AMDGPU/GlobalISel: Select llvm.amdgcn.raw.buffer.store

[nhaehnle]

Mostly seems reasonable to me, but two questions inline.

lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

745–755

Could we rearrange these to have a signature analogous to parseCachePolicy, to sanity-check that no bits in CachePolicy are set unexpectedly?

lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

1600–1611

What's the intention here?

[arsenm]

lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

745–755

I think that's a lot uglier. That kind of check belongs in the verifier

lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

1600–1611

This is leftover junk from when I thought it was a good idea to select this here

[arsenm]

Remove leftover junk

[arsenm]

Rebase

[arsenm]

ping

[nhaehnle]

LGTM

[arsenm]

Slightly confused, but I think this was ca2672accec386bd80804db659b5f06674ebd5be?

[nhaehnle]

In D65238#1842829, @arsenm wrote:

Slightly confused, but I think this was ca2672accec386bd80804db659b5f06674ebd5be?

I don't have that hash. I do have 67f1f6ff8c07b5eef7239679a6b534efe933ceaa though, which is quite old indeed. Either way, it's fine.

lib/Target/AMDGPU/AMDGPUInstructionSelector.h

 }
 
 namespace llvm {
 
 class AMDGPUInstrInfo;
 class AMDGPURegisterBankInfo;
 class GCNSubtarget;
 class MachineInstr;
+class MachineIRBuilder;
 class MachineOperand;
 class MachineRegisterInfo;
 class SIInstrInfo;
 class SIMachineFunctionInfo;
 class SIRegisterInfo;
 
 class AMDGPUInstructionSelector : public InstructionSelector {
 public:
private:
   bool selectG_ADD_SUB(MachineInstr &I) const;
   bool selectG_EXTRACT(MachineInstr &I) const;
   bool selectG_MERGE_VALUES(MachineInstr &I) const;
   bool selectG_UNMERGE_VALUES(MachineInstr &I) const;
   bool selectG_GEP(MachineInstr &I) const;
   bool selectG_IMPLICIT_DEF(MachineInstr &I) const;
   bool selectG_INSERT(MachineInstr &I) const;
   bool selectG_INTRINSIC(MachineInstr &I, CodeGenCoverage &CoverageInfo) const;
+
+  std::tuple<Register, unsigned, unsigned>
+  splitBufferOffsets(MachineIRBuilder &B, Register OrigOffset) const;
+
+  bool selectStoreIntrinsic(MachineInstr &MI, bool IsFormat) const;
+
   bool selectG_INTRINSIC_W_SIDE_EFFECTS(MachineInstr &I,
                                         CodeGenCoverage &CoverageInfo) const;
   int getS_CMPOpcode(CmpInst::Predicate P, unsigned Size) const;
   bool selectG_ICMP(MachineInstr &I) const;
   bool hasVgprParts(ArrayRef<GEPInfo> AddrInfo) const;
   void getAddrModeInfo(const MachineInstr &Load, const MachineRegisterInfo &MRI,
                        SmallVectorImpl<GEPInfo> &AddrInfo) const;
   bool selectSMRD(MachineInstr &I, ArrayRef<GEPInfo> AddrInfo) const;

lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

 #include "AMDGPURegisterBankInfo.h"
 #include "AMDGPURegisterInfo.h"
 #include "AMDGPUSubtarget.h"
 #include "AMDGPUTargetMachine.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/GlobalISel/InstructionSelector.h"
 #include "llvm/CodeGen/GlobalISel/InstructionSelectorImpl.h"
+#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/GlobalISel/Utils.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/IR/Type.h"
return BuildMI(BB, Insert, DL, TII.get(Opcode))
           .addReg(Reg1)
           .addReg(Reg2)
           .addReg(Reg3)
           .addImm(VM)
           .addImm(Compr)
           .addImm(Enabled);
 }
 
+static bool isZero(Register Reg, MachineRegisterInfo &MRI) {
+  int64_t C;
+  if (mi_match(Reg, MRI, m_ICst(C)) && C == 0)
+    return true;
+
+  // FIXME: matcher should ignore copies
+  return mi_match(Reg, MRI, m_Copy(m_ICst(C))) && C == 0;
+}
+
+static unsigned extractGLC(unsigned CachePolicy) {
+  return CachePolicy & 1;
+}
+
+static unsigned extractSLC(unsigned CachePolicy) {
+  return (CachePolicy >> 1) & 1;
+}
+
+static unsigned extractDLC(unsigned CachePolicy) {
+  return (CachePolicy >> 2) & 1;
+}
+
+// Returns Base register, constant offset, and offset def point.
+static std::tuple<Register, unsigned, MachineInstr *>
+getBaseWithConstantOffset(MachineRegisterInfo &MRI, Register Reg) {
+  MachineInstr *Def = getDefIgnoringCopies(Reg, MRI);
+  if (!Def)
+    return {Reg, 0, nullptr};
+
+  if (Def->getOpcode() == AMDGPU::G_CONSTANT) {
+    unsigned Offset;
+    const MachineOperand &Op = Def->getOperand(1);
+    if (Op.isImm())
+      Offset = Op.getImm();
+    else
+      Offset = Op.getCImm()->getZExtValue();
+
+    return {Register(), Offset, Def};
+  }
+
+  int64_t Offset;
+  if (Def->getOpcode() == AMDGPU::G_ADD) {
+    // TODO: Handle G_OR used for add case
+    if (mi_match(Def->getOperand(1).getReg(), MRI, m_ICst(Offset)))
+      return {Def->getOperand(0).getReg(), Offset, Def};
+
+    // FIXME: matcher should ignore copies
+    if (mi_match(Def->getOperand(1).getReg(), MRI, m_Copy(m_ICst(Offset))))
+      return {Def->getOperand(0).getReg(), Offset, Def};
+  }
+
+  return {Reg, 0, Def};
+}
+
+// TODO: Move this to combiner
+// Returns base register, imm offset, total constant offset.
+std::tuple<Register, unsigned, unsigned>
+AMDGPUInstructionSelector::splitBufferOffsets(MachineIRBuilder &B,
+                                              Register OrigOffset) const {
+  const unsigned MaxImm = 4095;
+  Register BaseReg;
+  unsigned TotalConstOffset;
+  MachineInstr *OffsetDef;
+  MachineRegisterInfo &MRI = *B.getMRI();
+
+  std::tie(BaseReg, TotalConstOffset, OffsetDef)
+    = getBaseWithConstantOffset(MRI, OrigOffset);
+
+  unsigned ImmOffset = TotalConstOffset;
+
+  // If the immediate value is too big for the immoffset field, put the value
+  // and -4096 into the immoffset field so that the value that is copied/added
+  // for the voffset field is a multiple of 4096, and it stands more chance
+  // of being CSEd with the copy/add for another similar load/store.f
+  // However, do not do that rounding down to a multiple of 4096 if that is a
+  // negative number, as it appears to be illegal to have a negative offset
+  // in the vgpr, even if adding the immediate offset makes it positive.
+  unsigned Overflow = ImmOffset & ~MaxImm;
+  ImmOffset -= Overflow;
+  if ((int32_t)Overflow < 0) {
+    Overflow += ImmOffset;
+    ImmOffset = 0;
+  }
+
+  if (Overflow != 0) {
+    // In case this is in a waterfall loop, insert offset code at the def point
+    // of the offset, not inside the loop.
+    MachineBasicBlock::iterator OldInsPt = B.getInsertPt();
+    MachineBasicBlock &OldMBB = B.getMBB();
+    B.setInstr(*OffsetDef);
+
+    if (!BaseReg) {
+      BaseReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+      B.buildInstr(AMDGPU::V_MOV_B32_e32)
+        .addDef(BaseReg)
+        .addImm(Overflow);
+    } else {
+      Register OverflowVal = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+      B.buildInstr(AMDGPU::V_MOV_B32_e32)
+        .addDef(OverflowVal)
+        .addImm(Overflow);
+
+      Register NewBaseReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+      TII.getAddNoCarry(B.getMBB(), B.getInsertPt(), B.getDebugLoc(), NewBaseReg)
+        .addReg(BaseReg)
+        .addReg(OverflowVal, RegState::Kill)
+        .addImm(0);
+      BaseReg = NewBaseReg;
+    }
+
+    B.setInsertPt(OldMBB, OldInsPt);
+  }
+
+  return {BaseReg, ImmOffset, TotalConstOffset};
+}
+
+bool AMDGPUInstructionSelector::selectStoreIntrinsic(MachineInstr &MI,
+                                                     bool IsFormat) const {
+  MachineIRBuilder B(MI);
+  MachineRegisterInfo &MRI = *B.getMRI();
+  MachineFunction &MF = B.getMF();
+  Register VData = MI.getOperand(1).getReg();
+  LLT Ty = MRI.getType(VData);
+
+  int Size = Ty.getSizeInBits();
+  if (Size % 32 != 0)
+    return false;
+
+  // FIXME: Verifier should enforce 1 MMO for these intrinsics.
+  MachineMemOperand *MMO = *MI.memoperands_begin();
+  const int MemSize = MMO->getSize();
+
+  Register RSrc = MI.getOperand(2).getReg();
+  Register VOffset = MI.getOperand(3).getReg();
+  Register SOffset = MI.getOperand(4).getReg();
+  unsigned CachePolicy = MI.getOperand(5).getImm();
+  unsigned ImmOffset;
+  unsigned TotalOffset;
+
+  std::tie(VOffset, ImmOffset, TotalOffset) = splitBufferOffsets(B, VOffset);
+  if (TotalOffset != 0)
+    MMO = MF.getMachineMemOperand(MMO, TotalOffset, MemSize);
+
+  const bool Offen = !isZero(VOffset, MRI);
+
+  unsigned Opc = AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact;
+  switch (8 * MemSize) {
+  case 8:
+    Opc = Offen ? AMDGPU::BUFFER_STORE_BYTE_OFFEN_exact :
+                  AMDGPU::BUFFER_STORE_BYTE_OFFSET_exact;
+    break;
+  case 16:
+    Opc = Offen ? AMDGPU::BUFFER_STORE_SHORT_OFFEN_exact :
+                  AMDGPU::BUFFER_STORE_SHORT_OFFSET_exact;
+    break;
+  default:
+    Opc = Offen ? AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact :
+                  AMDGPU::BUFFER_STORE_DWORD_OFFSET_exact;
+    if (Size > 32)
+      Opc = AMDGPU::getMUBUFOpcode(Opc, Size / 32);
+    break;
+  }
+
+  MachineInstrBuilder MIB = B.buildInstr(Opc)
+    .addUse(VData);
+
+  if (Offen)
+    MIB.addUse(VOffset);
+
+  MIB.addUse(RSrc)
+     .addUse(SOffset)
+     .addImm(ImmOffset)
+     .addImm(extractGLC(CachePolicy))
+     .addImm(extractSLC(CachePolicy))
+     .addImm(0) // tfe: FIXME: Remove from inst
+     .addImm(extractDLC(CachePolicy))
+     .addMemOperand(MMO);
+
+  MI.eraseFromParent();
+
+  return constrainSelectedInstRegOperands(*MIB, TII, TRI, RBI);
+}
+
 bool AMDGPUInstructionSelector::selectG_INTRINSIC_W_SIDE_EFFECTS(
   MachineInstr &I, CodeGenCoverage &CoverageInfo) const {
   MachineBasicBlock *BB = I.getParent();
   MachineFunction *MF = BB->getParent();
   MachineRegisterInfo &MRI = MF->getRegInfo();
 
   auto IntrinsicID = I.getOperand(I.getNumExplicitDefs()).getIntrinsicID();
   switch (IntrinsicID) {
case Intrinsic::amdgcn_end_cf: {
 
     Register Reg = I.getOperand(1).getReg();
     I.eraseFromParent();
 
     if (!MRI.getRegClassOrNull(Reg))
       MRI.setRegClass(Reg, TRI.getWaveMaskRegClass());
     return true;
   }
+  case Intrinsic::amdgcn_raw_buffer_store:
+    return selectStoreIntrinsic(I, false);
   default:
     return selectImpl(I, CoverageInfo);
   }
 }
 
 bool AMDGPUInstructionSelector::selectG_SELECT(MachineInstr &I) const {
   MachineBasicBlock *BB = I.getParent();
   MachineFunction *MF = BB->getParent();

lib/Target/AMDGPU/AMDGPULegalizerInfo.h

bool legalizePreloadedArgIntrin(
     MachineInstr &MI, MachineRegisterInfo &MRI, MachineIRBuilder &B,
     AMDGPUFunctionArgInfo::PreloadedValue ArgType) const;
 
   bool legalizeFDIVFast(MachineInstr &MI, MachineRegisterInfo &MRI,
                         MachineIRBuilder &B) const;
 
   bool legalizeImplicitArgPtr(MachineInstr &MI, MachineRegisterInfo &MRI,
                               MachineIRBuilder &B) const;
+
+  bool legalizeRawBufferStore(MachineInstr &MI, MachineRegisterInfo &MRI,
+                              MachineIRBuilder &B, bool IsFormat) const;
   bool legalizeIntrinsic(MachineInstr &MI, MachineRegisterInfo &MRI,
                          MachineIRBuilder &MIRBuilder) const override;
 
 };
 } // End llvm namespace.
 #endif

lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

bool AMDGPULegalizerInfo::legalizeImplicitArgPtr(MachineInstr &MI,
   if (!loadInputValue(KernargPtrReg, B, Arg))
     return false;
 
   B.buildGEP(DstReg, KernargPtrReg, B.buildConstant(IdxTy, Offset).getReg(0));
   MI.eraseFromParent();
   return true;
 }
 
+bool AMDGPULegalizerInfo::legalizeRawBufferStore(MachineInstr &MI,
+                                                 MachineRegisterInfo &MRI,
+                                                 MachineIRBuilder &B,
+                                                 bool IsFormat) const {
+  assert(!IsFormat && "format buffer stores not implemented");
+
+  // TODO: Reject f16 format on targets where unsupported.
+  Register VData = MI.getOperand(1).getReg();
+  LLT Ty = MRI.getType(VData);
+
+  B.setInstr(MI);
+
+  const LLT S32 = LLT::scalar(32);
+  const LLT S16 = LLT::scalar(16);
+
+  // Fixup illegal register types for i8 stores.
+  if (Ty == LLT::scalar(8) || Ty == S16) {
+    Register AnyExt = B.buildAnyExt(LLT::scalar(32), VData).getReg(0);
+    MI.getOperand(1).setReg(AnyExt);
+    return true;
+  }
+
+  if (Ty.isVector()) {
+    if (Ty.getElementType() == S16 && Ty.getNumElements() <= 4)
+      return true;
+
+    return Ty.getElementType() == S32 && Ty.getNumElements() <= 4;
+  }
+
+  return Ty == S32;
+}
+
 bool AMDGPULegalizerInfo::legalizeIntrinsic(MachineInstr &MI,
                                             MachineRegisterInfo &MRI,
                                             MachineIRBuilder &B) const {
   // Replace the use G_BRCOND with the exec manipulate and branch pseudos.
   switch (MI.getOperand(MI.getNumExplicitDefs()).getIntrinsicID()) {
   case Intrinsic::amdgcn_if: {
     if (MachineInstr *BrCond = verifyCFIntrinsic(MI, MRI)) {
       const SIRegisterInfo *TRI
bool AMDGPULegalizerInfo::legalizeIntrinsic(MachineInstr &MI,
   case Intrinsic::amdgcn_implicit_buffer_ptr:
     return legalizePreloadedArgIntrin(
       MI, MRI, B, AMDGPUFunctionArgInfo::IMPLICIT_BUFFER_PTR);
   case Intrinsic::amdgcn_dispatch_id:
     return legalizePreloadedArgIntrin(MI, MRI, B,
                                       AMDGPUFunctionArgInfo::DISPATCH_ID);
   case Intrinsic::amdgcn_fdiv_fast:
     return legalizeFDIVFast(MI, MRI, B);
+  case Intrinsic::amdgcn_raw_buffer_store:
+    return legalizeRawBufferStore(MI, MRI, B, false);
   default:
     return true;
   }
 
   return true;
 }

lib/Target/AMDGPU/AMDGPURegisterBankInfo.h

 
 #define GET_REGBANK_DECLARATIONS
 #include "AMDGPUGenRegisterBank.inc"
 #undef GET_REGBANK_DECLARATIONS
 
 namespace llvm {
 
 class LLT;
+class GCNSubtarget;
 class MachineIRBuilder;
+class SIInstrInfo;
 class SIRegisterInfo;
 class TargetRegisterInfo;
 
 /// This class provides the information for the target register banks.
 class AMDGPUGenRegisterBankInfo : public RegisterBankInfo {
 
 protected:
 
 #define GET_TARGET_REGBANK_CLASS
 #include "AMDGPUGenRegisterBank.inc"
 };
 class AMDGPURegisterBankInfo : public AMDGPUGenRegisterBankInfo {
+  const GCNSubtarget &Subtarget;
   const SIRegisterInfo *TRI;
+  const SIInstrInfo *TII;
 
-  void executeInWaterfallLoop(MachineInstr &MI,
+  bool executeInWaterfallLoop(MachineIRBuilder &B,
+                              MachineInstr &MI,
+                              MachineRegisterInfo &MRI,
+                              ArrayRef<unsigned> OpIndices) const;
+  bool executeInWaterfallLoop(MachineInstr &MI,
                               MachineRegisterInfo &MRI,
                               ArrayRef<unsigned> OpIndices) const;
 
   void constrainOpWithReadfirstlane(MachineInstr &MI, MachineRegisterInfo &MRI,
                                     unsigned OpIdx) const;
   bool applyMappingWideLoad(MachineInstr &MI,
                             const AMDGPURegisterBankInfo::OperandsMapper &OpdMapper,
                             MachineRegisterInfo &MRI) const;
   bool
   applyMappingImage(MachineInstr &MI,
                     const AMDGPURegisterBankInfo::OperandsMapper &OpdMapper,
                     MachineRegisterInfo &MRI, int RSrcIdx) const;
 
+  Register handleD16VData(MachineIRBuilder &B, MachineRegisterInfo &MRI,
+                          Register Reg) const;
+
+  std::pair<Register, unsigned>
+  splitBufferOffsets(MachineIRBuilder &B, Register Offset) const;
+
+  MachineInstr *selectStoreIntrinsic(MachineIRBuilder &B,
+                                     MachineInstr &MI) const;
+
   /// See RegisterBankInfo::applyMapping.
   void applyMappingImpl(const OperandsMapper &OpdMapper) const override;
 
   const RegisterBankInfo::InstructionMapping &
   getInstrMappingForLoad(const MachineInstr &MI) const;
 
   unsigned getRegBankID(Register Reg, const MachineRegisterInfo &MRI,
                         const TargetRegisterInfo &TRI,
class AMDGPURegisterBankInfo : public AMDGPUGenRegisterBankInfo {
   const InstructionMapping &getDefaultMappingAllVGPR(
     const MachineInstr &MI) const;
 
   const InstructionMapping &getImageMapping(const MachineRegisterInfo &MRI,
                                             const MachineInstr &MI,
                                             int RsrcIdx) const;
 
 public:
-  AMDGPURegisterBankInfo(const TargetRegisterInfo &TRI);
+  AMDGPURegisterBankInfo(const GCNSubtarget &STI);
 
   unsigned copyCost(const RegisterBank &A, const RegisterBank &B,
                     unsigned Size) const override;
 
   unsigned getBreakDownCost(const ValueMapping &ValMapping,
                             const RegisterBank *CurBank = nullptr) const override;
 
   const RegisterBank &

lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

 #include "AMDGPUInstrInfo.h"
 #include "AMDGPUSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIRegisterInfo.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/CodeGen/GlobalISel/LegalizerHelper.h"
 #include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
+#include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/GlobalISel/RegisterBank.h"
 #include "llvm/CodeGen/GlobalISel/RegisterBankInfo.h"
 #include "llvm/CodeGen/TargetRegisterInfo.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/IR/Constants.h"
 
 #define GET_TARGET_REGBANK_IMPL
 #include "AMDGPUGenRegisterBank.inc"
 
 // This file will be TableGen'ed at some point.
 #include "AMDGPUGenRegisterBankInfo.def"
 
 using namespace llvm;
+using namespace MIPatternMatch;
 
 namespace {
 
 // Observer to apply a register bank to new registers created by LegalizerHelper.
 class ApplyRegBankMapping final : public GISelChangeObserver {
 private:
   MachineRegisterInfo &MRI;
   const RegisterBank *NewBank;
void createdInstr(MachineInstr &MI) override {
     NewInsts.push_back(&MI);
   }
 
   void changingInstr(MachineInstr &MI) override {}
   void changedInstr(MachineInstr &MI) override {}
 };
 
 }
-AMDGPURegisterBankInfo::AMDGPURegisterBankInfo(const TargetRegisterInfo &TRI)
+AMDGPURegisterBankInfo::AMDGPURegisterBankInfo(const GCNSubtarget &ST)
     : AMDGPUGenRegisterBankInfo(),
-      TRI(static_cast<const SIRegisterInfo*>(&TRI)) {
+      Subtarget(ST),
+      TRI(Subtarget.getRegisterInfo()),
+      TII(Subtarget.getInstrInfo()) {
 
   // HACK: Until this is fully tablegen'd.
   static bool AlreadyInit = false;
   if (AlreadyInit)
     return;
 
   AlreadyInit = true;
 

 ///   Enable Lane, Disable all other lanes
 ///   SGPR = read SGPR value for current lane from VGPR
 ///   VGPRResult[Lane] = use_op SGPR
 /// }
 /// Restore Execution Mask
 ///
 /// There is additional complexity to try for compare values to identify the
 /// unique values used.
-void AMDGPURegisterBankInfo::executeInWaterfallLoop(
-  MachineInstr &MI, MachineRegisterInfo &MRI,
+bool AMDGPURegisterBankInfo::executeInWaterfallLoop(
+  MachineIRBuilder &B,
+  MachineInstr &MI,
+  MachineRegisterInfo &MRI,
   ArrayRef<unsigned> OpIndices) const {
   MachineFunction *MF = MI.getParent()->getParent();
   const GCNSubtarget &ST = MF->getSubtarget<GCNSubtarget>();
   const SIInstrInfo *TII = ST.getInstrInfo();
   MachineBasicBlock::iterator I(MI);
 
   MachineBasicBlock &MBB = *MI.getParent();
   const DebugLoc &DL = MI.getDebugLoc();
 
   // Use a set to avoid extra readfirstlanes in the case where multiple operands
   // are the same register.
   SmallSet<Register, 4> SGPROperandRegs;
   for (unsigned Op : OpIndices) {
     assert(MI.getOperand(Op).isUse());
     Register Reg = MI.getOperand(Op).getReg();
     const RegisterBank *OpBank = getRegBank(Reg, MRI, *TRI);
     if (OpBank->getID() == AMDGPU::VGPRRegBankID)
       SGPROperandRegs.insert(Reg);
   }
 
   // No operands need to be replaced, so no need to loop.
   if (SGPROperandRegs.empty())
-    return;
+    return false;
 
-  MachineIRBuilder B(MI);
   SmallVector<Register, 4> ResultRegs;
   SmallVector<Register, 4> InitResultRegs;
   SmallVector<Register, 4> PhiRegs;
   for (MachineOperand &Def : MI.defs()) {
     LLT ResTy = MRI.getType(Def.getReg());
     const RegisterBank *DefBank = getRegBank(Def.getReg(), MRI, *TRI);
     ResultRegs.push_back(Def.getReg());
     Register InitReg = B.buildUndef(ResTy).getReg(0);
bool AMDGPURegisterBankInfo::executeInWaterfallLoop(
   BuildMI(MBB, MBB.end(), DL, TII->get(AMDGPU::S_MOV_B64_term), SaveExecReg)
     .addReg(AMDGPU::EXEC);
 
   // Restore the EXEC mask after the loop.
   B.setMBB(*RestoreExecBB);
   B.buildInstr(AMDGPU::S_MOV_B64_term)
     .addDef(AMDGPU::EXEC)
     .addReg(SaveExecReg);
+
+  // Restore the insert point before the original instruction.
+  B.setInsertPt(MBB, MBB.end());
+
+  return true;
+}
+
+bool AMDGPURegisterBankInfo::executeInWaterfallLoop(
+  MachineInstr &MI, MachineRegisterInfo &MRI,
+  ArrayRef<unsigned> OpIndices) const {
+  MachineIRBuilder B(MI);
+  return executeInWaterfallLoop(B, MI, MRI, OpIndices);
 }
 
 // Legalize an operand that must be an SGPR by inserting a readfirstlane.
 void AMDGPURegisterBankInfo::constrainOpWithReadfirstlane(
     MachineInstr &MI, MachineRegisterInfo &MRI, unsigned OpIdx) const {
   Register Reg = MI.getOperand(OpIdx).getReg();
   const RegisterBank *Bank = getRegBank(Reg, MRI, *TRI);
   if (Bank != &AMDGPU::VGPRRegBank)
static void substituteSimpleCopyRegs(
   const AMDGPURegisterBankInfo::OperandsMapper &OpdMapper, unsigned OpIdx) {
   SmallVector<unsigned, 1> SrcReg(OpdMapper.getVRegs(OpIdx));
   if (!SrcReg.empty()) {
     assert(SrcReg.size() == 1);
     OpdMapper.getMI().getOperand(OpIdx).setReg(SrcReg[0]);
   }
 }
 
+/// Handle register layout difference for f16 images for some subtargets.
+Register AMDGPURegisterBankInfo::handleD16VData(MachineIRBuilder &B,
+                                                MachineRegisterInfo &MRI,
+                                                Register Reg) const {
+  if (!Subtarget.hasUnpackedD16VMem())
+    return Reg;
+
+  const LLT S16 = LLT::scalar(16);
+  LLT StoreVT = MRI.getType(Reg);
+  if (!StoreVT.isVector() || StoreVT.getElementType() != S16)
+    return Reg;
+
+  auto Unmerge = B.buildUnmerge(S16, Reg);
+
+
+  SmallVector<Register, 4> WideRegs;
+  for (int I = 0, E = Unmerge->getNumOperands() - 1; I != E; ++I)
+    WideRegs.push_back(Unmerge.getReg(I));
+
+  const LLT S32 = LLT::scalar(32);
+  int NumElts = StoreVT.getNumElements();
+
+  return B.buildMerge(LLT::vector(NumElts, S32), WideRegs).getReg(0);
+}
+
+static std::pair<Register, unsigned>
+getBaseWithConstantOffset(MachineRegisterInfo &MRI, Register Reg) {
+  int64_t Const;
+  if (mi_match(Reg, MRI, m_ICst(Const)))
+    return std::make_pair(Register(), Const);
+
+  Register Base;
+  if (mi_match(Reg, MRI, m_GAdd(m_Reg(Base), m_ICst(Const))))
+    return std::make_pair(Base, Const);
+
+  // TODO: Handle G_OR used for add case
+  return std::make_pair(Reg, 0);
+}
+
+std::pair<Register, unsigned>
+AMDGPURegisterBankInfo::splitBufferOffsets(MachineIRBuilder &B,
+                                           Register OrigOffset) const {
+  const unsigned MaxImm = 4095;
+  Register BaseReg;
+  unsigned ImmOffset;
+  const LLT S32 = LLT::scalar(32);
+
+  std::tie(BaseReg, ImmOffset) = getBaseWithConstantOffset(*B.getMRI(),
+                                                           OrigOffset);
+
+  unsigned C1 = 0;
+  if (ImmOffset != 0) {
+    // If the immediate value is too big for the immoffset field, put the value
+    // and -4096 into the immoffset field so that the value that is copied/added
+    // for the voffset field is a multiple of 4096, and it stands more chance
+    // of being CSEd with the copy/add for another similar load/store.
+    // However, do not do that rounding down to a multiple of 4096 if that is a
+    // negative number, as it appears to be illegal to have a negative offset
+    // in the vgpr, even if adding the immediate offset makes it positive.
+    unsigned Overflow = ImmOffset & ~MaxImm;
+    ImmOffset -= Overflow;
+    if ((int32_t)Overflow < 0) {
+      Overflow += ImmOffset;
+      ImmOffset = 0;
+    }
+
+    C1 = ImmOffset;
+    if (Overflow != 0) {
+      if (!BaseReg)
+        BaseReg = B.buildConstant(S32, Overflow).getReg(0);
+      else {
+        auto OverflowVal = B.buildConstant(S32, Overflow);
+        BaseReg = B.buildAdd(S32, BaseReg, OverflowVal).getReg(0);
+      }
+    }
+  }
+
+  if (!BaseReg)
+    BaseReg = B.buildConstant(S32, 0).getReg(0);
+
+  return {BaseReg, C1};
+}
+
+static bool isZero(Register Reg, MachineRegisterInfo &MRI) {
+  int64_t C;
+  return mi_match(Reg, MRI, m_ICst(C)) && C == 0;
+}
+
+static unsigned extractGLC(unsigned CachePolicy) {
+  return CachePolicy & 1;
+}
+
+static unsigned extractSLC(unsigned CachePolicy) {
+  return (CachePolicy >> 1) & 1;
+}
+
+static unsigned extractDLC(unsigned CachePolicy) {
+  return (CachePolicy >> 2) & 1;
+}
+
+MachineInstr *
+AMDGPURegisterBankInfo::selectStoreIntrinsic(MachineIRBuilder &B,
+                                             MachineInstr &MI) const {
+   MachineRegisterInfo &MRI = *B.getMRI();
+  executeInWaterfallLoop(B, MI, MRI, {2, 4});
+
+  // FIXME: DAG lowering brokenly changes opcode based on FP vs. integer.
+
+  Register VData = MI.getOperand(1).getReg();
+  LLT Ty = MRI.getType(VData);
+
+  int EltSize = Ty.getScalarSizeInBits();
+  int Size = Ty.getSizeInBits();
+
+  // FIXME: Broken integer truncstore.
+  if (EltSize != 32)
+    report_fatal_error("unhandled intrinsic store");
+
+  // FIXME: Verifier should enforce 1 MMO for these intrinsics.
+  const int MemSize = (*MI.memoperands_begin())->getSize();
+
+
+  Register RSrc = MI.getOperand(2).getReg();
+  Register VOffset = MI.getOperand(3).getReg();
+  Register SOffset = MI.getOperand(4).getReg();
+  unsigned CachePolicy = MI.getOperand(5).getImm();
+
+  unsigned ImmOffset;
+  std::tie(VOffset, ImmOffset) = splitBufferOffsets(B, VOffset);
+
+  const bool Offen = !isZero(VOffset, MRI);
+
+  unsigned Opc = AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact;
+  switch (8 * MemSize) {
+  case 8:
+    Opc = Offen ? AMDGPU::BUFFER_STORE_BYTE_OFFEN_exact :
+                  AMDGPU::BUFFER_STORE_BYTE_OFFSET_exact;
+    break;
+  case 16:
+    Opc = Offen ? AMDGPU::BUFFER_STORE_SHORT_OFFEN_exact :
+                  AMDGPU::BUFFER_STORE_SHORT_OFFSET_exact;
+    break;
+  default:
+    Opc = Offen ? AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact :
+                  AMDGPU::BUFFER_STORE_DWORD_OFFSET_exact;
+    if (Size > 32)
+      Opc = AMDGPU::getMUBUFOpcode(Opc, Size / 32);
+    break;
+  }
+
+
+  // Set the insertion point back to the instruction in case it was moved into a
+  // loop.
+  B.setInstr(MI);
+
+  MachineInstrBuilder MIB = B.buildInstr(Opc)
+    .addUse(VData);
+
+  if (Offen)
+    MIB.addUse(VOffset);
+
+  MIB.addUse(RSrc)
+     .addUse(SOffset)
+     .addImm(ImmOffset)
+     .addImm(extractGLC(CachePolicy))
+     .addImm(extractSLC(CachePolicy))
+     .addImm(0) // tfe: FIXME: Remove from inst
+     .addImm(extractDLC(CachePolicy))
+     .cloneMemRefs(MI);
+
+  // FIXME: We need a way to report failure from applyMappingImpl.
+  // Insert constrain copies before inserting the loop.
+  if (!constrainSelectedInstRegOperands(*MIB, *TII, *TRI, *this))
+    report_fatal_error("failed to constrain selected store intrinsic");
+
+  return MIB;
+}
+
 void AMDGPURegisterBankInfo::applyMappingImpl(
     const OperandsMapper &OpdMapper) const {
   MachineInstr &MI = OpdMapper.getMI();
   unsigned Opc = MI.getOpcode();
   MachineRegisterInfo &MRI = OpdMapper.getMRI();
   switch (Opc) {
   case AMDGPU::G_SELECT: {
     Register DstReg = MI.getOperand(0).getReg();
case AMDGPU::G_INTRINSIC_W_SIDE_EFFECTS: {
     }
     case Intrinsic::amdgcn_s_sendmsg:
     case Intrinsic::amdgcn_s_sendmsghalt: {
       // FIXME: Should this use a waterfall loop?
       constrainOpWithReadfirstlane(MI, MRI, 2); // M0
       return;
     }
     case Intrinsic::amdgcn_raw_buffer_load:
-    case Intrinsic::amdgcn_raw_buffer_store: {
+    case Intrinsic::amdgcn_raw_buffer_load_format:
+    case Intrinsic::amdgcn_raw_buffer_store:
+    case Intrinsic::amdgcn_raw_buffer_store_format: {
       applyDefaultMapping(OpdMapper);
       executeInWaterfallLoop(MI, MRI, {2, 4});
       return;
     }
     case Intrinsic::amdgcn_struct_buffer_load:
     case Intrinsic::amdgcn_struct_buffer_store: {
       applyDefaultMapping(OpdMapper);
       executeInWaterfallLoop(MI, MRI, {2, 5});
       return;
     }
     default: {
       if (const AMDGPU::RsrcIntrinsic *RSrcIntrin =
               AMDGPU::lookupRsrcIntrinsic(IntrID)) {
         // Non-images can have complications from operands that allow both SGPR
         // and VGPR. For now it's too complicated to figure out the final opcode
         // to derive the register bank from the MCInstrDesc.
         if (RSrcIntrin->IsImage) {
           applyMappingImage(MI, OpdMapper, MRI, RSrcIntrin->RsrcArg);
           return;
         }
       }
case Intrinsic::amdgcn_raw_buffer_load: {
       // FIXME: Should make intrinsic ID the last operand of the instruction,
       // then this would be the same as store
       OpdsMapping[0] = getVGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
       OpdsMapping[2] = getSGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
       OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI);
       OpdsMapping[4] = getSGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI);
       break;
     }
-    case Intrinsic::amdgcn_raw_buffer_store: {
+    case Intrinsic::amdgcn_raw_buffer_store:
+    case Intrinsic::amdgcn_raw_buffer_store_format: {
       OpdsMapping[1] = getVGPROpMapping(MI.getOperand(1).getReg(), MRI, *TRI);
       OpdsMapping[2] = getSGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
       OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI);
       OpdsMapping[4] = getSGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI);
       break;
     }
     case Intrinsic::amdgcn_struct_buffer_load: {
       OpdsMapping[0] = getVGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);

lib/Target/AMDGPU/AMDGPUSubtarget.cpp

GCNSubtarget::GCNSubtarget(const Triple &TT, StringRef GPU, StringRef FS,
 
     FeatureDisable(false),
     InstrInfo(initializeSubtargetDependencies(TT, GPU, FS)),
     TLInfo(TM, *this),
     FrameLowering(TargetFrameLowering::StackGrowsUp, getStackAlignment(), 0) {
   MaxWavesPerEU = AMDGPU::IsaInfo::getMaxWavesPerEU(this);
   CallLoweringInfo.reset(new AMDGPUCallLowering(*getTargetLowering()));
   Legalizer.reset(new AMDGPULegalizerInfo(*this, TM));
-  RegBankInfo.reset(new AMDGPURegisterBankInfo(*getRegisterInfo()));
+  RegBankInfo.reset(new AMDGPURegisterBankInfo(*this));
   InstSelector.reset(new AMDGPUInstructionSelector(
   *this, *static_cast<AMDGPURegisterBankInfo *>(RegBankInfo.get()), TM));
 }
 
 unsigned GCNSubtarget::getConstantBusLimit(unsigned Opcode) const {
   if (getGeneration() < GFX10)
     return 1;
 

test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.raw.buffer.store.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
+; FIXME: Test with SI when argument lowering not broken for f16
+
+; Natural mapping
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+; Copies for VGPR arguments
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float inreg %val, i32 inreg %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr6
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr7
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr8
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
+  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY7]], [[COPY8]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc
+define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; CHECK:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY7]], implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
+  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE3]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+; Waterfall for soffset
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
+  ; CHECK:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY6]], implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc and soffset
+define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; CHECK:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY7]], implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
+  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
+  ; CHECK:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec
+  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 1, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 1)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 1, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 2)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 1, 1, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 3)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_dlc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 1, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 4)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc_dlc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc_dlc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 1, 0, 1, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 6)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_dlc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_dlc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 1, 0, 0, 1, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 5)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc_dlc(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc_dlc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 1, 1, 0, 1, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 7)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32(<4 x i32> inreg %rsrc, <2 x float> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
+  ; CHECK:   BUFFER_STORE_DWORDX2_OFFEN_exact [[REG_SEQUENCE1]], [[COPY6]], [[REG_SEQUENCE]], [[COPY7]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v3f32(<4 x i32> inreg %rsrc, <3 x float> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v3f32
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY8:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2
+  ; CHECK:   BUFFER_STORE_DWORDX3_OFFEN_exact [[REG_SEQUENCE1]], [[COPY7]], [[REG_SEQUENCE]], [[COPY8]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 12 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v3f32(<3 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32(<4 x i32> inreg %rsrc, <4 x float> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[COPY9:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY7]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORDX4_OFFEN_exact [[REG_SEQUENCE1]], [[COPY8]], [[REG_SEQUENCE]], [[COPY9]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 16 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v4f32(<4 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i8(<4 x i32> inreg %rsrc, i32 %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i8
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_BYTE_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 1 into custom TargetCustom7, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %val.trunc = trunc i32 %val to i8
+  call void @llvm.amdgcn.raw.buffer.store.i8(i8 %val.trunc, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i16(<4 x i32> inreg %rsrc, i32 %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_SHORT_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 2 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %val.trunc = trunc i32 %val to i16
+  call void @llvm.amdgcn.raw.buffer.store.i16(i16 %val.trunc, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f16(<4 x i32> inreg %rsrc, half %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_SHORT_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 2 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16(<4 x i32> inreg %rsrc, <2 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16(<4 x i32> inreg %rsrc, <4 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
+  ; CHECK:   BUFFER_STORE_DWORDX2_OFFEN_exact [[REG_SEQUENCE1]], [[COPY6]], [[REG_SEQUENCE]], [[COPY7]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v4f16(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16(<4 x i32> %rsrc, <4 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; CHECK:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1
+  ; CHECK:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; CHECK:   [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[COPY9]], implicit $exec
+  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; CHECK:   [[REG_SEQUENCE4:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORDX2_OFFEN_exact [[REG_SEQUENCE1]], [[COPY6]], [[REG_SEQUENCE4]], [[COPY7]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v4f16(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4095(<4 x i32> inreg %rsrc, float %val, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4095
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], $noreg, [[REG_SEQUENCE]], [[COPY5]], 4095, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7 + 4095, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 4095, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4096(<4 x i32> inreg %rsrc, float %val, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4096
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[V_MOV_B32_e32_]], [[REG_SEQUENCE]], [[COPY5]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7 + 4096, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 4096, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_16(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 16
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %13:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 16
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4095(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4095
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 4095
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %13:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 4095
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4096(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4096
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 4096
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %13:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 4096
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4095(<4 x i32> inreg %rsrc, <2 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4095
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 4095, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4096(<4 x i32> inreg %rsrc, <2 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4096
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[COPY5]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 4096, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_16(<4 x i32> inreg %rsrc, <2 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_16
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 16
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %13:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 16
+  call void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4095(<4 x i32> inreg %rsrc, <2 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4095
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 4095
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %13:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 4095
+  call void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4096(<4 x i32> inreg %rsrc, <2 x half> %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4096
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 4096
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %13:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 4096
+  call void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+; An add of the offset is necessary, with a waterfall loop. Make sure the add is done outside of the waterfall loop.
+define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_offset_add_5000(<4 x i32> %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_offset_add_5000
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; CHECK:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 5000
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
+  ; CHECK:   %11:vgpr_32, dead %29:sreg_64_xexec = V_ADD_I32_e64 [[COPY5]], [[COPY7]], 0, implicit $exec
+  ; CHECK:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; CHECK:   [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY8]], implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY9]], implicit $exec
+  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], %11, [[REG_SEQUENCE3]], [[COPY6]], 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   S_ENDPGM 0
+  %voffset.add = add i32 %voffset, 5000
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+; An add of the offset is necessary, with a waterfall loop. Make sure the add is done outside of the waterfall loop.
+define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__5000_voffset__sgpr_soffset_offset(<4 x i32> %rsrc, float %val, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__5000_voffset__sgpr_soffset_offset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
+  ; CHECK:   [[COPY6:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; CHECK:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY6]], implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY7]], implicit $exec
+  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; CHECK:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY4]], [[V_MOV_B32_e32_]], [[REG_SEQUENCE3]], [[COPY5]], 904, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into custom TargetCustom7 + 5000, align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   S_ENDPGM 0
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 5000, i32 %soffset, i32 0)
+  ret void
+}
+
+declare void @llvm.amdgcn.raw.buffer.store.i8(i8, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.i16(i16, <4 x i32>, i32, i32, i32 immarg)
+
+declare void @llvm.amdgcn.raw.buffer.store.f16(half, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.v2f16(<2 x half>, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.v4f16(<4 x half>, <4 x i32>, i32, i32, i32 immarg)
+
+declare void @llvm.amdgcn.raw.buffer.store.f32(float, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.v2f32(<2 x float>, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.v3f32(<3 x float>, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32 immarg)

test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.raw.buffer.store.ll

-; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s
-
-; Natural mapping
-define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
-  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
-  ; CHECK:   G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY5]](s32), [[COPY6]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
-  ; CHECK:   S_ENDPGM 0
-  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
-  ret void
-}
-
-; Copies for VGPR arguments
-define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, float inreg %val, i32 inreg %voffset, i32 inreg %soffset) {
-  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
-  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY [[COPY4]](s32)
-  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY [[COPY5]](s32)
-  ; CHECK:   G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY7]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY8]](s32), [[COPY6]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
-  ; CHECK:   S_ENDPGM 0
-  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
-  ret void
-}
-
-; Waterfall for rsrc
-define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 inreg %soffset) {
-  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
-  ; CHECK:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.1, %13, %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR1]](<4 x s32>), [[COPY5]](s32), [[COPY6]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   S_ENDPGM 0
-  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
-  ret void
-}
-
-; Waterfall for soffset
-define amdgpu_ps void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> inreg %rsrc, float %val, i32 %voffset, i32 %soffset) {
-  ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr2
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.1, %13, %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY6]](s32), implicit $exec
-  ; CHECK:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]](s32), [[COPY6]](s32), implicit $exec
-  ; CHECK:   G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR]](<4 x s32>), [[COPY5]](s32), [[V_READFIRSTLANE_B32_]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   S_ENDPGM 0
-  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
-  ret void
-}
-
-; Waterfall for rsrc and soffset
-define amdgpu_ps void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
-  ; CHECK-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
-  ; CHECK:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
-  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr6
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64 = PHI [[DEF]], %bb.1, %13, %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY6]](s32), implicit $exec
-  ; CHECK:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]](s32), [[COPY6]](s32), implicit $exec
-  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
-  ; CHECK:   G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.raw.buffer.store), [[COPY4]](s32), [[BUILD_VECTOR1]](<4 x s32>), [[COPY5]](s32), [[V_READFIRSTLANE_B32_4]](s32), 0 :: (dereferenceable store 4 into custom TargetCustom7, align 1, addrspace 4)
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64 = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   S_ENDPGM 0
-  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
-  ret void
-}
-
-declare void @llvm.amdgcn.raw.buffer.store.f32(float, <4 x i32>, i32, i32, i32 immarg)