX8-NEXT:    s_mov_b32 s33, s8
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: test_call_v4bf16:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0x400
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX9-NEXT:    buffer_store_short v0, v2, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: test_call_v4bf16:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v3, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0x200
 ; GFX10-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_readlane_b32 s31, v3, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v3, 0
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v3, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfe00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %result = call <4 x bfloat> @test_arg_store_v2bf16(<4 x bfloat> %in)
   store volatile <4 x bfloat> %result, ptr addrspace(5) %out
   ret void
 }
 
 define void @test_call_v8bf16(<8 x bfloat> %in, ptr addrspace(5) %out) {
 ; GCN-LABEL: test_call_v8bf16:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s33
+; GCN-NEXT:    s_mov_b32 s10, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v9, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NEXT:    v_writelane_b32 v9, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v9, s31, 1

 ; GCN-NEXT:    buffer_store_short v0, v8, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s31, v9, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v9, 0
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v9, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s8
+; GCN-NEXT:    s_mov_b32 s33, s10
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-LABEL: test_call_v8bf16:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    s_mov_b32 s8, s33
+; GFX7-NEXT:    s_mov_b32 s10, s33
 ; GFX7-NEXT:    s_mov_b32 s33, s32
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_store_dword v9, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0x400
 ; GFX7-NEXT:    s_getpc_b64 s[4:5]
 ; GFX7-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX7-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX7-NEXT:    buffer_store_short v0, v8, s[0:3], 0 offen
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_readlane_b32 s31, v9, 1
 ; GFX7-NEXT:    v_readlane_b32 s30, v9, 0
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_load_dword v9, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX7-NEXT:    s_mov_b32 s33, s8
+; GFX7-NEXT:    s_mov_b32 s33, s10
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: test_call_v8bf16:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT:    s_mov_b32 s6, s33
+; GFX8-NEXT:    s_mov_b32 s8, s33
 ; GFX8-NEXT:    s_mov_b32 s33, s32
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_store_dword v5, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0x400
 ; GFX8-NEXT:    s_getpc_b64 s[4:5]
 ; GFX8-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX8-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX8-NEXT:    buffer_store_short v6, v0, s[0:3], 0 offen
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_readlane_b32 s31, v5, 1
 ; GFX8-NEXT:    v_readlane_b32 s30, v5, 0
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_load_dword v5, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX8-NEXT:    s_mov_b32 s33, s6
+; GFX8-NEXT:    s_mov_b32 s33, s8
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: test_call_v8bf16:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v5, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0x400
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX9-NEXT:    buffer_store_short v0, v4, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_readlane_b32 s31, v5, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v5, 0
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v5, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: test_call_v8bf16:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v5, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0x200
 ; GFX10-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_readlane_b32 s31, v5, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v5, 0
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v5, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfe00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %result = call <8 x bfloat> @test_arg_store_v2bf16(<8 x bfloat> %in)
   store volatile <8 x bfloat> %result, ptr addrspace(5) %out
   ret void
 }
 
 define void @test_call_v16bf16(<16 x bfloat> %in, ptr addrspace(5) %out) {
 ; GCN-LABEL: test_call_v16bf16:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s33
+; GCN-NEXT:    s_mov_b32 s10, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v17, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NEXT:    v_writelane_b32 v17, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v17, s31, 1

 ; GCN-NEXT:    buffer_store_short v0, v16, s[0:3], 0 offen
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s31, v17, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v17, 0
 ; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v17, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s8
+; GCN-NEXT:    s_mov_b32 s33, s10
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-LABEL: test_call_v16bf16:
 ; GFX7:       ; %bb.0: ; %entry
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT:    s_mov_b32 s8, s33
+; GFX7-NEXT:    s_mov_b32 s10, s33
 ; GFX7-NEXT:    s_mov_b32 s33, s32
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_store_dword v17, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0x400
 ; GFX7-NEXT:    s_getpc_b64 s[4:5]
 ; GFX7-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX7-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX7-NEXT:    buffer_store_short v0, v16, s[0:3], 0 offen
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_readlane_b32 s31, v17, 1
 ; GFX7-NEXT:    v_readlane_b32 s30, v17, 0
 ; GFX7-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX7-NEXT:    buffer_load_dword v17, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX7-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX7-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX7-NEXT:    s_mov_b32 s33, s8
+; GFX7-NEXT:    s_mov_b32 s33, s10
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX8-LABEL: test_call_v16bf16:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT:    s_mov_b32 s6, s33
+; GFX8-NEXT:    s_mov_b32 s8, s33
 ; GFX8-NEXT:    s_mov_b32 s33, s32
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_store_dword v9, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0x400
 ; GFX8-NEXT:    s_getpc_b64 s[4:5]
 ; GFX8-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX8-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX8-NEXT:    buffer_store_short v10, v0, s[0:3], 0 offen
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_readlane_b32 s31, v9, 1
 ; GFX8-NEXT:    v_readlane_b32 s30, v9, 0
 ; GFX8-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX8-NEXT:    buffer_load_dword v9, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX8-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX8-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX8-NEXT:    s_mov_b32 s33, s6
+; GFX8-NEXT:    s_mov_b32 s33, s8
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: test_call_v16bf16:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v9, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0x400
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, test_arg_store_v2bf16@gotpcrel32@hi+12

 ; GFX9-NEXT:    buffer_store_short v0, v8, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_readlane_b32 s31, v9, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v9, 0
 ; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v9, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: test_call_v16bf16:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v9, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0x200
 ; GFX10-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-NEXT:    s_add_u32 s4, s4, test_arg_store_v2bf16@gotpcrel32@lo+4

 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_readlane_b32 s31, v9, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v9, 0
 ; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v9, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfe00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %result = call <16 x bfloat> @test_arg_store_v2bf16(<16 x bfloat> %in)
   store volatile <16 x bfloat> %result, ptr addrspace(5) %out
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll

define void @callee_need_to_spill_fp_to_memory() #3 {
   ret void
 }
 
 ; If we have a reserved VGPR that can be used for SGPR spills, we may still
 ; need to spill the FP to memory if there are no free lanes in the reserved
 ; VGPR.
 ; GCN-LABEL: {{^}}callee_need_to_spill_fp_to_memory_full_reserved_vgpr:
 ; MUBUF:   s_mov_b32 [[FP_SCRATCH_COPY:s[0-9]+]], s33
-; FLATSCR: s_mov_b32 s33, s0
-; MUBUF:   s_mov_b32 s33, s32
+; GCN:     s_mov_b32 s33, s32
 ; MUBUF:   s_xor_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}}
 ; MUBUF:   s_mov_b64 exec, [[COPY_EXEC1]]
 ; MUBUF:   v_mov_b32_e32 [[TMP_VGPR1:v[0-9]+]], [[FP_SCRATCH_COPY]]
 ; MUBUF:   buffer_store_dword [[TMP_VGPR1]], off, s[0:3], s33 offset:[[OFF:[0-9]+]]
 ; GCN-NOT: v_writelane_b32 v40, s33
 ; GCN-NOT: v_readlane_b32 s33, v40
 ; GCN-NOT: v_readlane_b32 s33, v40
 ; MUBUF:   buffer_load_dword [[TMP_VGPR2:v[0-9]+]], off, s[0:3], s33 offset:[[OFF]]
define void @callee_need_to_spill_fp_to_memory_full_reserved_vgpr() #3 {
   ret void
 }
 
 ; When flat-scratch is enabled, we save the FP to s0. At the same time,
 ; the exec register is saved to s0 when saving CSR in the function prolog.
 ; Make sure that the FP save happens after restoring exec from the same
 ; register.
 ; GCN-LABEL: {{^}}callee_need_to_spill_fp_to_reg:
-; FLATSCR: s_mov_b32 s0, s33
+; FLATSCR: s_mov_b32 s2, s33
 ; FLATSCR: s_mov_b32 s33, s32
 ; GCN-NOT: v_writelane_b32 v40, s33
-; FLATSCR: s_or_saveexec_b64 s[2:3], -1
-; FLATSCR: s_mov_b64 exec, s[2:3]
-; FLATSCR: s_or_saveexec_b64 s[2:3], -1
+; FLATSCR: s_or_saveexec_b64 s[4:5], -1
+; FLATSCR: s_mov_b64 exec, s[4:5]
+; FLATSCR: s_or_saveexec_b64 s[4:5], -1
 ; GCN-NOT: v_readlane_b32 s33, v40
-; FLATSCR: s_mov_b32 s33, s0
+; FLATSCR: s_mov_b32 s33, s2
 ; GCN:     s_setpc_b64
 define void @callee_need_to_spill_fp_to_reg() #1 {
   call void asm sideeffect "; clobber nonpreserved SGPRs and 64 CSRs",
     "~{s4},~{s5},~{s6},~{s7},~{s8},~{s9}
     ,~{s10},~{s11},~{s12},~{s13},~{s14},~{s15},~{s16},~{s17},~{s18},~{s19}
     ,~{s20},~{s21},~{s22},~{s23},~{s24},~{s25},~{s26},~{s27},~{s28},~{s29}
     ,~{s40},~{s41},~{s42},~{s43},~{s44},~{s45},~{s46},~{s47},~{s48},~{s49}
     ,~{s50},~{s51},~{s52},~{s53},~{s54},~{s55},~{s56},~{s57},~{s58},~{s59}

llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=SDAG -enable-var-scope %s
 ; RUN: llc -global-isel -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GISEL -enable-var-scope %s
 
 declare void @extern_c_func()
 
 define amdgpu_gfx void @gfx_func() {
 ; SDAG-LABEL: gfx_func:
 ; SDAG:       ; %bb.0:
 ; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-NEXT:    s_mov_b32 s36, s33
+; SDAG-NEXT:    s_mov_b32 s38, s33
 ; SDAG-NEXT:    s_mov_b32 s33, s32
 ; SDAG-NEXT:    s_or_saveexec_b64 s[34:35], -1
 ; SDAG-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; SDAG-NEXT:    s_mov_b64 exec, s[34:35]
 ; SDAG-NEXT:    v_writelane_b32 v40, s4, 0
 ; SDAG-NEXT:    v_writelane_b32 v40, s5, 1
 ; SDAG-NEXT:    v_writelane_b32 v40, s6, 2
 ; SDAG-NEXT:    v_writelane_b32 v40, s7, 3

 ; SDAG-NEXT:    v_readlane_b32 s7, v40, 3
 ; SDAG-NEXT:    v_readlane_b32 s6, v40, 2
 ; SDAG-NEXT:    v_readlane_b32 s5, v40, 1
 ; SDAG-NEXT:    v_readlane_b32 s4, v40, 0
 ; SDAG-NEXT:    s_or_saveexec_b64 s[34:35], -1
 ; SDAG-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; SDAG-NEXT:    s_mov_b64 exec, s[34:35]
 ; SDAG-NEXT:    s_addk_i32 s32, 0xfc00
-; SDAG-NEXT:    s_mov_b32 s33, s36
+; SDAG-NEXT:    s_mov_b32 s33, s38
 ; SDAG-NEXT:    s_waitcnt vmcnt(0)
 ; SDAG-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GISEL-LABEL: gfx_func:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT:    s_mov_b32 s36, s33
+; GISEL-NEXT:    s_mov_b32 s38, s33
 ; GISEL-NEXT:    s_mov_b32 s33, s32
 ; GISEL-NEXT:    s_or_saveexec_b64 s[34:35], -1
 ; GISEL-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GISEL-NEXT:    s_mov_b64 exec, s[34:35]
 ; GISEL-NEXT:    v_writelane_b32 v40, s4, 0
 ; GISEL-NEXT:    v_writelane_b32 v40, s5, 1
 ; GISEL-NEXT:    v_writelane_b32 v40, s6, 2
 ; GISEL-NEXT:    v_writelane_b32 v40, s7, 3

 ; GISEL-NEXT:    v_readlane_b32 s7, v40, 3
 ; GISEL-NEXT:    v_readlane_b32 s6, v40, 2
 ; GISEL-NEXT:    v_readlane_b32 s5, v40, 1
 ; GISEL-NEXT:    v_readlane_b32 s4, v40, 0
 ; GISEL-NEXT:    s_or_saveexec_b64 s[34:35], -1
 ; GISEL-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GISEL-NEXT:    s_mov_b64 exec, s[34:35]
 ; GISEL-NEXT:    s_addk_i32 s32, 0xfc00
-; GISEL-NEXT:    s_mov_b32 s33, s36
+; GISEL-NEXT:    s_mov_b32 s33, s38
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_setpc_b64 s[30:31]
   call void @extern_c_func()
   ret void
 }

llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll

; GFX10-SCRATCH-NEXT:    s_setpc_b64 s[30:31]
   call amdgpu_gfx void @external_void_func_v16i8(<16 x i8> %val)
   ret void
 }
 
 define void @tail_call_byval_align16(<32 x i32> %val, double %tmp) #0 {
 ; GFX9-LABEL: tail_call_byval_align16:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s6, s33
+; GFX9-NEXT:    s_mov_b32 s8, s33
 ; GFX9-NEXT:    s_mov_b32 s33, s32
 ; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    buffer_load_dword v32, off, s[0:3], s33 offset:20
 ; GFX9-NEXT:    buffer_load_dword v33, off, s[0:3], s33 offset:16
 ; GFX9-NEXT:    buffer_load_dword v31, off, s[0:3], s33
 ; GFX9-NEXT:    v_writelane_b32 v40, s30, 0

 ; GFX9-NEXT:    v_readlane_b32 s35, v40, 3
 ; GFX9-NEXT:    v_readlane_b32 s34, v40, 2
 ; GFX9-NEXT:    v_readlane_b32 s31, v40, 1
 ; GFX9-NEXT:    v_readlane_b32 s30, v40, 0
 ; GFX9-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GFX9-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
 ; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX9-NEXT:    s_addk_i32 s32, 0xf800
-; GFX9-NEXT:    s_mov_b32 s33, s6
+; GFX9-NEXT:    s_mov_b32 s33, s8
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: tail_call_byval_align16:
 ; GFX10:       ; %bb.0: ; %entry
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_mov_b32 s6, s33
+; GFX10-NEXT:    s_mov_b32 s7, s33
 ; GFX10-NEXT:    s_mov_b32 s33, s32
 ; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_clause 0x2
 ; GFX10-NEXT:    buffer_load_dword v32, off, s[0:3], s33 offset:20
 ; GFX10-NEXT:    buffer_load_dword v33, off, s[0:3], s33 offset:16

 ; GFX10-NEXT:    v_readlane_b32 s34, v40, 2
 ; GFX10-NEXT:    v_readlane_b32 s31, v40, 1
 ; GFX10-NEXT:    v_readlane_b32 s30, v40, 0
 ; GFX10-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX10-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX10-NEXT:    s_addk_i32 s32, 0xfc00
-; GFX10-NEXT:    s_mov_b32 s33, s6
+; GFX10-NEXT:    s_mov_b32 s33, s7
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: tail_call_byval_align16:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s4, s33
+; GFX11-NEXT:    s_mov_b32 s5, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v40, s33 offset:24 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_clause 0x1
 ; GFX11-NEXT:    scratch_load_b64 v[32:33], off, s33 offset:16
 ; GFX11-NEXT:    scratch_load_b32 v31, off, s33
 ; GFX11-NEXT:    v_writelane_b32 v40, s30, 0

 ; GFX11-NEXT:    v_readlane_b32 s35, v40, 3
 ; GFX11-NEXT:    v_readlane_b32 s34, v40, 2
 ; GFX11-NEXT:    v_readlane_b32 s31, v40, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v40, 0
 ; GFX11-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v40, off, s33 offset:24 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_addk_i32 s32, 0xffe0
-; GFX11-NEXT:    s_mov_b32 s33, s4
+; GFX11-NEXT:    s_mov_b32 s33, s5
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-SCRATCH-LABEL: tail_call_byval_align16:
 ; GFX10-SCRATCH:       ; %bb.0: ; %entry
 ; GFX10-SCRATCH-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-SCRATCH-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-SCRATCH-NEXT:    s_mov_b32 s4, s33
+; GFX10-SCRATCH-NEXT:    s_mov_b32 s5, s33
 ; GFX10-SCRATCH-NEXT:    s_mov_b32 s33, s32
 ; GFX10-SCRATCH-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX10-SCRATCH-NEXT:    scratch_store_dword off, v40, s33 offset:24 ; 4-byte Folded Spill
 ; GFX10-SCRATCH-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-SCRATCH-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX10-SCRATCH-NEXT:    s_clause 0x1
 ; GFX10-SCRATCH-NEXT:    scratch_load_dwordx2 v[32:33], off, s33 offset:16
 ; GFX10-SCRATCH-NEXT:    scratch_load_dword v31, off, s33

 ; GFX10-SCRATCH-NEXT:    v_readlane_b32 s34, v40, 2
 ; GFX10-SCRATCH-NEXT:    v_readlane_b32 s31, v40, 1
 ; GFX10-SCRATCH-NEXT:    v_readlane_b32 s30, v40, 0
 ; GFX10-SCRATCH-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX10-SCRATCH-NEXT:    scratch_load_dword v40, off, s33 offset:24 ; 4-byte Folded Reload
 ; GFX10-SCRATCH-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-SCRATCH-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX10-SCRATCH-NEXT:    s_addk_i32 s32, 0xffe0
-; GFX10-SCRATCH-NEXT:    s_mov_b32 s33, s4
+; GFX10-SCRATCH-NEXT:    s_mov_b32 s33, s5
 ; GFX10-SCRATCH-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-SCRATCH-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %alloca = alloca double, align 8, addrspace(5)
   tail call amdgpu_gfx void @byval_align16_f64_arg(<32 x i32> %val, ptr addrspace(5) byval(double) align 16 %alloca)
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/gfx-callable-preserved-registers.ll

 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: void_func_void_clobber_s28_s29:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
+; GFX11-NEXT:    s_xor_saveexec_b32 s1, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s32 ; 4-byte Folded Spill
-; GFX11-NEXT:    s_mov_b32 exec_lo, s0
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    v_writelane_b32 v0, s28, 0
 ; GFX11-NEXT:    v_writelane_b32 v0, s29, 1
 ; GFX11-NEXT:    v_writelane_b32 v0, s30, 2
 ; GFX11-NEXT:    v_writelane_b32 v0, s31, 3
 ; GFX11-NEXT:    ;;#ASMSTART
 ; GFX11-NEXT:    ; clobber
 ; GFX11-NEXT:    ;;#ASMEND
 ; GFX11-NEXT:    ;;#ASMSTART
 ; GFX11-NEXT:    ; clobber
 ; GFX11-NEXT:    ;;#ASMEND
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s31, v0, 3
 ; GFX11-NEXT:    v_readlane_b32 s30, v0, 2
 ; GFX11-NEXT:    v_readlane_b32 s29, v0, 1
 ; GFX11-NEXT:    v_readlane_b32 s28, v0, 0
-; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
+; GFX11-NEXT:    s_xor_saveexec_b32 s1, -1
 ; GFX11-NEXT:    scratch_load_b32 v0, off, s32 ; 4-byte Folded Reload
-; GFX11-NEXT:    s_mov_b32 exec_lo, s0
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "; clobber", "~{s[30:31]}"() #0
   call void asm sideeffect "; clobber", "~{s[28:29]}"() #0
   ret void
 }
 
; GFX11-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "; use $0", "{v40}"(i32 %v40)
   ret void
 }
 
 define hidden void @void_func_void_clobber_s33() #1 {
 ; GFX9-LABEL: void_func_void_clobber_s33:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    s_xor_saveexec_b64 s[6:7], -1
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
-; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX9-NEXT:    v_writelane_b32 v0, s33, 0
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; clobber
 ; GFX9-NEXT:    ;;#ASMEND
 ; GFX9-NEXT:    v_readlane_b32 s33, v0, 0
-; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    s_xor_saveexec_b64 s[6:7], -1
 ; GFX9-NEXT:    buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
-; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: void_func_void_clobber_s33:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
+; GFX10-NEXT:    s_xor_saveexec_b32 s5, -1
 ; GFX10-NEXT:    buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    v_writelane_b32 v0, s33, 0
 ; GFX10-NEXT:    ;;#ASMSTART
 ; GFX10-NEXT:    ; clobber
 ; GFX10-NEXT:    ;;#ASMEND
 ; GFX10-NEXT:    v_readlane_b32 s33, v0, 0
-; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
+; GFX10-NEXT:    s_xor_saveexec_b32 s5, -1
 ; GFX10-NEXT:    buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: void_func_void_clobber_s33:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
+; GFX11-NEXT:    s_xor_saveexec_b32 s1, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s32 ; 4-byte Folded Spill
-; GFX11-NEXT:    s_mov_b32 exec_lo, s0
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    v_writelane_b32 v0, s33, 0
 ; GFX11-NEXT:    ;;#ASMSTART
 ; GFX11-NEXT:    ; clobber
 ; GFX11-NEXT:    ;;#ASMEND
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s33, v0, 0
-; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
+; GFX11-NEXT:    s_xor_saveexec_b32 s1, -1
 ; GFX11-NEXT:    scratch_load_b32 v0, off, s32 ; 4-byte Folded Reload
-; GFX11-NEXT:    s_mov_b32 exec_lo, s0
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "; clobber", "~{s33}"() #0
   ret void
 }
 
 define hidden void @void_func_void_clobber_s34() #1 {
 ; GFX9-LABEL: void_func_void_clobber_s34:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    s_xor_saveexec_b64 s[6:7], -1
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
-; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX9-NEXT:    v_writelane_b32 v0, s34, 0
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; clobber
 ; GFX9-NEXT:    ;;#ASMEND
 ; GFX9-NEXT:    v_readlane_b32 s34, v0, 0
-; GFX9-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; GFX9-NEXT:    s_xor_saveexec_b64 s[6:7], -1
 ; GFX9-NEXT:    buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
-; GFX9-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX9-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: void_func_void_clobber_s34:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
+; GFX10-NEXT:    s_xor_saveexec_b32 s5, -1
 ; GFX10-NEXT:    buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    v_writelane_b32 v0, s34, 0
 ; GFX10-NEXT:    ;;#ASMSTART
 ; GFX10-NEXT:    ; clobber
 ; GFX10-NEXT:    ;;#ASMEND
 ; GFX10-NEXT:    v_readlane_b32 s34, v0, 0
-; GFX10-NEXT:    s_xor_saveexec_b32 s4, -1
+; GFX10-NEXT:    s_xor_saveexec_b32 s5, -1
 ; GFX10-NEXT:    buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_mov_b32 exec_lo, s4
+; GFX10-NEXT:    s_mov_b32 exec_lo, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: void_func_void_clobber_s34:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
+; GFX11-NEXT:    s_xor_saveexec_b32 s1, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v0, s32 ; 4-byte Folded Spill
-; GFX11-NEXT:    s_mov_b32 exec_lo, s0
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    v_writelane_b32 v0, s34, 0
 ; GFX11-NEXT:    ;;#ASMSTART
 ; GFX11-NEXT:    ; clobber
 ; GFX11-NEXT:    ;;#ASMEND
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s34, v0, 0
-; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
+; GFX11-NEXT:    s_xor_saveexec_b32 s1, -1
 ; GFX11-NEXT:    scratch_load_b32 v0, off, s32 ; 4-byte Folded Reload
-; GFX11-NEXT:    s_mov_b32 exec_lo, s0
+; GFX11-NEXT:    s_mov_b32 exec_lo, s1
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "; clobber", "~{s34}"() #0
   ret void
 }
 
 define amdgpu_gfx void @test_call_void_func_void_clobber_s33() #0 {

llvm/test/CodeGen/AMDGPU/gfx-callable-return-types.ll

 ; GFX10-NEXT:    s_mov_b32 s33, s36
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: call_i1:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s2, s33
+; GFX11-NEXT:    s_mov_b32 s3, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v1, s33 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, 16
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, return_i1@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, return_i1@gotpcrel32@hi+12
 ; GFX11-NEXT:    v_writelane_b32 v1, s30, 0
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_writelane_b32 v1, s31, 1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s31, v1, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v1, 0
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v1, off, s33 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, -16
-; GFX11-NEXT:    s_mov_b32 s33, s2
+; GFX11-NEXT:    s_mov_b32 s33, s3
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 entry:
   call amdgpu_gfx i1 @return_i1()
   ret void
 }
 
 define amdgpu_gfx i16 @return_i16() #0 {

 ; GFX10-NEXT:    s_mov_b32 s33, s36
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: call_i16:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s2, s33
+; GFX11-NEXT:    s_mov_b32 s3, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v1, s33 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, 16
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, return_i16@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, return_i16@gotpcrel32@hi+12
 ; GFX11-NEXT:    v_writelane_b32 v1, s30, 0
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_writelane_b32 v1, s31, 1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s31, v1, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v1, 0
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v1, off, s33 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, -16
-; GFX11-NEXT:    s_mov_b32 s33, s2
+; GFX11-NEXT:    s_mov_b32 s33, s3
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 entry:
   call amdgpu_gfx i16 @return_i16()
   ret void
 }
 
 define amdgpu_gfx <2 x i16> @return_2xi16() #0 {

 ; GFX10-NEXT:    s_mov_b32 s33, s36
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: call_2xi16:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s2, s33
+; GFX11-NEXT:    s_mov_b32 s3, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v1, s33 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, 16
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, return_2xi16@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, return_2xi16@gotpcrel32@hi+12
 ; GFX11-NEXT:    v_writelane_b32 v1, s30, 0
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_writelane_b32 v1, s31, 1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s31, v1, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v1, 0
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v1, off, s33 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, -16
-; GFX11-NEXT:    s_mov_b32 s33, s2
+; GFX11-NEXT:    s_mov_b32 s33, s3
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 entry:
   call amdgpu_gfx <2 x i16> @return_2xi16()
   ret void
 }
 
 define amdgpu_gfx <3 x i16> @return_3xi16() #0 {

 ; GFX10-NEXT:    s_mov_b32 s33, s36
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: call_3xi16:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s2, s33
+; GFX11-NEXT:    s_mov_b32 s3, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v2, s33 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, 16
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, return_3xi16@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, return_3xi16@gotpcrel32@hi+12
 ; GFX11-NEXT:    v_writelane_b32 v2, s30, 0
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_writelane_b32 v2, s31, 1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s31, v2, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v2, 0
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v2, off, s33 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, -16
-; GFX11-NEXT:    s_mov_b32 s33, s2
+; GFX11-NEXT:    s_mov_b32 s33, s3
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 entry:
   call amdgpu_gfx <3 x i16> @return_3xi16()
   ret void
 }
 
 ; Check that return values that overlap CSRs are correctly handled

 ; GFX10-NEXT:    s_mov_b32 s33, s36
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: call_100xi32:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s2, s33
+; GFX11-NEXT:    s_mov_b32 s3, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v100, s33 offset:128 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_addk_i32 s32, 0x90
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, return_100xi32@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, return_100xi32@gotpcrel32@hi+12

 ; GFX11-NEXT:    scratch_load_b32 v41, off, s33 offset:120
 ; GFX11-NEXT:    scratch_load_b32 v40, off, s33 offset:124
 ; GFX11-NEXT:    v_readlane_b32 s31, v100, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v100, 0
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v100, off, s33 offset:128 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_addk_i32 s32, 0xff70
-; GFX11-NEXT:    s_mov_b32 s33, s2
+; GFX11-NEXT:    s_mov_b32 s33, s3
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 entry:
   call amdgpu_gfx <100 x i32> @return_100xi32()
   ret void
 }
 
 ; Check that return values that do not fit in registers do not crash

llvm/test/CodeGen/AMDGPU/indirect-call.ll

; GISEL-NEXT:    s_setpc_b64 s[30:31]
   call amdgpu_gfx void %fptr(i32 inreg 123)
   ret void
 }
 
 define i32 @test_indirect_call_vgpr_ptr_arg_and_reuse(i32 %i, ptr %fptr) {
 ; GCN-LABEL: test_indirect_call_vgpr_ptr_arg_and_reuse:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s10, s33
+; GCN-NEXT:    s_mov_b32 s12, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v40, s31, 1

 ; GCN-NEXT:    v_readlane_b32 s34, v40, 2
 ; GCN-NEXT:    v_readlane_b32 s31, v40, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v40, 0
 ; GCN-NEXT:    buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s10
+; GCN-NEXT:    s_mov_b32 s33, s12
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GISEL-LABEL: test_indirect_call_vgpr_ptr_arg_and_reuse:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT:    s_mov_b32 s10, s33
+; GISEL-NEXT:    s_mov_b32 s12, s33
 ; GISEL-NEXT:    s_mov_b32 s33, s32
 ; GISEL-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GISEL-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
 ; GISEL-NEXT:    s_mov_b64 exec, s[4:5]
 ; GISEL-NEXT:    s_addk_i32 s32, 0x400
 ; GISEL-NEXT:    buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GISEL-NEXT:    v_writelane_b32 v40, s30, 0
 ; GISEL-NEXT:    v_writelane_b32 v40, s31, 1

 ; GISEL-NEXT:    v_readlane_b32 s34, v40, 2
 ; GISEL-NEXT:    v_readlane_b32 s31, v40, 1
 ; GISEL-NEXT:    v_readlane_b32 s30, v40, 0
 ; GISEL-NEXT:    buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GISEL-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GISEL-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
 ; GISEL-NEXT:    s_mov_b64 exec, s[4:5]
 ; GISEL-NEXT:    s_addk_i32 s32, 0xfc00
-; GISEL-NEXT:    s_mov_b32 s33, s10
+; GISEL-NEXT:    s_mov_b32 s33, s12
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_setpc_b64 s[30:31]
   call amdgpu_gfx void %fptr(i32 %i)
   ret i32 %i
 }
 
 ; Use a variable inside a waterfall loop and use the return variable after the loop.
 ; TODO The argument and return variable could be in the same physical register, but the register
 ; allocator is not able to do that because the return value clashes with the liverange of an
 ; IMPLICIT_DEF of the argument.
 define i32 @test_indirect_call_vgpr_ptr_arg_and_return(i32 %i, ptr %fptr) {
 ; GCN-LABEL: test_indirect_call_vgpr_ptr_arg_and_return:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s10, s33
+; GCN-NEXT:    s_mov_b32 s12, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v40, s31, 1
 ; GCN-NEXT:    v_writelane_b32 v40, s34, 2

 ; GCN-NEXT:    v_readlane_b32 s35, v40, 3
 ; GCN-NEXT:    v_readlane_b32 s34, v40, 2
 ; GCN-NEXT:    v_readlane_b32 s31, v40, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v40, 0
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s10
+; GCN-NEXT:    s_mov_b32 s33, s12
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GISEL-LABEL: test_indirect_call_vgpr_ptr_arg_and_return:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT:    s_mov_b32 s10, s33
+; GISEL-NEXT:    s_mov_b32 s12, s33
 ; GISEL-NEXT:    s_mov_b32 s33, s32
 ; GISEL-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GISEL-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GISEL-NEXT:    s_mov_b64 exec, s[4:5]
 ; GISEL-NEXT:    s_addk_i32 s32, 0x400
 ; GISEL-NEXT:    v_writelane_b32 v40, s30, 0
 ; GISEL-NEXT:    v_writelane_b32 v40, s31, 1
 ; GISEL-NEXT:    v_writelane_b32 v40, s34, 2

 ; GISEL-NEXT:    v_readlane_b32 s35, v40, 3
 ; GISEL-NEXT:    v_readlane_b32 s34, v40, 2
 ; GISEL-NEXT:    v_readlane_b32 s31, v40, 1
 ; GISEL-NEXT:    v_readlane_b32 s30, v40, 0
 ; GISEL-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GISEL-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GISEL-NEXT:    s_mov_b64 exec, s[4:5]
 ; GISEL-NEXT:    s_addk_i32 s32, 0xfc00
-; GISEL-NEXT:    s_mov_b32 s33, s10
+; GISEL-NEXT:    s_mov_b32 s33, s12
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_setpc_b64 s[30:31]
   %ret = call amdgpu_gfx i32 %fptr(i32 %i)
   ret i32 %ret
 }
 
 ; Calling a vgpr can never be a tail call.
 define void @test_indirect_tail_call_vgpr_ptr(ptr %fptr) {
 ; GCN-LABEL: test_indirect_tail_call_vgpr_ptr:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s10, s33
+; GCN-NEXT:    s_mov_b32 s12, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0x400
 ; GCN-NEXT:    v_writelane_b32 v40, s30, 0
 ; GCN-NEXT:    v_writelane_b32 v40, s31, 1
 ; GCN-NEXT:    v_writelane_b32 v40, s34, 2

 ; GCN-NEXT:    v_readlane_b32 s35, v40, 3
 ; GCN-NEXT:    v_readlane_b32 s34, v40, 2
 ; GCN-NEXT:    v_readlane_b32 s31, v40, 1
 ; GCN-NEXT:    v_readlane_b32 s30, v40, 0
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_addk_i32 s32, 0xfc00
-; GCN-NEXT:    s_mov_b32 s33, s10
+; GCN-NEXT:    s_mov_b32 s33, s12
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GISEL-LABEL: test_indirect_tail_call_vgpr_ptr:
 ; GISEL:       ; %bb.0:
 ; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT:    s_mov_b32 s10, s33
+; GISEL-NEXT:    s_mov_b32 s12, s33
 ; GISEL-NEXT:    s_mov_b32 s33, s32
 ; GISEL-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GISEL-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GISEL-NEXT:    s_mov_b64 exec, s[4:5]
 ; GISEL-NEXT:    s_addk_i32 s32, 0x400
 ; GISEL-NEXT:    v_writelane_b32 v40, s30, 0
 ; GISEL-NEXT:    v_writelane_b32 v40, s31, 1
 ; GISEL-NEXT:    v_writelane_b32 v40, s34, 2

 ; GISEL-NEXT:    v_readlane_b32 s35, v40, 3
 ; GISEL-NEXT:    v_readlane_b32 s34, v40, 2
 ; GISEL-NEXT:    v_readlane_b32 s31, v40, 1
 ; GISEL-NEXT:    v_readlane_b32 s30, v40, 0
 ; GISEL-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GISEL-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; GISEL-NEXT:    s_mov_b64 exec, s[4:5]
 ; GISEL-NEXT:    s_addk_i32 s32, 0xfc00
-; GISEL-NEXT:    s_mov_b32 s33, s10
+; GISEL-NEXT:    s_mov_b32 s33, s12
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    s_setpc_b64 s[30:31]
   tail call amdgpu_gfx void %fptr()
   ret void
 }
 
 !llvm.module.flags = !{!0}
 !0 = !{i32 1, !"amdgpu_code_object_version", i32 200}

llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs %s -o - | FileCheck %s -check-prefix=GFX11
 
 declare i32 @llvm.amdgcn.workitem.id.x()
 
 define void @f0() {
 ; GFX11-LABEL: f0:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-NEXT:    s_mov_b32 s2, s33
+; GFX11-NEXT:    s_mov_b32 s3, s33
 ; GFX11-NEXT:    s_mov_b32 s33, s32
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_store_b32 off, v4, s33 ; 4-byte Folded Spill
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, 16
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, f1@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, f1@gotpcrel32@hi+12
 ; GFX11-NEXT:    v_writelane_b32 v4, s30, 0
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    v_writelane_b32 v4, s31, 1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_readlane_b32 s31, v4, 1
 ; GFX11-NEXT:    v_readlane_b32 s30, v4, 0
 ; GFX11-NEXT:    s_xor_saveexec_b32 s0, -1
 ; GFX11-NEXT:    scratch_load_b32 v4, off, s33 ; 4-byte Folded Reload
 ; GFX11-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX11-NEXT:    s_add_i32 s32, s32, -16
-; GFX11-NEXT:    s_mov_b32 s33, s2
+; GFX11-NEXT:    s_mov_b32 s33, s3
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %i = call <2 x i64> @f1()
   ret void
 }
 
 define <2 x i64> @f1() #0 {
; GFX11-NEXT:    s_setpc_b64 s[30:31]
   ret <2 x i64> zeroinitializer
 }
 
 ; FIXME: This generates "instid1(/* invalid instid value */)".
 define amdgpu_kernel void @f2(i32 %arg, i32 %arg1, i32 %arg2, i1 %arg3, i32 %arg4, i1 %arg5, ptr %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10, i1 %arg11) {
 ; GFX11-LABEL: f2:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_mov_b64 s[16:17], s[4:5]
+; GFX11-NEXT:    s_mov_b64 s[10:11], s[6:7]
+; GFX11-NEXT:    s_mov_b64 s[6:7], s[2:3]
+; GFX11-NEXT:    s_load_b32 s2, s[16:17], 0x24
 ; GFX11-NEXT:    v_mov_b32_e32 v31, v0
-; GFX11-NEXT:    s_load_b32 s24, s[16:17], 0x24
 ; GFX11-NEXT:    s_mov_b32 s18, s14
 ; GFX11-NEXT:    s_mov_b32 s12, s13
-; GFX11-NEXT:    s_mov_b64 s[10:11], s[6:7]
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v31
-; GFX11-NEXT:    s_mov_b64 s[6:7], s[2:3]
 ; GFX11-NEXT:    s_mov_b64 s[4:5], s[0:1]
-; GFX11-NEXT:    s_mov_b32 s3, 0
+; GFX11-NEXT:    s_mov_b32 s20, 0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v31
 ; GFX11-NEXT:    s_mov_b32 s0, -1
 ; GFX11-NEXT:    s_mov_b32 s19, exec_lo
 ; GFX11-NEXT:    s_mov_b32 s32, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_mul_lo_u32 v0, s24, v0
+; GFX11-NEXT:    v_mul_lo_u32 v0, s2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
 ; GFX11-NEXT:    v_cmpx_eq_u32_e32 0, v0
 ; GFX11-NEXT:    s_cbranch_execz .LBB2_13
 ; GFX11-NEXT:  ; %bb.1: ; %bb14
 ; GFX11-NEXT:    s_load_b128 s[20:23], s[16:17], 0x2c
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_bitcmp1_b32 s21, 0
-; GFX11-NEXT:    s_cselect_b32 s25, -1, 0
+; GFX11-NEXT:    s_cselect_b32 s24, -1, 0
 ; GFX11-NEXT:    s_bitcmp0_b32 s21, 0
 ; GFX11-NEXT:    s_mov_b32 s21, 0
 ; GFX11-NEXT:    s_cbranch_scc0 .LBB2_3
 ; GFX11-NEXT:  ; %bb.2: ; %bb15
 ; GFX11-NEXT:    s_add_u32 s8, s16, 0x58
 ; GFX11-NEXT:    s_addc_u32 s9, s17, 0
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, f0@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, f0@gotpcrel32@hi+12
 ; GFX11-NEXT:    s_mov_b32 s13, s18
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    s_mov_b32 s14, s15
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX11-NEXT:    s_mov_b32 s1, -1
-; GFX11-NEXT:    s_and_not1_b32 vcc_lo, exec_lo, s3
-; GFX11-NEXT:    s_cbranch_vccz .LBB2_4
+; GFX11-NEXT:    s_cbranch_execz .LBB2_4
 ; GFX11-NEXT:    s_branch .LBB2_12
 ; GFX11-NEXT:  .LBB2_3:
 ; GFX11-NEXT:    s_mov_b32 s1, 0
 ; GFX11-NEXT:  .LBB2_4: ; %bb16
-; GFX11-NEXT:    s_load_b32 s2, s[16:17], 0x54
+; GFX11-NEXT:    s_load_b32 s3, s[16:17], 0x54
 ; GFX11-NEXT:    s_bitcmp1_b32 s23, 0
 ; GFX11-NEXT:    s_cselect_b32 s0, -1, 0
-; GFX11-NEXT:    s_and_b32 s3, s23, 1
+; GFX11-NEXT:    s_and_b32 s9, s23, 1
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_bitcmp1_b32 s2, 0
-; GFX11-NEXT:    s_mov_b32 s2, -1
+; GFX11-NEXT:    s_bitcmp1_b32 s3, 0
+; GFX11-NEXT:    s_mov_b32 s3, -1
 ; GFX11-NEXT:    s_cselect_b32 s8, -1, 0
-; GFX11-NEXT:    s_cmp_eq_u32 s3, 0
+; GFX11-NEXT:    s_cmp_eq_u32 s9, 0
 ; GFX11-NEXT:    s_cbranch_scc0 .LBB2_8
 ; GFX11-NEXT:  ; %bb.5: ; %bb18.preheader
 ; GFX11-NEXT:    s_load_b128 s[28:31], s[16:17], 0x44
 ; GFX11-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mul_hi_u32 s2, s29, s28
-; GFX11-NEXT:    s_mul_i32 s3, s29, s28
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_alignbit_b32 v0, s2, s3, 1
-; GFX11-NEXT:    s_mov_b32 s3, 0
-; GFX11-NEXT:    v_readfirstlane_b32 s2, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s25
+; GFX11-NEXT:    s_mul_hi_u32 s3, s29, s28
+; GFX11-NEXT:    s_mul_i32 s9, s29, s28
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-NEXT:    v_alignbit_b32 v0, s3, s9, 1
+; GFX11-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s24
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_or_b32 s2, s2, 1
-; GFX11-NEXT:    s_lshr_b32 s2, s2, s30
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_mul_i32 s2, s2, s22
-; GFX11-NEXT:    s_mul_i32 s2, s2, s20
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    s_or_b32 s2, s24, s2
+; GFX11-NEXT:    s_or_b32 s3, s3, 1
+; GFX11-NEXT:    s_lshr_b32 s3, s3, s30
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    s_mul_i32 s9, s3, s22
+; GFX11-NEXT:    s_mov_b32 s3, 0
+; GFX11-NEXT:    s_mul_i32 s9, s9, s20
+; GFX11-NEXT:    s_or_b32 s2, s2, s9
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_lshl_b64 s[22:23], s[2:3], 1
 ; GFX11-NEXT:    global_load_u16 v2, v1, s[22:23]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0, v2
 ; GFX11-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc_lo
 ; GFX11-NEXT:    .p2align 6
 ; GFX11-NEXT:  .LBB2_6: ; %bb18
 ; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1

 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-NEXT:    v_and_b32_e32 v1, 1, v1
 ; GFX11-NEXT:    s_bitcmp1_b32 s2, 0
 ; GFX11-NEXT:    s_cselect_b32 s2, 0x100, 0
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_or_b32 s3, s2, s3
 ; GFX11-NEXT:    s_cbranch_vccz .LBB2_6
 ; GFX11-NEXT:  ; %bb.7: ; %Flow
-; GFX11-NEXT:    s_mov_b32 s2, 0
+; GFX11-NEXT:    s_mov_b32 s3, 0
 ; GFX11-NEXT:  .LBB2_8: ; %Flow12
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_and_b32 vcc_lo, exec_lo, s2
+; GFX11-NEXT:    s_and_b32 vcc_lo, exec_lo, s3
 ; GFX11-NEXT:    s_cbranch_vccz .LBB2_12
 ; GFX11-NEXT:  ; %bb.9:
 ; GFX11-NEXT:    s_xor_b32 s0, s8, -1
 ; GFX11-NEXT:  .LBB2_10: ; %bb17
 ; GFX11-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_and_b32 vcc_lo, exec_lo, s0
 ; GFX11-NEXT:    s_cbranch_vccz .LBB2_10
 ; GFX11-NEXT:  ; %bb.11: ; %Flow6
 ; GFX11-NEXT:    s_mov_b32 s21, -1
 ; GFX11-NEXT:  .LBB2_12: ; %Flow11
-; GFX11-NEXT:    s_and_b32 s3, s1, exec_lo
+; GFX11-NEXT:    s_and_b32 s20, s1, exec_lo
 ; GFX11-NEXT:    s_or_not1_b32 s0, s21, exec_lo
 ; GFX11-NEXT:  .LBB2_13: ; %Flow9
 ; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s19
-; GFX11-NEXT:    s_and_saveexec_b32 s19, s0
+; GFX11-NEXT:    s_and_saveexec_b32 s2, s0
 ; GFX11-NEXT:    s_cbranch_execz .LBB2_15
 ; GFX11-NEXT:  ; %bb.14: ; %bb43
 ; GFX11-NEXT:    s_add_u32 s8, s16, 0x58
 ; GFX11-NEXT:    s_addc_u32 s9, s17, 0
 ; GFX11-NEXT:    s_getpc_b64 s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, f0@gotpcrel32@lo+4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, f0@gotpcrel32@hi+12
 ; GFX11-NEXT:    s_mov_b32 s13, s18
 ; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
 ; GFX11-NEXT:    s_mov_b32 s14, s15
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
-; GFX11-NEXT:    s_or_b32 s3, s3, exec_lo
+; GFX11-NEXT:    s_or_b32 s20, s20, exec_lo
 ; GFX11-NEXT:  .LBB2_15: ; %Flow14
-; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s19
-; GFX11-NEXT:    s_and_saveexec_b32 s0, s3
+; GFX11-NEXT:    s_or_b32 exec_lo, exec_lo, s2
+; GFX11-NEXT:    s_and_saveexec_b32 s0, s20
 ; GFX11-NEXT:  ; %bb.16: ; %UnifiedUnreachableBlock
 ; GFX11-NEXT:    ; divergent unreachable
 ; GFX11-NEXT:  ; %bb.17: ; %UnifiedReturnBlock
 ; GFX11-NEXT:    s_endpgm
 bb:
   %i = tail call i32 @llvm.amdgcn.workitem.id.x()
   %i12 = mul i32 %arg, %i
   %i13 = icmp ult i32 %i12, 1

llvm/test/CodeGen/AMDGPU/load-constant-i16.ll

 ; GCN-HSA-NEXT:    s_lshr_b32 s23, s2, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s24, s5, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s26, s4, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s28, s7, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s30, s6, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s33, s9, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s35, s8, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s37, s11, 16
-; GCN-HSA-NEXT:    s_lshr_b32 s38, s10, 16
-; GCN-HSA-NEXT:    s_lshr_b32 s39, s13, 16
-; GCN-HSA-NEXT:    s_lshr_b32 s40, s12, 16
-; GCN-HSA-NEXT:    s_lshr_b32 s41, s15, 16
-; GCN-HSA-NEXT:    s_lshr_b32 s42, s14, 16
+; GCN-HSA-NEXT:    s_lshr_b32 s39, s10, 16
+; GCN-HSA-NEXT:    s_lshr_b32 s40, s13, 16
+; GCN-HSA-NEXT:    s_lshr_b32 s41, s12, 16
+; GCN-HSA-NEXT:    s_lshr_b32 s42, s15, 16
+; GCN-HSA-NEXT:    s_lshr_b32 s43, s14, 16
 ; GCN-HSA-NEXT:    s_and_b32 s25, s1, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s27, s0, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s29, s3, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s31, s2, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s34, s5, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s36, s4, 0xffff
-; GCN-HSA-NEXT:    s_and_b32 s43, s7, 0xffff
+; GCN-HSA-NEXT:    s_and_b32 s38, s7, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s44, s6, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s45, s9, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s46, s8, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s47, s11, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s48, s10, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s49, s13, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s50, s12, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s51, s15, 0xffff

 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[30:31], v[16:19]
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[32:33], v[20:23]
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[24:25], v[0:3]
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[34:35], v[4:7]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s52
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x60
-; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s42
+; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s43
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s51
-; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s41
+; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s42
 ; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x50
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s50
-; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s40
+; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s41
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s49
-; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s39
+; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s40
 ; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-HSA-NEXT:    s_add_u32 s0, s16, 64
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s48
-; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s38
+; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s39
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s47
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s37
 ; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-HSA-NEXT:    s_add_u32 s0, s16, 48
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s46
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s35
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s45
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s33
 ; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-HSA-NEXT:    s_add_u32 s0, s16, 32
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s44
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s30
-; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s43
+; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s38
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v3, s28
 ; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-HSA-NEXT:    s_add_u32 s0, s16, 16
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s36
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v1, s26

 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s1, s1, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s0, s0, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s2, s2, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s4, s4, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s7, s7, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s6, s6, 0xffff
-; GCN-NOHSA-VI-NEXT:    s_and_b32 s67, s9, 0xffff
-; GCN-NOHSA-VI-NEXT:    s_and_b32 s68, s8, 0xffff
-; GCN-NOHSA-VI-NEXT:    s_and_b32 s11, s11, 0xffff
-; GCN-NOHSA-VI-NEXT:    s_and_b32 s10, s10, 0xffff
+; GCN-NOHSA-VI-NEXT:    s_and_b32 s9, s9, 0xffff
+; GCN-NOHSA-VI-NEXT:    s_and_b32 s8, s8, 0xffff
+; GCN-NOHSA-VI-NEXT:    s_and_b32 s67, s11, 0xffff
+; GCN-NOHSA-VI-NEXT:    s_and_b32 s68, s10, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s13, s13, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s12, s12, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s15, s15, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s14, s14, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s17, s17, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s16, s16, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s19, s19, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s18, s18, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s21, s21, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s20, s20, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s23, s23, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s22, s22, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s25, s25, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s24, s24, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s27, s27, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s26, s26, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s29, s29, 0xffff
 ; GCN-NOHSA-VI-NEXT:    s_and_b32 s28, s28, 0xffff
-; GCN-NOHSA-VI-NEXT:    s_and_b32 s8, s31, 0xffff
-; GCN-NOHSA-VI-NEXT:    s_and_b32 s9, s30, 0xffff
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s8
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0xf0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s9
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_and_b32 s10, s31, 0xffff
+; GCN-NOHSA-VI-NEXT:    s_and_b32 s11, s30, 0xffff
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s10
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0xf0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s11
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s66
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s65
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0xe0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0xe0
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s28
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s64
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s29
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s63
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0xd0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0xd0
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s26
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s62
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s27
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s61
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0xc0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0xc0
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s24
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s60
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s25
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s59
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0xb0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0xb0
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s22
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s58
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s23
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s57
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0xa0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0xa0
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s20
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s56
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s21
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s55
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0x90
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0x90
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s18
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s54
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s19
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s53
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0x80
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0x80
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s16
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s52
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s17
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s51
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0x70
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0x70
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s14
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s50
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s15
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s49
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0x60
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0x60
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s12
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s39
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s13
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s38
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 0x50
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s10, s36, 0x50
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-NOHSA-VI-NEXT:    s_addc_u32 s11, s37, 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s10
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s68
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s48
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s11
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s67
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s47
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
-; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 64
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s11
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GCN-NOHSA-VI-NEXT:    s_nop 0
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s8
+; GCN-NOHSA-VI-NEXT:    s_add_u32 s8, s36, 64
+; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s9
 ; GCN-NOHSA-VI-NEXT:    s_addc_u32 s9, s37, 0
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v4, s8
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s68
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v1, s46
-; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s67
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v3, s45
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v5, s9
 ; GCN-NOHSA-VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NOHSA-VI-NEXT:    s_nop 0
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v0, s6
 ; GCN-NOHSA-VI-NEXT:    s_add_u32 s6, s36, 48
 ; GCN-NOHSA-VI-NEXT:    v_mov_b32_e32 v2, s7
 ; GCN-NOHSA-VI-NEXT:    s_addc_u32 s7, s37, 0

 ; GCN-HSA-NEXT:    s_lshr_b32 s27, s14, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s28, s12, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s29, s10, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s30, s8, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s31, s6, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s33, s4, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s34, s2, 16
 ; GCN-HSA-NEXT:    s_lshr_b32 s18, s0, 16
-; GCN-HSA-NEXT:    s_and_b32 s35, s0, 0xffff
-; GCN-HSA-NEXT:    s_and_b32 s2, s2, 0xffff
+; GCN-HSA-NEXT:    s_and_b32 s0, s0, 0xffff
+; GCN-HSA-NEXT:    s_and_b32 s35, s2, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s4, s4, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s6, s6, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s8, s8, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s10, s10, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s12, s12, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s14, s14, 0xffff
-; GCN-HSA-NEXT:    s_and_b32 s36, s1, 0xffff
-; GCN-HSA-NEXT:    s_and_b32 s3, s3, 0xffff
+; GCN-HSA-NEXT:    s_and_b32 s1, s1, 0xffff
+; GCN-HSA-NEXT:    s_and_b32 s36, s3, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s7, s7, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s9, s9, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s11, s11, 0xffff
 ; GCN-HSA-NEXT:    s_and_b32 s13, s13, 0xffff
-; GCN-HSA-NEXT:    s_and_b32 s0, s15, 0xffff
-; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0xf0
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0xd0
+; GCN-HSA-NEXT:    s_and_b32 s2, s15, 0xffff
+; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0xf0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0xd0
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s26
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0xb0
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0xb0
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s13
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s25
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x90
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0x90
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s11
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s24
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x70
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0x70
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s9
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s23
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x50
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0x50
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s7
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s22
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 48
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 48
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s5
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s21
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 16
-; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 16
+; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s36
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s20
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0xe0
-; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s36
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0xe0
+; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s1
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s19
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0xc0
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0xc0
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s14
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s27
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0xa0
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0xa0
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s12
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s28
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x80
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0x80
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s10
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s29
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 0x60
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 0x60
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s8
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s30
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 64
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 64
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s6
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s31
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
-; GCN-HSA-NEXT:    s_add_u32 s0, s16, 32
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
+; GCN-HSA-NEXT:    s_add_u32 s2, s16, 32
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s4
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s33
-; GCN-HSA-NEXT:    s_addc_u32 s1, s17, 0
+; GCN-HSA-NEXT:    s_addc_u32 s3, s17, 0
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s2
+; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s3
+; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s35
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s34
-; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s0
+; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s2
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v4, s16
-; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s35
+; GCN-HSA-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v2, s18
 ; GCN-HSA-NEXT:    v_mov_b32_e32 v5, s17
 ; GCN-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-HSA-NEXT:    s_endpgm
 ;
 ; GCN-NOHSA-VI-LABEL: constant_zextload_v32i16_to_v32i64:
 ; GCN-NOHSA-VI:       ; %bb.0:
 ; GCN-NOHSA-VI-NEXT:    s_load_dwordx4 s[16:19], s[0:1], 0x24

 ; GCN-NOHSA-SI:       ; %bb.0:
 ; GCN-NOHSA-SI-NEXT:    s_load_dwordx4 s[16:19], s[0:1], 0x9
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NOHSA-SI-NEXT:    s_load_dwordx16 s[0:15], s[18:19], 0x0
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NOHSA-SI-NEXT:    s_mov_b32 s18, s15
 ; GCN-NOHSA-SI-NEXT:    s_mov_b32 s20, s13
 ; GCN-NOHSA-SI-NEXT:    s_mov_b32 s22, s11
-; GCN-NOHSA-SI-NEXT:    s_mov_b32 s24, s9
-; GCN-NOHSA-SI-NEXT:    s_mov_b32 s26, s7
-; GCN-NOHSA-SI-NEXT:    s_mov_b32 s28, s5
-; GCN-NOHSA-SI-NEXT:    s_mov_b32 s30, s3
-; GCN-NOHSA-SI-NEXT:    s_mov_b32 s34, s1
-; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s36, s14, 16
+; GCN-NOHSA-SI-NEXT:    s_mov_b32 s28, s9
+; GCN-NOHSA-SI-NEXT:    s_mov_b32 s24, s7
+; GCN-NOHSA-SI-NEXT:    s_mov_b32 s30, s5
+; GCN-NOHSA-SI-NEXT:    s_mov_b32 s34, s3
+; GCN-NOHSA-SI-NEXT:    s_mov_b32 s36, s1
+; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s26, s14, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s38, s12, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s40, s10, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s42, s8, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s44, s6, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s46, s4, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s48, s2, 16
 ; GCN-NOHSA-SI-NEXT:    s_lshr_b32 s50, s0, 16
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[52:53], s[20:21], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[54:55], s[18:19], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[24:25], s[24:25], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[28:29], s[28:29], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[56:57], s[22:23], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[18:19], s[0:1], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[20:21], s[2:3], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[22:23], s[4:5], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[58:59], s[6:7], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[60:61], s[8:9], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[62:63], s[10:11], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[64:65], s[12:13], 0x100000

 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v4, s52
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v5, s53
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v6, s4
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v7, s5
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v8, s56
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v9, s57
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v10, s10
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v11, s11
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v12, s24
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v13, s25
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v12, s28
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v13, s29
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v14, s2
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v15, s3
 ; GCN-NOHSA-SI-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-NOHSA-SI-NEXT:    s_mov_b32 s2, -1
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[8:9], s[34:35], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[10:11], s[30:31], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[14:15], s[28:29], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[16:17], s[26:27], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[10:11], s[36:37], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[14:15], s[34:35], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[16:17], s[30:31], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[24:25], s[24:25], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[4:5], s[50:51], 0x100000
 ; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[6:7], s[48:49], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[24:25], s[46:47], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[26:27], s[44:45], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[28:29], s[42:43], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[30:31], s[40:41], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[34:35], s[38:39], 0x100000
-; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[36:37], s[36:37], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[8:9], s[46:47], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[28:29], s[44:45], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[30:31], s[42:43], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[34:35], s[40:41], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[36:37], s[38:39], 0x100000
+; GCN-NOHSA-SI-NEXT:    s_bfe_i64 s[26:27], s[26:27], 0x100000
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:240
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:208
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], 0 offset:176
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[0:3], 0 offset:144
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(3)
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s16
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s17
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s24
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s25
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s12
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s13
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:112
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s14
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s15
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s16
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s17
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s72
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s73
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:80
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s10
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s11
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s14
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s15
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s70
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s71
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:48
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s8
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s9
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s11
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s68
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s69
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:16
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s66
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s67
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v4, s64
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v5, s65
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v8, s62
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v9, s63
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v12, s60
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v13, s61
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v16, s58
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v17, s59
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s36
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s37
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s26
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s27
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:224
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v0, s22
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v1, s23
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v6, s34
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v7, s35
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v6, s36
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v7, s37
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:192
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v4, s20
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v5, s21
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v10, s30
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v11, s31
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v10, s34
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v11, s35
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], 0 offset:160
 ; GCN-NOHSA-SI-NEXT:    s_waitcnt expcnt(0)
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v8, s18
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v9, s19
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v14, s28
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v15, s29
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v14, s30
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v15, s31
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[0:3], 0 offset:128
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v18, s26
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v19, s27
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v18, s28
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v19, s29
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[16:19], off, s[0:3], 0 offset:96
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s24
-; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s25
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v2, s8
+; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v3, s9
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:64
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v6, s6
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v7, s7
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:32
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v10, s4
 ; GCN-NOHSA-SI-NEXT:    v_mov_b32_e32 v11, s5
 ; GCN-NOHSA-SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], 0
 ; GCN-NOHSA-SI-NEXT:    s_endpgm

llvm/test/CodeGen/AMDGPU/load-constant-i8.ll

 ; GFX8-NOHSA-NEXT:    s_load_dwordx16 s[0:15], s[18:19], 0x0
 ; GFX8-NOHSA-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NOHSA-NEXT:    s_lshr_b32 s18, s0, 24
 ; GFX8-NOHSA-NEXT:    s_lshr_b32 s19, s1, 24
 ; GFX8-NOHSA-NEXT:    s_lshr_b32 s21, s2, 24
 ; GFX8-NOHSA-NEXT:    s_lshr_b32 s23, s3, 24
 ; GFX8-NOHSA-NEXT:    s_lshr_b32 s25, s4, 24
 ; GFX8-NOHSA-NEXT:    s_lshr_b32 s27, s5, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s30, s6, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s31, s7, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s33, s8, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s34, s9, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s35, s10, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s36, s11, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s37, s12, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s38, s13, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s39, s14, 24
-; GFX8-NOHSA-NEXT:    s_lshr_b32 s28, s15, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s29, s6, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s33, s7, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s34, s8, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s35, s9, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s36, s10, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s37, s11, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s38, s12, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s39, s13, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s40, s14, 24
+; GFX8-NOHSA-NEXT:    s_lshr_b32 s30, s15, 24
 ; GFX8-NOHSA-NEXT:    s_and_b32 s20, s0, 0xff
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v1, 8, s0
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s0, s0, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s22, s1, 0xff
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v3, 8, s1
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s1, s1, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s24, s2, 0xff
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v5, 8, s2
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s2, s2, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s26, s3, 0xff
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v7, 8, s3
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s3, s3, 0x80010
-; GFX8-NOHSA-NEXT:    s_and_b32 s40, s4, 0xff
+; GFX8-NOHSA-NEXT:    s_and_b32 s28, s4, 0xff
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v9, 8, s4
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s41, s4, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s4, s5, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s42, s5, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s43, s6, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s44, s6, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s45, s7, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s46, s7, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s47, s8, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s48, s8, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s49, s9, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s50, s9, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s51, s10, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s52, s10, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s53, s11, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s54, s11, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s55, s12, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s56, s12, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s57, s13, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s58, s13, 0x80010
 ; GFX8-NOHSA-NEXT:    s_and_b32 s59, s14, 0xff
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s60, s14, 0x80010
-; GFX8-NOHSA-NEXT:    s_and_b32 s29, s15, 0xff
+; GFX8-NOHSA-NEXT:    s_and_b32 s31, s15, 0xff
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s15
 ; GFX8-NOHSA-NEXT:    s_bfe_u32 s15, s15, 0x80010
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s28
-; GFX8-NOHSA-NEXT:    s_add_u32 s28, s16, 0xf0
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s29
-; GFX8-NOHSA-NEXT:    s_addc_u32 s29, s17, 0
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s28
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s30
+; GFX8-NOHSA-NEXT:    s_add_u32 s30, s16, 0xf0
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s31
+; GFX8-NOHSA-NEXT:    s_addc_u32 s31, s17, 0
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s30
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s15
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s29
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s31
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v8, s40
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v8, s28
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s14
 ; GFX8-NOHSA-NEXT:    s_add_u32 s14, s16, 0xe0
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s15, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s14
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s59
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s60
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s39
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s40
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s15
 ; GFX8-NOHSA-NEXT:    s_add_u32 s14, s16, 0xd0
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s15, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s14
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s13
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s57
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s58
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s38
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s39
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s15
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v6, s26
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s12
 ; GFX8-NOHSA-NEXT:    s_add_u32 s12, s16, 0xc0
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s13, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s13
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s12
 ; GFX8-NOHSA-NEXT:    s_add_u32 s12, s16, 0xb0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s55
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s56
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s37
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s38
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s13, s17, 0
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s13
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s11
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s53
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s54
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s36
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s37
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s12
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v4, s24
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s10
 ; GFX8-NOHSA-NEXT:    s_add_u32 s10, s16, 0xa0
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s11, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s11
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s10
 ; GFX8-NOHSA-NEXT:    s_add_u32 s10, s16, 0x90
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s51
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s52
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s35
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s36
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s11, s17, 0
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s11
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s9
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s49
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s50
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s34
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s35
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s10
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v2, s22
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s8
 ; GFX8-NOHSA-NEXT:    s_add_u32 s8, s16, 0x80
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s9, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s9
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s8
 ; GFX8-NOHSA-NEXT:    s_add_u32 s8, s16, 0x70
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s47
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s48
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s33
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s34
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s9, s17, 0
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s9
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s7
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s45
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s46
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s31
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s33
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s8
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v0, s20
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s6
 ; GFX8-NOHSA-NEXT:    s_add_u32 s6, s16, 0x60
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s7, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s7
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s43
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v12, s44
-; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s30
+; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v13, s29
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v14, s6
 ; GFX8-NOHSA-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
 ; GFX8-NOHSA-NEXT:    s_nop 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v10, s4
 ; GFX8-NOHSA-NEXT:    s_add_u32 s4, s16, 0x50
 ; GFX8-NOHSA-NEXT:    v_lshrrev_b16_e64 v11, 8, s5
 ; GFX8-NOHSA-NEXT:    s_addc_u32 s5, s17, 0
 ; GFX8-NOHSA-NEXT:    v_mov_b32_e32 v15, s5

 ; GFX6-NOHSA-NEXT:    s_endpgm
 ;
 ; GFX7-HSA-LABEL: constant_sextload_v16i8_to_v16i64:
 ; GFX7-HSA:       ; %bb.0:
 ; GFX7-HSA-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-HSA-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x0
 ; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-HSA-NEXT:    s_lshr_b32 s8, s7, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s10, s7, 8
-; GFX7-HSA-NEXT:    s_mov_b32 s12, s7
-; GFX7-HSA-NEXT:    s_lshr_b32 s14, s6, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s16, s6, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s18, s6, 8
-; GFX7-HSA-NEXT:    s_lshr_b32 s20, s5, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s22, s5, 8
-; GFX7-HSA-NEXT:    s_mov_b32 s24, s5
-; GFX7-HSA-NEXT:    s_lshr_b32 s26, s4, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s10, s7, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s12, s7, 8
+; GFX7-HSA-NEXT:    s_mov_b32 s14, s7
+; GFX7-HSA-NEXT:    s_lshr_b32 s16, s6, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s18, s6, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s20, s6, 8
+; GFX7-HSA-NEXT:    s_lshr_b32 s22, s5, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s24, s5, 8
+; GFX7-HSA-NEXT:    s_mov_b32 s26, s5
+; GFX7-HSA-NEXT:    s_lshr_b32 s8, s4, 16
 ; GFX7-HSA-NEXT:    s_lshr_b32 s28, s4, 24
 ; GFX7-HSA-NEXT:    s_lshr_b32 s30, s4, 8
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[2:3], s[4:5], 0x80000
 ; GFX7-HSA-NEXT:    s_ashr_i64 s[34:35], s[4:5], 56
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[36:37], s[6:7], 0x80000
 ; GFX7-HSA-NEXT:    s_ashr_i64 s[38:39], s[6:7], 56
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[4:5], s[30:31], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[6:7], s[28:29], 0x80000
+; GFX7-HSA-NEXT:    s_bfe_i64 s[8:9], s[8:9], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[26:27], s[26:27], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[24:25], s[24:25], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[22:23], s[22:23], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[20:21], s[20:21], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[18:19], s[18:19], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[16:17], s[16:17], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[14:15], s[14:15], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[12:13], s[12:13], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[10:11], s[10:11], 0x80000
-; GFX7-HSA-NEXT:    s_bfe_i64 s[8:9], s[8:9], 0x80000
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-HSA-NEXT:    s_add_u32 s8, s0, 0x70
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s9
-; GFX7-HSA-NEXT:    s_addc_u32 s9, s1, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s8
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s10
+; GFX7-HSA-NEXT:    s_add_u32 s10, s0, 0x70
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s11
+; GFX7-HSA-NEXT:    s_addc_u32 s11, s1, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s10
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s38
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s39
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s9
-; GFX7-HSA-NEXT:    s_add_u32 s8, s0, 0x60
-; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GFX7-HSA-NEXT:    s_addc_u32 s9, s1, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s8
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s12
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s13
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s10
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s11
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s9
-; GFX7-HSA-NEXT:    s_add_u32 s8, s0, 0x50
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-HSA-NEXT:    s_add_u32 s10, s0, 0x60
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GFX7-HSA-NEXT:    s_addc_u32 s9, s1, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s8
+; GFX7-HSA-NEXT:    s_addc_u32 s11, s1, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s10
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s14
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s15
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s16
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s17
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s9
-; GFX7-HSA-NEXT:    s_add_u32 s8, s0, 64
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s12
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s13
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-HSA-NEXT:    s_add_u32 s10, s0, 0x50
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GFX7-HSA-NEXT:    s_addc_u32 s9, s1, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s8
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s36
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s37
+; GFX7-HSA-NEXT:    s_addc_u32 s11, s1, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s10
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s16
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s17
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s18
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s19
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s9
-; GFX7-HSA-NEXT:    s_add_u32 s8, s0, 48
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-HSA-NEXT:    s_add_u32 s10, s0, 64
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GFX7-HSA-NEXT:    s_addc_u32 s9, s1, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s8
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s20
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s21
+; GFX7-HSA-NEXT:    s_addc_u32 s11, s1, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s10
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s36
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s37
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s20
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s21
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-HSA-NEXT:    s_add_u32 s10, s0, 48
+; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX7-HSA-NEXT:    s_addc_u32 s11, s1, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s10
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s22
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s23
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s34
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s35
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s9
-; GFX7-HSA-NEXT:    s_add_u32 s8, s0, 32
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-HSA-NEXT:    s_add_u32 s10, s0, 32
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
-; GFX7-HSA-NEXT:    s_addc_u32 s9, s1, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s8
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s24
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s25
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s22
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s23
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s9
+; GFX7-HSA-NEXT:    s_addc_u32 s11, s1, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s10
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s26
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s27
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s24
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s25
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s11
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_nop 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-HSA-NEXT:    s_add_u32 s6, s0, 16
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX7-HSA-NEXT:    s_addc_u32 s7, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s6
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s26
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s27
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s7
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s0

 ; GFX7-HSA-LABEL: constant_zextload_v32i8_to_v32i64:
 ; GFX7-HSA:       ; %bb.0:
 ; GFX7-HSA-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, v1
 ; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-HSA-NEXT:    s_load_dwordx8 s[4:11], s[2:3], 0x0
 ; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-HSA-NEXT:    s_lshr_b32 s3, s4, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s12, s5, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s13, s6, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s14, s7, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s15, s8, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s16, s9, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s17, s10, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s18, s11, 24
-; GFX7-HSA-NEXT:    s_bfe_u32 s19, s11, 0x80008
-; GFX7-HSA-NEXT:    s_bfe_u32 s20, s10, 0x80008
-; GFX7-HSA-NEXT:    s_bfe_u32 s21, s9, 0x80008
-; GFX7-HSA-NEXT:    s_bfe_u32 s22, s8, 0x80008
-; GFX7-HSA-NEXT:    s_bfe_u32 s23, s7, 0x80008
-; GFX7-HSA-NEXT:    s_bfe_u32 s24, s6, 0x80008
-; GFX7-HSA-NEXT:    s_bfe_u32 s25, s5, 0x80008
+; GFX7-HSA-NEXT:    s_lshr_b32 s12, s4, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s13, s5, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s14, s6, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s15, s7, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s16, s8, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s17, s9, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s18, s10, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s19, s11, 24
+; GFX7-HSA-NEXT:    s_bfe_u32 s20, s11, 0x80008
+; GFX7-HSA-NEXT:    s_bfe_u32 s21, s10, 0x80008
+; GFX7-HSA-NEXT:    s_bfe_u32 s22, s9, 0x80008
+; GFX7-HSA-NEXT:    s_bfe_u32 s23, s8, 0x80008
+; GFX7-HSA-NEXT:    s_bfe_u32 s24, s7, 0x80008
+; GFX7-HSA-NEXT:    s_bfe_u32 s25, s6, 0x80008
+; GFX7-HSA-NEXT:    s_bfe_u32 s26, s5, 0x80008
 ; GFX7-HSA-NEXT:    s_bfe_u32 s2, s4, 0x80008
-; GFX7-HSA-NEXT:    s_and_b32 s26, s4, 0xff
+; GFX7-HSA-NEXT:    s_and_b32 s3, s4, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s27, s5, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s28, s6, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s29, s7, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s30, s8, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s31, s9, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s33, s10, 0xff
 ; GFX7-HSA-NEXT:    s_and_b32 s34, s11, 0xff
 ; GFX7-HSA-NEXT:    s_bfe_u32 s35, s4, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s36, s5, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s6, s6, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s7, s7, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s8, s8, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s9, s9, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s10, s10, 0x80010
 ; GFX7-HSA-NEXT:    s_bfe_u32 s4, s11, 0x80010
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0xf0
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s18
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s19
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0xd0
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s10
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s17
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s18
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0xb0
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s9
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s16
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s17
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0x90
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s8
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s15
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s16
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0x70
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s7
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s14
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s15
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0x50
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s6
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s13
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 48
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s36
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s12
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s13
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 16
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s35
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s3
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s12
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0xe0
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s34
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s19
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0xc0
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s33
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s20
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s21
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0xa0
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s31
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s21
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s22
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0x80
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s30
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s22
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s23
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 0x60
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s29
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s23
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s24
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 64
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s28
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s24
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s25
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    s_add_u32 s4, s0, 32
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_addc_u32 s5, s1, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s27
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s25
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s26
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s1
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s26
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX7-HSA-NEXT:    s_endpgm
 ;
 ; GFX8-NOHSA-LABEL: constant_zextload_v32i8_to_v32i64:
 ; GFX8-NOHSA:       ; %bb.0:
 ; GFX8-NOHSA-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24

 ;
 ; GFX7-HSA-LABEL: constant_sextload_v32i8_to_v32i64:
 ; GFX7-HSA:       ; %bb.0:
 ; GFX7-HSA-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x0
 ; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-HSA-NEXT:    s_load_dwordx8 s[0:7], s[10:11], 0x0
 ; GFX7-HSA-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-HSA-NEXT:    s_lshr_b32 s14, s7, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s38, s7, 8
-; GFX7-HSA-NEXT:    s_mov_b32 s40, s7
-; GFX7-HSA-NEXT:    s_lshr_b32 s42, s6, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s44, s6, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s46, s6, 8
-; GFX7-HSA-NEXT:    s_lshr_b32 s48, s5, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s50, s5, 8
-; GFX7-HSA-NEXT:    s_mov_b32 s52, s5
-; GFX7-HSA-NEXT:    s_lshr_b32 s54, s4, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s56, s4, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s58, s4, 8
+; GFX7-HSA-NEXT:    s_lshr_b32 s40, s7, 8
+; GFX7-HSA-NEXT:    s_mov_b32 s42, s7
+; GFX7-HSA-NEXT:    s_lshr_b32 s44, s6, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s46, s6, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s48, s6, 8
+; GFX7-HSA-NEXT:    s_lshr_b32 s50, s5, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s52, s5, 8
+; GFX7-HSA-NEXT:    s_mov_b32 s54, s5
+; GFX7-HSA-NEXT:    s_lshr_b32 s56, s4, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s58, s4, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s60, s4, 8
 ; GFX7-HSA-NEXT:    s_lshr_b32 s36, s3, 16
 ; GFX7-HSA-NEXT:    s_lshr_b32 s30, s3, 8
 ; GFX7-HSA-NEXT:    s_mov_b32 s34, s3
 ; GFX7-HSA-NEXT:    s_lshr_b32 s28, s2, 16
 ; GFX7-HSA-NEXT:    s_lshr_b32 s26, s2, 24
 ; GFX7-HSA-NEXT:    s_lshr_b32 s22, s2, 8
 ; GFX7-HSA-NEXT:    s_lshr_b32 s18, s1, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s60, s1, 8
+; GFX7-HSA-NEXT:    s_lshr_b32 s62, s1, 8
 ; GFX7-HSA-NEXT:    s_mov_b32 s16, s1
-; GFX7-HSA-NEXT:    s_lshr_b32 s62, s0, 16
-; GFX7-HSA-NEXT:    s_lshr_b32 s64, s0, 24
-; GFX7-HSA-NEXT:    s_lshr_b32 s66, s0, 8
+; GFX7-HSA-NEXT:    s_lshr_b32 s64, s0, 16
+; GFX7-HSA-NEXT:    s_lshr_b32 s66, s0, 24
+; GFX7-HSA-NEXT:    s_lshr_b32 s68, s0, 8
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[12:13], s[2:3], 0x80000
 ; GFX7-HSA-NEXT:    s_ashr_i64 s[20:21], s[2:3], 56
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[24:25], s[4:5], 0x80000
-; GFX7-HSA-NEXT:    s_ashr_i64 s[68:69], s[4:5], 56
+; GFX7-HSA-NEXT:    s_ashr_i64 s[38:39], s[4:5], 56
 ; GFX7-HSA-NEXT:    s_ashr_i64 s[2:3], s[6:7], 56
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[4:5], s[14:15], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[10:11], s[0:1], 0x80000
 ; GFX7-HSA-NEXT:    s_ashr_i64 s[0:1], s[0:1], 56
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[70:71], s[6:7], 0x80000
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s3
-; GFX7-HSA-NEXT:    s_bfe_i64 s[2:3], s[66:67], 0x80000
-; GFX7-HSA-NEXT:    s_bfe_i64 s[4:5], s[64:65], 0x80000
-; GFX7-HSA-NEXT:    s_bfe_i64 s[6:7], s[62:63], 0x80000
+; GFX7-HSA-NEXT:    s_bfe_i64 s[2:3], s[68:69], 0x80000
+; GFX7-HSA-NEXT:    s_bfe_i64 s[4:5], s[66:67], 0x80000
+; GFX7-HSA-NEXT:    s_bfe_i64 s[6:7], s[64:65], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[16:17], s[16:17], 0x80000
-; GFX7-HSA-NEXT:    s_bfe_i64 s[14:15], s[60:61], 0x80000
+; GFX7-HSA-NEXT:    s_bfe_i64 s[14:15], s[62:63], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[18:19], s[18:19], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[22:23], s[22:23], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[26:27], s[26:27], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[28:29], s[28:29], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[34:35], s[34:35], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[30:31], s[30:31], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[36:37], s[36:37], 0x80000
+; GFX7-HSA-NEXT:    s_bfe_i64 s[60:61], s[60:61], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[58:59], s[58:59], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[56:57], s[56:57], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[54:55], s[54:55], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[52:53], s[52:53], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[50:51], s[50:51], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[48:49], s[48:49], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[46:47], s[46:47], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[44:45], s[44:45], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[42:43], s[42:43], 0x80000
 ; GFX7-HSA-NEXT:    s_bfe_i64 s[40:41], s[40:41], 0x80000
-; GFX7-HSA-NEXT:    s_bfe_i64 s[38:39], s[38:39], 0x80000
-; GFX7-HSA-NEXT:    s_add_u32 s60, s8, 0xf0
-; GFX7-HSA-NEXT:    s_addc_u32 s61, s9, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v6, s38
-; GFX7-HSA-NEXT:    s_add_u32 s38, s8, 0xe0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v7, s39
-; GFX7-HSA-NEXT:    s_addc_u32 s39, s9, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v24, s38
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v25, s39
-; GFX7-HSA-NEXT:    s_add_u32 s38, s8, 0xd0
-; GFX7-HSA-NEXT:    s_addc_u32 s39, s9, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v26, s38
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v27, s39
-; GFX7-HSA-NEXT:    s_add_u32 s38, s8, 0xc0
-; GFX7-HSA-NEXT:    s_addc_u32 s39, s9, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v28, s38
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v29, s39
+; GFX7-HSA-NEXT:    s_add_u32 s62, s8, 0xf0
+; GFX7-HSA-NEXT:    s_addc_u32 s63, s9, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v6, s40
+; GFX7-HSA-NEXT:    s_add_u32 s40, s8, 0xe0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v7, s41
+; GFX7-HSA-NEXT:    s_addc_u32 s41, s9, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v24, s40
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v25, s41
+; GFX7-HSA-NEXT:    s_add_u32 s40, s8, 0xd0
+; GFX7-HSA-NEXT:    s_addc_u32 s41, s9, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v26, s40
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v27, s41
+; GFX7-HSA-NEXT:    s_add_u32 s40, s8, 0xc0
+; GFX7-HSA-NEXT:    s_addc_u32 s41, s9, 0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v18, s38
 ; GFX7-HSA-NEXT:    s_add_u32 s38, s8, 0xb0
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v19, s39
 ; GFX7-HSA-NEXT:    s_addc_u32 s39, s9, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v30, s38
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v31, s39
 ; GFX7-HSA-NEXT:    s_add_u32 s38, s8, 0xa0
 ; GFX7-HSA-NEXT:    s_addc_u32 s39, s9, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v32, s38
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v33, s39
 ; GFX7-HSA-NEXT:    s_add_u32 s38, s8, 0x90
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v22, s60
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s40
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s41
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v22, s62
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s42
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s43
 ; GFX7-HSA-NEXT:    s_addc_u32 s39, s9, 0
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v23, s61
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v8, s42
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v9, s43
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v10, s44
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v11, s45
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v23, s63
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v8, s44
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v9, s45
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v10, s46
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v11, s47
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v28, s40
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[24:25], v[4:7]
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v12, s70
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v4, s24
 ; GFX7-HSA-NEXT:    s_add_u32 s24, s8, 0x80
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v13, s71
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v14, s46
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v15, s47
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v16, s48
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v17, s49
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v18, s68
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v19, s69
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v14, s48
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v15, s49
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v29, s41
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v16, s50
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v17, s51
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[22:23], v[0:3]
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[26:27], v[8:11]
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s56
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s58
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v5, s25
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v8, s38
 ; GFX7-HSA-NEXT:    s_addc_u32 s25, s9, 0
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v10, s24
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v20, s52
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v21, s53
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v22, s50
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v23, s51
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s54
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s55
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s57
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v20, s54
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v21, s55
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v22, s52
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v23, s53
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v0, s56
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v1, s57
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s59
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v9, s39
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[28:29], v[12:15]
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v6, s58
-; GFX7-HSA-NEXT:    v_mov_b32_e32 v7, s59
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v6, s60
+; GFX7-HSA-NEXT:    v_mov_b32_e32 v7, s61
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v11, s25
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[30:31], v[16:19]
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[32:33], v[20:23]
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
 ; GFX7-HSA-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-HSA-NEXT:    s_add_u32 s20, s8, 0x70
 ; GFX7-HSA-NEXT:    v_mov_b32_e32 v3, s21

llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll

 
 ; Has no stack objects, but introduces them due to the CSR spill. We
 ; see the FP modified in the callee with IPRA. We should not have
 ; redundant spills of s33 or assert.
 define internal fastcc void @csr_vgpr_spill_fp_callee() #0 {
 ; CHECK-LABEL: csr_vgpr_spill_fp_callee:
 ; CHECK:       ; %bb.0: ; %bb
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT:    s_mov_b32 s18, s33
+; CHECK-NEXT:    s_mov_b32 s24, s33
 ; CHECK-NEXT:    s_mov_b32 s33, s32
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[16:17], -1
 ; CHECK-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
 ; CHECK-NEXT:    s_mov_b64 exec, s[16:17]
 ; CHECK-NEXT:    s_add_i32 s32, s32, 0x400
 ; CHECK-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; CHECK-NEXT:    v_writelane_b32 v1, s30, 0
 ; CHECK-NEXT:    v_writelane_b32 v1, s31, 1

 ; CHECK-NEXT:    ;;#ASMEND
 ; CHECK-NEXT:    v_readlane_b32 s31, v1, 1
 ; CHECK-NEXT:    v_readlane_b32 s30, v1, 0
 ; CHECK-NEXT:    buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; CHECK-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
 ; CHECK-NEXT:    s_mov_b64 exec, s[4:5]
 ; CHECK-NEXT:    s_add_i32 s32, s32, 0xfffffc00
-; CHECK-NEXT:    s_mov_b32 s33, s18
+; CHECK-NEXT:    s_mov_b32 s33, s24
 ; CHECK-NEXT:    s_waitcnt vmcnt(0)
 ; CHECK-NEXT:    s_setpc_b64 s[30:31]
 bb:
   call fastcc void @callee_has_fp()
   call void asm sideeffect "; clobber csr v40", "~{v40}"()
   ret void
 }
 

 ; CHECK-NEXT:    ;;#ASMSTART
 ; CHECK-NEXT:    ; clobber csr v40
 ; CHECK-NEXT:    ;;#ASMEND
 ; CHECK-NEXT:    s_getpc_b64 s[16:17]
 ; CHECK-NEXT:    s_add_u32 s16, s16, callee_has_fp@rel32@lo+4
 ; CHECK-NEXT:    s_addc_u32 s17, s17, callee_has_fp@rel32@hi+12
 ; CHECK-NEXT:    v_readlane_b32 s33, v1, 0
 ; CHECK-NEXT:    buffer_load_dword v40, off, s[0:3], s32 ; 4-byte Folded Reload
-; CHECK-NEXT:    s_xor_saveexec_b64 s[18:19], -1
+; CHECK-NEXT:    s_xor_saveexec_b64 s[20:21], -1
 ; CHECK-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
-; CHECK-NEXT:    s_mov_b64 exec, s[18:19]
+; CHECK-NEXT:    s_mov_b64 exec, s[20:21]
 ; CHECK-NEXT:    s_setpc_b64 s[16:17]
 bb:
   call void asm sideeffect "; clobber csr v40", "~{v40}"()
   tail call fastcc void @callee_has_fp()
   ret void
 }
 
 define amdgpu_kernel void @kernel_tailcall() {

 entry:
   ret i32 0
 }
 
 define hidden i32 @caller_save_vgpr_spill_fp_tail_call() #0 {
 ; CHECK-LABEL: caller_save_vgpr_spill_fp_tail_call:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT:    s_mov_b32 s18, s33
+; CHECK-NEXT:    s_mov_b32 s24, s33
 ; CHECK-NEXT:    s_mov_b32 s33, s32
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[16:17], -1
 ; CHECK-NEXT:    buffer_store_dword v1, off, s[0:3], s33 ; 4-byte Folded Spill
 ; CHECK-NEXT:    s_mov_b64 exec, s[16:17]
 ; CHECK-NEXT:    s_add_i32 s32, s32, 0x400
 ; CHECK-NEXT:    v_writelane_b32 v1, s30, 0
 ; CHECK-NEXT:    v_writelane_b32 v1, s31, 1
 ; CHECK-NEXT:    s_getpc_b64 s[16:17]
 ; CHECK-NEXT:    s_add_u32 s16, s16, tail_call@rel32@lo+4
 ; CHECK-NEXT:    s_addc_u32 s17, s17, tail_call@rel32@hi+12
 ; CHECK-NEXT:    s_mov_b64 s[22:23], s[2:3]
 ; CHECK-NEXT:    s_mov_b64 s[20:21], s[0:1]
 ; CHECK-NEXT:    s_mov_b64 s[0:1], s[20:21]
 ; CHECK-NEXT:    s_mov_b64 s[2:3], s[22:23]
 ; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; CHECK-NEXT:    v_readlane_b32 s31, v1, 1
 ; CHECK-NEXT:    v_readlane_b32 s30, v1, 0
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; CHECK-NEXT:    buffer_load_dword v1, off, s[0:3], s33 ; 4-byte Folded Reload
 ; CHECK-NEXT:    s_mov_b64 exec, s[4:5]
 ; CHECK-NEXT:    s_add_i32 s32, s32, 0xfffffc00
-; CHECK-NEXT:    s_mov_b32 s33, s18
+; CHECK-NEXT:    s_mov_b32 s33, s24
 ; CHECK-NEXT:    s_waitcnt vmcnt(0)
 ; CHECK-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %call = call i32 @tail_call()
   ret i32 %call
 }
 
 define hidden i32 @caller_save_vgpr_spill_fp() #0 {
 ; CHECK-LABEL: caller_save_vgpr_spill_fp:
 ; CHECK:       ; %bb.0: ; %entry
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT:    s_mov_b32 s19, s33
+; CHECK-NEXT:    s_mov_b32 s25, s33
 ; CHECK-NEXT:    s_mov_b32 s33, s32
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[16:17], -1
 ; CHECK-NEXT:    buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
 ; CHECK-NEXT:    s_mov_b64 exec, s[16:17]
 ; CHECK-NEXT:    s_add_i32 s32, s32, 0x400
 ; CHECK-NEXT:    v_writelane_b32 v2, s30, 0
 ; CHECK-NEXT:    v_writelane_b32 v2, s31, 1
 ; CHECK-NEXT:    s_getpc_b64 s[16:17]
 ; CHECK-NEXT:    s_add_u32 s16, s16, caller_save_vgpr_spill_fp_tail_call@rel32@lo+4
 ; CHECK-NEXT:    s_addc_u32 s17, s17, caller_save_vgpr_spill_fp_tail_call@rel32@hi+12
 ; CHECK-NEXT:    s_mov_b64 s[22:23], s[2:3]
 ; CHECK-NEXT:    s_mov_b64 s[20:21], s[0:1]
 ; CHECK-NEXT:    s_mov_b64 s[0:1], s[20:21]
 ; CHECK-NEXT:    s_mov_b64 s[2:3], s[22:23]
 ; CHECK-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; CHECK-NEXT:    v_readlane_b32 s31, v2, 1
 ; CHECK-NEXT:    v_readlane_b32 s30, v2, 0
 ; CHECK-NEXT:    s_xor_saveexec_b64 s[4:5], -1
 ; CHECK-NEXT:    buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
 ; CHECK-NEXT:    s_mov_b64 exec, s[4:5]
 ; CHECK-NEXT:    s_add_i32 s32, s32, 0xfffffc00
-; CHECK-NEXT:    s_mov_b32 s33, s19
+; CHECK-NEXT:    s_mov_b32 s33, s25
 ; CHECK-NEXT:    s_waitcnt vmcnt(0)
 ; CHECK-NEXT:    s_setpc_b64 s[30:31]
 entry:
   %call = call i32 @caller_save_vgpr_spill_fp_tail_call()
   ret i32 %call
 }
 
 define protected amdgpu_kernel void @kernel() {

llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll

; GCN-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "", "~{vcc}" () #0
   ret void
 }
 
 define void @spill_sgpr_with_no_lower_vgpr_available() #0 {
 ; GCN-LABEL: spill_sgpr_with_no_lower_vgpr_available:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s18, s33
+; GCN-NEXT:    s_mov_b32 s24, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_or_saveexec_b64 s[16:17], -1
 ; GCN-NEXT:    buffer_store_dword v255, off, s[0:3], s33 offset:452 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[16:17]
 ; GCN-NEXT:    s_add_i32 s32, s32, 0x7400
 ; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:440 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v41, off, s[0:3], s33 offset:436 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v42, off, s[0:3], s33 offset:432 ; 4-byte Folded Spill

 ; GCN-NEXT:    buffer_load_dword v43, off, s[0:3], s33 offset:428 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v42, off, s[0:3], s33 offset:432 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v41, off, s[0:3], s33 offset:436 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:440 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v255, off, s[0:3], s33 offset:452 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_add_i32 s32, s32, 0xffff8c00
-; GCN-NEXT:    s_mov_b32 s33, s18
+; GCN-NEXT:    s_mov_b32 s33, s24
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   %alloca = alloca i32, align 4, addrspace(5)
   store volatile i32 0, ptr addrspace(5) %alloca
 
   call void asm sideeffect "",
   "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9}
   ,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19}
; GCN-NEXT:    s_setpc_b64 s[30:31]
   call void @child_function()
   ret void
 }
 
 define void @spill_to_lowest_available_vgpr() #0 {
 ; GCN-LABEL: spill_to_lowest_available_vgpr:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s18, s33
+; GCN-NEXT:    s_mov_b32 s24, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_or_saveexec_b64 s[16:17], -1
 ; GCN-NEXT:    buffer_store_dword v254, off, s[0:3], s33 offset:448 ; 4-byte Folded Spill
 ; GCN-NEXT:    s_mov_b64 exec, s[16:17]
 ; GCN-NEXT:    s_add_i32 s32, s32, 0x7400
 ; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:436 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v41, off, s[0:3], s33 offset:432 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v42, off, s[0:3], s33 offset:428 ; 4-byte Folded Spill

 ; GCN-NEXT:    buffer_load_dword v43, off, s[0:3], s33 offset:424 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v42, off, s[0:3], s33 offset:428 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v41, off, s[0:3], s33 offset:432 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:436 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_or_saveexec_b64 s[4:5], -1
 ; GCN-NEXT:    buffer_load_dword v254, off, s[0:3], s33 offset:448 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_add_i32 s32, s32, 0xffff8c00
-; GCN-NEXT:    s_mov_b32 s33, s18
+; GCN-NEXT:    s_mov_b32 s33, s24
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   %alloca = alloca i32, align 4, addrspace(5)
   store volatile i32 0, ptr addrspace(5) %alloca
 
   call void asm sideeffect "",
   "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9}
   ,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19}
; GCN-NEXT:    s_setpc_b64 s[30:31]
   ,~{v250},~{v251},~{v252},~{v253},~{v254},~{v255}" () #0
   ret void
 }
 
 define void @spill_sgpr_no_free_vgpr_ipra() #0 {
 ; GCN-LABEL: spill_sgpr_no_free_vgpr_ipra:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s18, s33
+; GCN-NEXT:    s_mov_b32 s24, s33
 ; GCN-NEXT:    s_mov_b32 s33, s32
 ; GCN-NEXT:    s_add_i32 s32, s32, 0x7400
 ; GCN-NEXT:    buffer_store_dword v40, off, s[0:3], s33 offset:444 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v41, off, s[0:3], s33 offset:440 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v42, off, s[0:3], s33 offset:436 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v43, off, s[0:3], s33 offset:432 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v44, off, s[0:3], s33 offset:428 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v45, off, s[0:3], s33 offset:424 ; 4-byte Folded Spill

 ; GCN-NEXT:    buffer_load_dword v46, off, s[0:3], s33 offset:420 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v45, off, s[0:3], s33 offset:424 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v44, off, s[0:3], s33 offset:428 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v43, off, s[0:3], s33 offset:432 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v42, off, s[0:3], s33 offset:436 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v41, off, s[0:3], s33 offset:440 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v40, off, s[0:3], s33 offset:444 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_add_i32 s32, s32, 0xffff8c00
-; GCN-NEXT:    s_mov_b32 s33, s18
+; GCN-NEXT:    s_mov_b32 s33, s24
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   call void @child_function_ipra()
   ret void
 }
 
 define internal void @child_function_ipra_tail_call() #0 {
 ; GCN-LABEL: child_function_ipra_tail_call:

llvm/test/CodeGen/AMDGPU/spill-offset-calculation.ll

 define void @test_sgpr_offset_function_scavenge_fail_func() #2 {
 ; MUBUF-LABEL: test_sgpr_offset_function_scavenge_fail_func:
 ; MUBUF:       ; %bb.0: ; %entry
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:8 glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    v_mov_b32_e32 v1, 0x1004
-; MUBUF-NEXT:    buffer_store_dword v0, v1, s[0:3], s32 offen ; 4-byte Folded Spill
+; MUBUF-NEXT:    s_add_i32 s10, s32, 0x40100
+; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s10 ; 4-byte Folded Spill
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
-; MUBUF-NEXT:    v_mov_b32_e32 v1, 0x1004
-; MUBUF-NEXT:    buffer_load_dword v0, v1, s[0:3], s32 offen ; 4-byte Folded Reload
+; MUBUF-NEXT:    s_add_i32 s10, s32, 0x40100
+; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s10 ; 4-byte Folded Reload
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; FLATSCR-LABEL: test_sgpr_offset_function_scavenge_fail_func:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)

 ; MUBUF-LABEL: test_sgpr_offset_function_scavenge_fail_kernel:
 ; MUBUF:       ; %bb.0: ; %entry
 ; MUBUF-NEXT:    s_add_u32 s0, s0, s7
 ; MUBUF-NEXT:    s_addc_u32 s1, s1, 0
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    v_mov_b32_e32 v1, 0x1004
-; MUBUF-NEXT:    buffer_store_dword v0, v1, s[0:3], 0 offen ; 4-byte Folded Spill
+; MUBUF-NEXT:    s_mov_b32 s10, 0x40100
+; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s10 ; 4-byte Folded Spill
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
-; MUBUF-NEXT:    v_mov_b32_e32 v1, 0x1004
-; MUBUF-NEXT:    buffer_load_dword v0, v1, s[0:3], 0 offen ; 4-byte Folded Reload
+; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s10 ; 4-byte Folded Reload
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: test_sgpr_offset_function_scavenge_fail_kernel:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
entry:
   ; Ensure the spill is of the full super-reg.
   call void asm sideeffect "; $0", "r"(<2 x i32> %a)
 
   ret void
 }
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind "amdgpu-num-sgpr"="17" "amdgpu-num-vgpr"="8" }
-attributes #2 = { nounwind "amdgpu-num-sgpr"="14" "amdgpu-num-vgpr"="8" }
-attributes #3 = { nounwind "amdgpu-num-sgpr"="16" "amdgpu-num-vgpr"="8" }
+attributes #2 = { nounwind "amdgpu-num-sgpr"="16" "amdgpu-num-vgpr"="8" }
+attributes #3 = { nounwind "amdgpu-num-sgpr"="18" "amdgpu-num-vgpr"="8" }

llvm/test/CodeGen/AMDGPU/spill-scavenge-offset.ll

 ; GFX6-NEXT:    s_mov_b32 s41, SCRATCH_RSRC_DWORD1
 ; GFX6-NEXT:    s_mov_b32 s42, -1
 ; GFX6-NEXT:    s_mov_b32 s43, 0xe8f000
 ; GFX6-NEXT:    s_add_u32 s40, s40, s3
 ; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GFX6-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, -1, 0
 ; GFX6-NEXT:    v_mbcnt_hi_u32_b32_e32 v5, -1, v0
 ; GFX6-NEXT:    v_mov_b32_e32 v6, 0
-; GFX6-NEXT:    s_mov_b32 s38, 0
-; GFX6-NEXT:    s_mov_b32 s39, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_mov_b64 s[36:37], s[2:3]
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v7, 8, v5
 ; GFX6-NEXT:    v_mov_b32_e32 v8, v6
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:240
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:240
 ; GFX6-NEXT:    s_addc_u32 s41, s41, 0
+; GFX6-NEXT:    s_mov_b32 s2, 0x83800
+; GFX6-NEXT:    s_mov_b64 s[8:9], exec
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT:    s_waitcnt expcnt(0)
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:224
 ; GFX6-NEXT:    s_mov_b32 s2, 0x83400
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:224
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:208
 ; GFX6-NEXT:    s_mov_b32 s2, 0x83000
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:208
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:192
 ; GFX6-NEXT:    s_mov_b32 s2, 0x82c00
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:192
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:176
 ; GFX6-NEXT:    s_mov_b32 s2, 0x82800
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:176
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:160
 ; GFX6-NEXT:    s_mov_b32 s2, 0x82400
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:160
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:144
 ; GFX6-NEXT:    s_mov_b32 s2, 0x82000
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:144
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:128
 ; GFX6-NEXT:    s_mov_b32 s2, 0x81c00
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:128
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:112
 ; GFX6-NEXT:    s_mov_b32 s2, 0x81800
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:112
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:96
 ; GFX6-NEXT:    s_mov_b32 s2, 0x81400
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:96
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:80
 ; GFX6-NEXT:    s_mov_b32 s2, 0x81000
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:80
-; GFX6-NEXT:    s_mov_b32 s2, 0x80c00
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
-; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
-; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:64
-; GFX6-NEXT:    s_mov_b32 s2, 0x80400
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64 offset:64
+; GFX6-NEXT:    s_mov_b32 s2, 0x80800
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64
-; GFX6-NEXT:    buffer_load_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:16
-; GFX6-NEXT:    s_mov_b32 s2, 0x80800
+; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[4:7], 0 addr64
+; GFX6-NEXT:    buffer_load_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:16
+; GFX6-NEXT:    s_mov_b32 s2, 0x80c00
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v9, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v10, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v11, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v12, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
-; GFX6-NEXT:    buffer_load_dwordx4 v[13:16], v[7:8], s[36:39], 0 addr64 offset:32
-; GFX6-NEXT:    buffer_load_dwordx4 v[17:20], v[7:8], s[36:39], 0 addr64 offset:48
+; GFX6-NEXT:    buffer_load_dwordx4 v[13:16], v[7:8], s[4:7], 0 addr64 offset:32
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_mov_b64 exec, 15
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
+; GFX6-NEXT:    s_waitcnt expcnt(0)
+; GFX6-NEXT:    v_writelane_b32 v4, s0, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s1, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s2, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s3, 3
+; GFX6-NEXT:    s_mov_b32 s10, 0x80400
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s10 ; 4-byte Folded Spill
+; GFX6-NEXT:    s_waitcnt expcnt(0)
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_mov_b64 exec, s[8:9]
+; GFX6-NEXT:    buffer_load_dwordx4 v[17:20], v[7:8], s[4:7], 0 addr64 offset:48
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v4, 13, v0
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 16, v4
 ; GFX6-NEXT:    v_mov_b32_e32 v7, 1
 ; GFX6-NEXT:    s_mov_b64 s[2:3], exec
 ; GFX6-NEXT:    buffer_store_dword v7, v4, s[40:43], 0 offen
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[4:11]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s4, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s5, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s6, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s7, 3
 ; GFX6-NEXT:    v_writelane_b32 v4, s8, 4
 ; GFX6-NEXT:    v_writelane_b32 v4, s9, 5
 ; GFX6-NEXT:    v_writelane_b32 v4, s10, 6
 ; GFX6-NEXT:    v_writelane_b32 v4, s11, 7
-; GFX6-NEXT:    s_mov_b32 s12, 0x83800
+; GFX6-NEXT:    s_mov_b32 s12, 0x83c00
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[2:3]
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[8:15]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[16:23]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[24:31]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[4:7]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[2:3]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
-; GFX6-NEXT:    ; def s[36:37]
-; GFX6-NEXT:    ;;#ASMEND
-; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s33
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    s_and_saveexec_b64 s[34:35], vcc
 ; GFX6-NEXT:    s_cbranch_execz .LBB1_2
 ; GFX6-NEXT:  ; %bb.1: ; %bb0
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s8, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s9, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s10, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s11, 3
 ; GFX6-NEXT:    v_writelane_b32 v4, s12, 4
 ; GFX6-NEXT:    v_writelane_b32 v4, s13, 5
 ; GFX6-NEXT:    v_writelane_b32 v4, s14, 6
 ; GFX6-NEXT:    v_writelane_b32 v4, s15, 7
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2100
-; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    s_mov_b32 s38, 0x84400
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s38 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x20e0
+; GFX6-NEXT:    s_mov_b32 s38, 0x83c00
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s38 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s8, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s9, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s10, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s11, v4, 3
 ; GFX6-NEXT:    v_readlane_b32 s12, v4, 4
 ; GFX6-NEXT:    v_readlane_b32 s13, v4, 5
 ; GFX6-NEXT:    v_readlane_b32 s14, v4, 6
 ; GFX6-NEXT:    v_readlane_b32 s15, v4, 7
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s16, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s17, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s18, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s19, 3
 ; GFX6-NEXT:    v_writelane_b32 v4, s20, 4
 ; GFX6-NEXT:    v_writelane_b32 v4, s21, 5
 ; GFX6-NEXT:    v_writelane_b32 v4, s22, 6
 ; GFX6-NEXT:    v_writelane_b32 v4, s23, 7
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2120
-; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    s_mov_b32 s38, 0x84c00
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s38 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2100
+; GFX6-NEXT:    s_mov_b32 s38, 0x84400
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s38 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s16, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s17, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s18, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s19, v4, 3
 ; GFX6-NEXT:    v_readlane_b32 s20, v4, 4
 ; GFX6-NEXT:    v_readlane_b32 s21, v4, 5
 ; GFX6-NEXT:    v_readlane_b32 s22, v4, 6
 ; GFX6-NEXT:    v_readlane_b32 s23, v4, 7
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s24, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s25, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s26, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s27, 3
 ; GFX6-NEXT:    v_writelane_b32 v4, s28, 4
 ; GFX6-NEXT:    v_writelane_b32 v4, s29, 5
 ; GFX6-NEXT:    v_writelane_b32 v4, s30, 6
 ; GFX6-NEXT:    v_writelane_b32 v4, s31, 7
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2140
-; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    s_mov_b32 s38, 0x85400
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s38 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2120
+; GFX6-NEXT:    s_mov_b32 s38, 0x84c00
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s38 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s24, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s25, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s26, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s27, v4, 3
 ; GFX6-NEXT:    v_readlane_b32 s28, v4, 4
 ; GFX6-NEXT:    v_readlane_b32 s29, v4, 5
 ; GFX6-NEXT:    v_readlane_b32 s30, v4, 6
 ; GFX6-NEXT:    v_readlane_b32 s31, v4, 7
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s0, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s1, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s2, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s3, 3
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2160
-; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    s_mov_b32 s38, 0x85c00
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s38 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
 ; GFX6-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s4, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s5, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s6, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s7, 3
-; GFX6-NEXT:    s_mov_b32 s44, 0x85c00
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s44 ; 4-byte Folded Spill
+; GFX6-NEXT:    s_mov_b32 s36, 0x86000
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s36 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX6-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 3
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s2, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s3, 1
-; GFX6-NEXT:    s_mov_b32 s4, 0x86600
+; GFX6-NEXT:    s_mov_b32 s4, 0x86400
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 s[36:37], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2140
+; GFX6-NEXT:    s_mov_b32 s38, 0x85400
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s38 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s0, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s1, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s2, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s3, v4, 3
 ; GFX6-NEXT:    v_readlane_b32 s4, v4, 4
 ; GFX6-NEXT:    v_readlane_b32 s5, v4, 5
 ; GFX6-NEXT:    v_readlane_b32 s6, v4, 6
 ; GFX6-NEXT:    v_readlane_b32 s7, v4, 7
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
+; GFX6-NEXT:    s_mov_b64 exec, s[36:37]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v4, s36, 0
-; GFX6-NEXT:    v_writelane_b32 v4, s37, 1
-; GFX6-NEXT:    v_writelane_b32 v4, s38, 2
-; GFX6-NEXT:    v_writelane_b32 v4, s39, 3
 ; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2180
-; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[38:39], exec
-; GFX6-NEXT:    s_mov_b64 exec, 3
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v4, s36, 0
-; GFX6-NEXT:    v_writelane_b32 v4, s37, 1
-; GFX6-NEXT:    s_mov_b32 s44, 0x86400
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s44 ; 4-byte Folded Spill
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[38:39]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
-; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2170
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s36, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s37, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s38, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s39, v4, 3
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2190
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
-; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2190
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
-; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s44, v4, 0
-; GFX6-NEXT:    v_readlane_b32 s45, v4, 1
-; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_not_b64 exec, exec
 ; GFX6-NEXT:    s_mov_b64 vcc, s[34:35]
-; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2198
+; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 exec, 3
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2190
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
-; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2198
-; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
-; GFX6-NEXT:    s_not_b64 exec, exec
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s34, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s35, v4, 1
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_not_b64 exec, exec
+; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    ;;#ASMSTART
-; GFX6-NEXT:    ; use s[8:15],s[16:23],s[24:31],s[0:7],s[36:39],s[34:35],s[44:45]
+; GFX6-NEXT:    ; use s[8:15],s[16:23],s[24:31],s[0:7],s[36:39],s[34:35]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    s_mov_b64 s[34:35], vcc
-; GFX6-NEXT:    s_mov_b64 s[0:1], exec
-; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    s_mov_b32 s2, 0x86000
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s2 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s36, v4, 0
-; GFX6-NEXT:    v_readlane_b32 s37, v4, 1
-; GFX6-NEXT:    v_readlane_b32 s38, v4, 2
-; GFX6-NEXT:    v_readlane_b32 s39, v4, 3
-; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
-; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX6-NEXT:    s_mov_b64 s[4:5], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    s_mov_b32 s6, 0x85800
+; GFX6-NEXT:    s_mov_b32 s6, 0x85c00
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s6 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s0, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s1, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s2, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s3, v4, 3
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[4:5]
-; GFX6-NEXT:    s_mov_b32 s2, 0x83800
+; GFX6-NEXT:    s_mov_b32 s2, 0x83c00
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
-; GFX6-NEXT:    s_mov_b32 s2, 0x84000
+; GFX6-NEXT:    s_mov_b32 s2, 0x84400
 ; GFX6-NEXT:    buffer_store_dword v13, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v14, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v15, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v16, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
-; GFX6-NEXT:    s_mov_b32 s2, 0x84800
+; GFX6-NEXT:    s_mov_b32 s2, 0x84c00
 ; GFX6-NEXT:    buffer_store_dword v17, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v18, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v19, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v20, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    buffer_load_dword v17, off, s[40:43], s2 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v18, off, s[40:43], s2 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v19, off, s[40:43], s2 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v20, off, s[40:43], s2 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s2, 0x84000
+; GFX6-NEXT:    s_mov_b32 s2, 0x84400
 ; GFX6-NEXT:    buffer_load_dword v13, off, s[40:43], s2 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v14, off, s[40:43], s2 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v15, off, s[40:43], s2 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v16, off, s[40:43], s2 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s2, 0x83800
+; GFX6-NEXT:    s_mov_b32 s2, 0x83c00
 ; GFX6-NEXT:    buffer_load_dword v0, off, s[40:43], s2 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Reload
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:  .LBB1_2: ; %ret
 ; GFX6-NEXT:    s_or_b64 exec, exec, s[34:35]
-; GFX6-NEXT:    s_mov_b32 s4, 0x83400
+; GFX6-NEXT:    s_mov_b64 s[2:3], exec
+; GFX6-NEXT:    s_mov_b64 exec, 15
+; GFX6-NEXT:    s_mov_b32 s8, 0x80400
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
+; GFX6-NEXT:    s_waitcnt expcnt(0)
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s8 ; 4-byte Folded Reload
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_readlane_b32 s4, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s5, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s6, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s7, v4, 3
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_mov_b64 exec, s[2:3]
+; GFX6-NEXT:    s_mov_b32 s4, 0x83800
 ; GFX6-NEXT:    v_lshl_b64 v[4:5], v[5:6], 8
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b64 s[2:3], s[38:39]
-; GFX6-NEXT:    s_mov_b32 s4, 0x83000
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_mov_b32 s4, 0x83400
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:240
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x82c00
+; GFX6-NEXT:    s_mov_b32 s4, 0x83000
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:224
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x82800
+; GFX6-NEXT:    s_mov_b32 s4, 0x82c00
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:208
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x82400
+; GFX6-NEXT:    s_mov_b32 s4, 0x82800
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:192
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x82000
+; GFX6-NEXT:    s_mov_b32 s4, 0x82400
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:176
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x81c00
+; GFX6-NEXT:    s_mov_b32 s4, 0x82000
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:160
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x81800
+; GFX6-NEXT:    s_mov_b32 s4, 0x81c00
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:144
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x81400
+; GFX6-NEXT:    s_mov_b32 s4, 0x81800
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:128
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x81000
+; GFX6-NEXT:    s_mov_b32 s4, 0x81400
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:112
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x80c00
+; GFX6-NEXT:    s_mov_b32 s4, 0x81000
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:96
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x80400
+; GFX6-NEXT:    s_mov_b32 s4, 0x80800
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:80
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
-; GFX6-NEXT:    s_mov_b32 s4, 0x80800
+; GFX6-NEXT:    s_mov_b32 s4, 0x80c00
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:64
 ; GFX6-NEXT:    buffer_store_dwordx4 v[17:20], v[4:5], s[0:3], 0 addr64 offset:48
 ; GFX6-NEXT:    buffer_store_dwordx4 v[13:16], v[4:5], s[0:3], 0 addr64 offset:32
 ; GFX6-NEXT:    s_waitcnt expcnt(2)
 ; GFX6-NEXT:    buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
 ; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload

 ; GFX9-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX9-FLATSCR-NEXT:    ;;#ASMSTART
 ; GFX9-FLATSCR-NEXT:    ; def s[40:43]
 ; GFX9-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX9-FLATSCR-NEXT:    ;;#ASMSTART
 ; GFX9-FLATSCR-NEXT:    ; def s[38:39]
 ; GFX9-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX9-FLATSCR-NEXT:    ;;#ASMSTART
-; GFX9-FLATSCR-NEXT:    ; def s[44:45]
-; GFX9-FLATSCR-NEXT:    ;;#ASMEND
-; GFX9-FLATSCR-NEXT:    ;;#ASMSTART
 ; GFX9-FLATSCR-NEXT:    ; def s33
 ; GFX9-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX9-FLATSCR-NEXT:    s_and_saveexec_b64 s[34:35], vcc
 ; GFX9-FLATSCR-NEXT:    s_cbranch_execz .LBB1_2
 ; GFX9-FLATSCR-NEXT:  ; %bb.1: ; %bb0
 ; GFX9-FLATSCR-NEXT:    ;;#ASMSTART
-; GFX9-FLATSCR-NEXT:    ; use s[0:7],s[8:15],s[16:23],s[24:31],s[40:43],s[38:39],s[44:45]
+; GFX9-FLATSCR-NEXT:    ; use s[0:7],s[8:15],s[16:23],s[24:31],s[40:43],s[38:39]
 ; GFX9-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX9-FLATSCR-NEXT:    s_movk_i32 s0, 0x20d0
 ; GFX9-FLATSCR-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 ; 16-byte Folded Spill
 ; GFX9-FLATSCR-NEXT:    s_movk_i32 s0, 0x20e0
 ; GFX9-FLATSCR-NEXT:    scratch_store_dwordx4 off, v[16:19], s0 ; 16-byte Folded Spill
 ; GFX9-FLATSCR-NEXT:    s_movk_i32 s0, 0x20f0
 ; GFX9-FLATSCR-NEXT:    scratch_store_dwordx4 off, v[20:23], s0 ; 16-byte Folded Spill
 ; GFX9-FLATSCR-NEXT:    s_movk_i32 s0, 0x2100

 ; GFX10-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX10-FLATSCR-NEXT:    ;;#ASMSTART
 ; GFX10-FLATSCR-NEXT:    ; def s[40:43]
 ; GFX10-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX10-FLATSCR-NEXT:    ;;#ASMSTART
 ; GFX10-FLATSCR-NEXT:    ; def s[34:35]
 ; GFX10-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX10-FLATSCR-NEXT:    ;;#ASMSTART
-; GFX10-FLATSCR-NEXT:    ; def s[38:39]
-; GFX10-FLATSCR-NEXT:    ;;#ASMEND
-; GFX10-FLATSCR-NEXT:    ;;#ASMSTART
-; GFX10-FLATSCR-NEXT:    ; def s44
+; GFX10-FLATSCR-NEXT:    ; def s38
 ; GFX10-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX10-FLATSCR-NEXT:    v_cmpx_eq_u32_e32 0, v0
 ; GFX10-FLATSCR-NEXT:    s_cbranch_execz .LBB1_2
 ; GFX10-FLATSCR-NEXT:  ; %bb.1: ; %bb0
 ; GFX10-FLATSCR-NEXT:    ;;#ASMSTART
-; GFX10-FLATSCR-NEXT:    ; use s[0:7],s[8:15],s[16:23],s[24:31],s[40:43],s[34:35],s[38:39]
+; GFX10-FLATSCR-NEXT:    ; use s[0:7],s[8:15],s[16:23],s[24:31],s[40:43],s[34:35]
 ; GFX10-FLATSCR-NEXT:    ;;#ASMEND
 ; GFX10-FLATSCR-NEXT:    s_movk_i32 s0, 0x2010
 ; GFX10-FLATSCR-NEXT:    v_mov_b32_e32 v88, v59
 ; GFX10-FLATSCR-NEXT:    v_mov_b32_e32 v92, v63
 ; GFX10-FLATSCR-NEXT:    v_mov_b32_e32 v87, v58
 ; GFX10-FLATSCR-NEXT:    v_mov_b32_e32 v86, v57
 ; GFX10-FLATSCR-NEXT:    v_mov_b32_e32 v85, v56
 ; GFX10-FLATSCR-NEXT:    v_mov_b32_e32 v91, v62

 
 ; fill up SGPRs
   %sgpr0 = call <8 x i32> asm sideeffect "; def $0", "=s" ()
   %sgpr1 = call <8 x i32> asm sideeffect "; def $0", "=s" ()
   %sgpr2 = call <8 x i32> asm sideeffect "; def $0", "=s" ()
   %sgpr3 = call <8 x i32> asm sideeffect "; def $0", "=s" ()
   %sgpr4 = call <4 x i32> asm sideeffect "; def $0", "=s" ()
   %sgpr5 = call <2 x i32> asm sideeffect "; def $0", "=s" ()
-  %sgpr6 = call <2 x i32> asm sideeffect "; def $0", "=s" ()
-  %sgpr7 = call i32 asm sideeffect "; def $0", "=s" ()
+  %sgpr6 = call i32 asm sideeffect "; def $0", "=s" ()
 
   %cmp = icmp eq i32 %x, 0
   br i1 %cmp, label %bb0, label %ret
 
 bb0:
 ; create SGPR pressure
-  call void asm sideeffect "; use $0,$1,$2,$3,$4,$5,$6", "s,s,s,s,s,s,s,s"(<8 x i32> %sgpr0, <8 x i32> %sgpr1, <8 x i32> %sgpr2, <8 x i32> %sgpr3, <4 x i32> %sgpr4, <2 x i32> %sgpr5, <2 x i32> %sgpr6, i32 %sgpr7)
+  call void asm sideeffect "; use $0,$1,$2,$3,$4,$5", "s,s,s,s,s,s,s"(<8 x i32> %sgpr0, <8 x i32> %sgpr1, <8 x i32> %sgpr2, <8 x i32> %sgpr3, <4 x i32> %sgpr4, <2 x i32> %sgpr5, i32 %sgpr6)
 
 ; mark most VGPR registers as used to increase register pressure
   call void asm sideeffect "", "~{v4},~{v8},~{v12},~{v16},~{v20},~{v24},~{v28},~{v32}" ()
   call void asm sideeffect "", "~{v36},~{v40},~{v44},~{v48},~{v52},~{v56},~{v60},~{v64}" ()
   call void asm sideeffect "", "~{v68},~{v72},~{v76},~{v80},~{v84},~{v88},~{v92},~{v96}" ()
   call void asm sideeffect "", "~{v100},~{v104},~{v108},~{v112},~{v116},~{v120},~{v124},~{v128}" ()
   call void asm sideeffect "", "~{v132},~{v136},~{v140},~{v144},~{v148},~{v152},~{v156},~{v160}" ()
   call void asm sideeffect "", "~{v164},~{v168},~{v172},~{v176},~{v180},~{v184},~{v188},~{v192}" ()

llvm/test/CodeGen/AMDGPU/spill-sgpr-stack-no-sgpr.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX10 %s
 
-; Spill an SGPR to scratch without having spare SGPRs available to save exec
+; The test was originally written to spill an SGPR to scratch without having spare SGPRs
+; available to save exec. This scenario won't be true anymore as we reseve SGPR(s)
+; upfront for saving exec.
 
 define amdgpu_kernel void @test() #1 {
 ; GFX10-LABEL: test:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_mov_b32 s8, SCRATCH_RSRC_DWORD0
 ; GFX10-NEXT:    s_mov_b32 s9, SCRATCH_RSRC_DWORD1
 ; GFX10-NEXT:    s_mov_b32 s10, -1
 ; GFX10-NEXT:    s_mov_b32 s11, 0x31e16000
 ; GFX10-NEXT:    s_add_u32 s8, s8, s1
 ; GFX10-NEXT:    s_addc_u32 s9, s9, 0
 ; GFX10-NEXT:    ;;#ASMSTART
 ; GFX10-NEXT:    ; def s[0:7]
 ; GFX10-NEXT:    ;;#ASMEND
 ; GFX10-NEXT:    ;;#ASMSTART
 ; GFX10-NEXT:    ; def s[8:12]
 ; GFX10-NEXT:    ;;#ASMEND
-; GFX10-NEXT:    s_not_b64 exec, exec
-; GFX10-NEXT:    buffer_store_dword v0, off, s[8:11], 0
-; GFX10-NEXT:    v_writelane_b32 v0, s8, 0
-; GFX10-NEXT:    v_writelane_b32 v0, s9, 1
-; GFX10-NEXT:    v_writelane_b32 v0, s10, 2
-; GFX10-NEXT:    v_writelane_b32 v0, s11, 3
-; GFX10-NEXT:    v_writelane_b32 v0, s12, 4
-; GFX10-NEXT:    buffer_store_dword v0, off, s[8:11], 0 offset:4 ; 4-byte Folded Spill
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_not_b64 exec, exec
-; GFX10-NEXT:    buffer_store_dword v0, off, s[8:11], 0 offset:4 ; 4-byte Folded Spill
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_not_b64 exec, exec
-; GFX10-NEXT:    buffer_load_dword v0, off, s[8:11], 0
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_not_b64 exec, exec
 ; GFX10-NEXT:    ;;#ASMSTART
 ; GFX10-NEXT:    ;;#ASMEND
 ; GFX10-NEXT:    ;;#ASMSTART
 ; GFX10-NEXT:    ; use s[0:7]
 ; GFX10-NEXT:    ;;#ASMEND
-; GFX10-NEXT:    s_mov_b64 s[6:7], exec
-; GFX10-NEXT:    s_mov_b64 exec, 31
-; GFX10-NEXT:    buffer_store_dword v0, off, s[8:11], 0
-; GFX10-NEXT:    buffer_load_dword v0, off, s[8:11], 0 offset:4 ; 4-byte Folded Reload
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    v_readlane_b32 s0, v0, 0
-; GFX10-NEXT:    v_readlane_b32 s1, v0, 1
-; GFX10-NEXT:    v_readlane_b32 s2, v0, 2
-; GFX10-NEXT:    v_readlane_b32 s3, v0, 3
-; GFX10-NEXT:    v_readlane_b32 s4, v0, 4
-; GFX10-NEXT:    buffer_load_dword v0, off, s[8:11], 0
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX10-NEXT:    ;;#ASMSTART
-; GFX10-NEXT:    ; use s[0:4]
+; GFX10-NEXT:    ; use s[8:12]
 ; GFX10-NEXT:    ;;#ASMEND
 ; GFX10-NEXT:    s_endpgm
   %wide.sgpr0 = call <8 x i32> asm sideeffect "; def $0", "={s[0:7]}" () #0
   %wide.sgpr2 = call <5 x i32> asm sideeffect "; def $0", "={s[8:12]}" () #0
   call void asm sideeffect "", "~{v[0:7]}" () #0
   call void asm sideeffect "; use $0", "s"(<8 x i32> %wide.sgpr0) #0
   call void asm sideeffect "; use $0", "s"(<5 x i32> %wide.sgpr2) #0
   ret void
 }
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind "amdgpu-num-sgpr"="16" "amdgpu-num-vgpr"="8" }

llvm/test/CodeGen/AMDGPU/spill-writelane-vgprs.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
 
 ; Callee must preserve the VGPR modified by writelane even if it is marked Caller-saved.
 
 declare i32 @llvm.amdgcn.writelane(i32, i32, i32)
 
 define void @sgpr_spill_writelane() {
 ; GCN-LABEL: sgpr_spill_writelane:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; GCN-NEXT:    s_xor_saveexec_b64 s[6:7], -1
 ; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
-; GCN-NEXT:    s_mov_b64 exec, s[4:5]
+; GCN-NEXT:    s_mov_b64 exec, s[6:7]
 ; GCN-NEXT:    v_writelane_b32 v0, s35, 0
 ; GCN-NEXT:    ;;#ASMSTART
 ; GCN-NEXT:    ;;#ASMEND
 ; GCN-NEXT:    v_readlane_b32 s35, v0, 0
-; GCN-NEXT:    s_xor_saveexec_b64 s[4:5], -1
+; GCN-NEXT:    s_xor_saveexec_b64 s[6:7], -1
 ; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
-; GCN-NEXT:    s_mov_b64 exec, s[4:5]
+; GCN-NEXT:    s_mov_b64 exec, s[6:7]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "", "~{s35}"()
   ret void
 }
 
 ; FIXME: The writelane intrinsic doesn't really overwrite any inactive lanes
 ; and hence there is no need to preserve the VGPR it modifies.

llvm/test/CodeGen/MIR/AMDGPU/machine-function-info-after-pei.ll

 ; AFTER-PEI-NEXT:   fp32-input-denormals: true
 ; AFTER-PEI-NEXT:   fp32-output-denormals: true
 ; AFTER-PEI-NEXT:   fp64-fp16-input-denormals: true
 ; AFTER-PEI-NEXT:   fp64-fp16-output-denormals: true
 ; AFTER-PEI-NEXT: highBitsOf32BitAddress: 0
 ; AFTER-PEI-NEXT: occupancy: 5
 ; AFTER-PEI-NEXT: scavengeFI: '%fixed-stack.0'
 ; AFTER-PEI-NEXT: vgprForAGPRCopy: ''
+; AFTER-PEI-NEXT: sgprForEXECCopy: ''
 ; AFTER-PEI-NEXT: body:
 define amdgpu_kernel void @scavenge_fi(ptr addrspace(1) %out, i32 %in) #0 {
   %wide.sgpr0 = call <32 x i32>  asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr1 = call <32 x i32>  asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr2 = call <32 x i32>  asm sideeffect "; def $0", "=s" () #0
   %wide.sgpr3 = call <32 x i32>  asm sideeffect "; def $0", "=s" () #0
 
   call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr0) #0
   call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr1) #0
   call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr2) #0
   call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr3) #0
   ret void
 }
 
 attributes #0 = { nounwind }

llvm/test/CodeGen/MIR/AMDGPU/machine-function-info-no-ir.mir

 # FULL-NEXT: dx10-clamp: true
 # FULL-NEXT: fp32-input-denormals: true
 # FULL-NEXT: fp32-output-denormals: true
 # FULL-NEXT: fp64-fp16-input-denormals: true
 # FULL-NEXT: fp64-fp16-output-denormals: true
 # FULL-NEXT:  highBitsOf32BitAddress: 0
 # FULL-NEXT:  occupancy: 8
 # FULL-NEXT:  vgprForAGPRCopy: ''
+# FULL-NEXT:  sgprForEXECCopy: ''
 # FULL-NEXT: body:
 
 # SIMPLE: machineFunctionInfo:
 # SIMPLE-NEXT: explicitKernArgSize: 128
 # SIMPLE-NEXT: maxKernArgAlign: 64
 # SIMPLE-NEXT: ldsSize: 2048
 # SIMPLE-NEXT: gdsSize: 256
 # SIMPLE-NEXT: isEntryFunction: true

 # FULL-NEXT: dx10-clamp: true
 # FULL-NEXT: fp32-input-denormals: true
 # FULL-NEXT: fp32-output-denormals: true
 # FULL-NEXT: fp64-fp16-input-denormals: true
 # FULL-NEXT: fp64-fp16-output-denormals: true
 # FULL-NEXT:  highBitsOf32BitAddress: 0
 # FULL-NEXT:  occupancy: 8
 # FULL-NEXT: vgprForAGPRCopy: ''
+# FULL-NEXT: sgprForEXECCopy: ''
 # FULL-NEXT: body:
 
 # SIMPLE: machineFunctionInfo:
 # SIMPLE-NEXT: maxKernArgAlign: 1
 # SIMPLE-NEXT: argumentInfo:
 # SIMPLE-NEXT: privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
 # SIMPLE-NEXT: dispatchPtr:     { reg: '$sgpr4_sgpr5' }
 # SIMPLE-NEXT: queuePtr:        { reg: '$sgpr6_sgpr7' }

 # FULL-NEXT: dx10-clamp: true
 # FULL-NEXT: fp32-input-denormals: true
 # FULL-NEXT: fp32-output-denormals: true
 # FULL-NEXT: fp64-fp16-input-denormals: true
 # FULL-NEXT: fp64-fp16-output-denormals: true
 # FULL-NEXT:  highBitsOf32BitAddress: 0
 # FULL-NEXT:  occupancy: 8
 # FULL-NEXT: vgprForAGPRCopy: ''
+# FULL-NEXT: sgprForEXECCopy: ''
 # FULL-NEXT: body:
 
 # SIMPLE: machineFunctionInfo:
 # SIMPLE-NEXT: maxKernArgAlign: 1
 # SIMPLE-NEXT: argumentInfo:
 # SIMPLE-NEXT: privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
 # SIMPLE-NEXT: dispatchPtr:     { reg: '$sgpr4_sgpr5' }
 # SIMPLE-NEXT: queuePtr:        { reg: '$sgpr6_sgpr7' }

 # FULL-NEXT: dx10-clamp: true
 # FULL-NEXT: fp32-input-denormals: true
 # FULL-NEXT: fp32-output-denormals: true
 # FULL-NEXT: fp64-fp16-input-denormals: true
 # FULL-NEXT: fp64-fp16-output-denormals: true
 # FULL-NEXT:  highBitsOf32BitAddress: 0
 # FULL-NEXT:  occupancy: 8
 # FULL-NEXT: vgprForAGPRCopy: ''
+# FULL-NEXT: sgprForEXECCopy: ''
 # FULL-NEXT: body:
 
 # SIMPLE: machineFunctionInfo:
 # SIMPLE-NEXT: maxKernArgAlign: 1
 # SIMPLE-NEXT: isEntryFunction: true
 # SIMPLE-NEXT: argumentInfo:
 # SIMPLE-NEXT: privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
 # SIMPLE-NEXT: dispatchPtr:     { reg: '$sgpr4_sgpr5' }

 name: vgpr_for_agpr_copy_noreg
 machineFunctionInfo:
   vgprForAGPRCopy: '$noreg'
 body:             |
   bb.0:
     SI_RETURN
 
 ...
+
+---
+# ALL-LABEL: name: sgpr_for_exec_copy
+# ALL: sgprForEXECCopy: '$sgpr2_sgpr3'
+name: sgpr_for_exec_copy
+machineFunctionInfo:
+  sgprForEXECCopy: '$sgpr2_sgpr3'
+body:             |
+  bb.0:
+    SI_RETURN
+
+...
+
+---
+# ALL-LABEL: name: sgpr_for_exec_copy_noreg
+# FULL: sgprForEXECCopy: ''
+# SIMPLE-NOT: sgprForEXECCopy
+name: sgpr_for_exec_copy_noreg
+machineFunctionInfo:
+  sgprForEXECCopy: '$noreg'
+body:             |
+  bb.0:
+    SI_RETURN
+
+...

llvm/test/CodeGen/MIR/AMDGPU/machine-function-info.ll

 ; CHECK-NEXT: dx10-clamp: true
 ; CHECK-NEXT: fp32-input-denormals: true
 ; CHECK-NEXT: fp32-output-denormals: true
 ; CHECK-NEXT: fp64-fp16-input-denormals: true
 ; CHECK-NEXT: fp64-fp16-output-denormals: true
 ; CHECK-NEXT: highBitsOf32BitAddress: 0
 ; CHECK-NEXT: occupancy: 8
 ; CHECK-NEXT: vgprForAGPRCopy: ''
+; CHECK-NEXT: sgprForEXECCopy: '$sgpr100_sgpr101'
 ; CHECK-NEXT: body:
 define amdgpu_kernel void @kernel(i32 %arg0, i64 %arg1, <16 x i32> %arg2) {
   %gep = getelementptr inbounds [512 x float], ptr addrspace(3) @lds, i32 0, i32 %arg0
   store float 0.0, ptr addrspace(3) %gep, align 4
   ret void
 }
 
 @gds = addrspace(2) global [128 x i32] undef, align 4

 ; CHECK-NEXT: dx10-clamp: true
 ; CHECK-NEXT: fp32-input-denormals: true
 ; CHECK-NEXT: fp32-output-denormals: true
 ; CHECK-NEXT: fp64-fp16-input-denormals: true
 ; CHECK-NEXT: fp64-fp16-output-denormals: true
 ; CHECK-NEXT: highBitsOf32BitAddress: 0
 ; CHECK-NEXT: occupancy: 10
 ; CHECK-NEXT: vgprForAGPRCopy: ''
+; CHECK-NEXT: sgprForEXECCopy: '$sgpr100_sgpr101'
 ; CHECK-NEXT: body:
 define amdgpu_ps void @ps_shader(i32 %arg0, i32 inreg %arg1) {
   %gep = getelementptr inbounds [128 x i32], ptr addrspace(2) @gds, i32 0, i32 %arg0
   atomicrmw add ptr addrspace(2) %gep, i32 8 seq_cst
   ret void
 }
 
 ; CHECK-LABEL: {{^}}name: ps_shader_ps_input_enable

 ; CHECK-NEXT: dx10-clamp: true
 ; CHECK-NEXT: fp32-input-denormals: true
 ; CHECK-NEXT: fp32-output-denormals: true
 ; CHECK-NEXT: fp64-fp16-input-denormals: true
 ; CHECK-NEXT: fp64-fp16-output-denormals: true
 ; CHECK-NEXT: highBitsOf32BitAddress: 0
 ; CHECK-NEXT: occupancy: 8
 ; CHECK-NEXT: vgprForAGPRCopy: ''
+; CHECK-NEXT: sgprForEXECCopy: '$sgpr100_sgpr101'
 ; CHECK-NEXT: body:
 define void @function() {
   ret void
 }
 
 ; CHECK-LABEL: {{^}}name: function_nsz
 ; CHECK: machineFunctionInfo:
 ; CHECK-NEXT: explicitKernArgSize: 0

 ; CHECK-NEXT: dx10-clamp: true
 ; CHECK-NEXT: fp32-input-denormals: true
 ; CHECK-NEXT: fp32-output-denormals: true
 ; CHECK-NEXT: fp64-fp16-input-denormals: true
 ; CHECK-NEXT: fp64-fp16-output-denormals: true
 ; CHECK-NEXT: highBitsOf32BitAddress: 0
 ; CHECK-NEXT: occupancy: 8
 ; CHECK-NEXT: vgprForAGPRCopy: ''
+; CHECK-NEXT: sgprForEXECCopy: '$sgpr100_sgpr101'
 ; CHECK-NEXT: body:
 define void @function_nsz() #0 {
   ret void
 }
 
 ; CHECK-LABEL: {{^}}name: function_dx10_clamp_off
 ; CHECK: mode:
 ; CHECK-NEXT: ieee: true

llvm/test/CodeGen/MIR/AMDGPU/sgpr-for-exec-copy-invalid-reg.mir

+# RUN: not llc -mtriple=amdgcn-amd-amdhsa -run-pass=none %s -o /dev/null 2>&1 | FileCheck -check-prefix=ERR %s
+
+---
+name: invalid_reg
+machineFunctionInfo:
+# ERR: [[@LINE+1]]:21: unknown register name 'srst'
+  sgprForEXECCopy: '$srst'
+body:             |
+  bb.0:
+    S_ENDPGM 0
+
+...