[AMDGPU] Legalize soffset of buffer instruction. Use Waterfall loop logic.

[skc7]

Summary

Legalize soffset of buffer instructions. If vgpr is assigned to soffset, use waterfall loop logic to legalize.

Update loadSRsrcFromVGPR method to use for rsrc and soffset operands.

[skc7]

Use loadMBUFScalarOperandFromVGPR to build waterfall loop for soffset and srsrc.

[skc7]

Rebase

[skc7]

Rebase

[arsenm]

I think this will mishandle the case where both the SRD and the soffset are VGPRs. You need to handle both at the same time in one waterfall loop (this should show up if your tests used a meaningful SRD). You can also just look into the globalisel tests for these intrinsics, they test all the permutations already

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

5774

Don't understand the isIdenticalTo check, this will miss flag mismatches. Register and subregister equality should be sufficient (maybe it doesn't matter based on context)

llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll

35

All of these tests are using poison for the SRD, these should use a meaningful value

[skc7]

Make changes to legalize soffset and rsrc together.

[skc7]

In D141030#4059355, @arsenm wrote:

I think this will mishandle the case where both the SRD and the soffset are VGPRs. You need to handle both at the same time in one waterfall loop (this should show up if your tests used a meaningful SRD). You can also just look into the globalisel tests for these intrinsics, they test all the permutations already

Made changes to legalize and use single water fall loop for rsrc and soffset. As per my understanding, SiFixSGPRCopies pass makes call to moveToVALU sequentially (not together for soffset and rsrc) for operands. So we see two water fall loops in the tests.

[skc7]

Rebase. Update test.

[skc7]

Port MBUF related tests from global-isel containing all the test combinations.

[skc7]

llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll

35

These tests are checking if a waterfall loop is created if only soffset is illegal. So, SRD has been made poison.

[skc7]

Rebase.

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

for (auto &Succ : RemainderBB->successors()) {
       if (MDT->properlyDominates(&MBB, Succ)) {
         MDT->changeImmediateDominator(Succ, RemainderBB);
       }
     }
   }
 
   emitLoadSRsrcFromVGPRLoop(TII, MRI, MBB, *LoopBB, *BodyBB, DL, Rsrc);
 
   // Restore the EXEC mask
   MachineBasicBlock::iterator First = RemainderBB->begin();
   BuildMI(*RemainderBB, First, DL, TII.get(MovExecOpc), Exec).addReg(SaveExec);
   return BodyBB;
 }
 
+// Emit the actual waterfall loop, executing the wrapped instruction for each
+// unique value of \p Soffset across all lanes. In the best case we execute 1
+// iteration, in the worst case we execute 64 (once per lane).
+static void
+emitLoadSoffsetFromVGPRLoop(const SIInstrInfo &TII, MachineRegisterInfo &MRI,
+                          MachineBasicBlock &OrigBB, MachineBasicBlock &LoopBB,
+                          MachineBasicBlock &BodyBB, const DebugLoc &DL,
+                          MachineOperand &Soffset) {
+  MachineFunction &MF = *OrigBB.getParent();
+  const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
+  const SIRegisterInfo *TRI = ST.getRegisterInfo();
+  unsigned Exec = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
+  unsigned SaveExecOpc =
+      ST.isWave32() ? AMDGPU::S_AND_SAVEEXEC_B32 : AMDGPU::S_AND_SAVEEXEC_B64;
+  unsigned XorTermOpc =
+      ST.isWave32() ? AMDGPU::S_XOR_B32_term : AMDGPU::S_XOR_B64_term;
+  const auto *BoolXExecRC = TRI->getRegClass(AMDGPU::SReg_1_XEXECRegClassID);
+
+  MachineBasicBlock::iterator I = LoopBB.begin();
+  Register VSoffset = Soffset.getReg();
+  unsigned RegSize = TRI->getRegSizeInBits(Soffset.getReg(), MRI);
+  unsigned NumSubRegs =  RegSize / 32;
+  assert(NumSubRegs == 1 && "Unhandled register size");
+
+  Register CurReg = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+
+  // Read the next variant <- also loop target.
+  BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_READFIRSTLANE_B32), CurReg)
+            .addReg(VSoffset);
+
+  Register CondReg = MRI.createVirtualRegister(BoolXExecRC);
+  BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_CMP_EQ_U32_e64), CondReg)
+          .addReg(CurReg)
+          .addReg(VSoffset);
+
+  // Update Soffset operand to use the SGPR Soffset.
+  Soffset.setReg(CurReg);
+  Soffset.setIsKill();
+
+  Register SaveExec = MRI.createVirtualRegister(BoolXExecRC);
+  MRI.setSimpleHint(SaveExec, CondReg);
+
+  // Update EXEC to matching lanes, saving original to SaveExec.
+  BuildMI(LoopBB, I, DL, TII.get(SaveExecOpc), SaveExec)
+      .addReg(CondReg, RegState::Kill);
+
+  // The original instruction is here; we insert the terminators after it.
+  I = BodyBB.end();
+
+  // Update EXEC, switch all done bits to 0 and all todo bits to 1.
+  BuildMI(BodyBB, I, DL, TII.get(XorTermOpc), Exec)
+      .addReg(Exec)
+      .addReg(SaveExec);
+
+  BuildMI(BodyBB, I, DL, TII.get(AMDGPU::SI_WATERFALL_LOOP)).addMBB(&LoopBB);
+}
+
+// Build a waterfall loop around \p MI, replacing the VGPR \p Soffset register
+// with SGPRs by iterating over all unique values across all lanes.
+// Returns the loop basic block that now contains \p MI.
+static MachineBasicBlock *
+loadSoffsetFromVGPR(const SIInstrInfo &TII, MachineInstr &MI,
+                  MachineOperand &Rsrc, MachineDominatorTree *MDT,
+                  MachineBasicBlock::iterator Begin = nullptr,
+                  MachineBasicBlock::iterator End = nullptr) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
+  const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
+  const SIRegisterInfo *TRI = ST.getRegisterInfo();
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  if (!Begin.isValid())
+    Begin = &MI;
+  if (!End.isValid()) {
+    End = &MI;
+    ++End;
+  }
+  const DebugLoc &DL = MI.getDebugLoc();
+  unsigned Exec = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
+  unsigned MovExecOpc = ST.isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
+  const auto *BoolXExecRC = TRI->getRegClass(AMDGPU::SReg_1_XEXECRegClassID);
+
+  Register SaveExec = MRI.createVirtualRegister(BoolXExecRC);
+
+  // Save the EXEC mask
+  BuildMI(MBB, Begin, DL, TII.get(MovExecOpc), SaveExec).addReg(Exec);
+
+  // Killed uses in the instruction we are waterfalling around will be
+  // incorrect due to the added control-flow.
+  MachineBasicBlock::iterator AfterMI = MI;
+  ++AfterMI;
+  for (auto I = Begin; I != AfterMI; I++) {
+    for (auto &MO : I->uses()) {
+      if (MO.isReg() && MO.isUse()) {
+        MRI.clearKillFlags(MO.getReg());
+      }
+    }
+  }
+
+  // To insert the loop we need to split the block. Move everything after this
+  // point to a new block, and insert a new empty block between the two.
+  MachineBasicBlock *LoopBB = MF.CreateMachineBasicBlock();
+  MachineBasicBlock *BodyBB = MF.CreateMachineBasicBlock();
+  MachineBasicBlock *RemainderBB = MF.CreateMachineBasicBlock();
+  MachineFunction::iterator MBBI(MBB);
+  ++MBBI;
+
+  MF.insert(MBBI, LoopBB);
+  MF.insert(MBBI, BodyBB);
+  MF.insert(MBBI, RemainderBB);
+
+  LoopBB->addSuccessor(BodyBB);
+  BodyBB->addSuccessor(LoopBB);
+  BodyBB->addSuccessor(RemainderBB);
+
+  // Move Begin to MI to the BodyBB, and the remainder of the block to
+  // RemainderBB.
+  RemainderBB->transferSuccessorsAndUpdatePHIs(&MBB);
+  RemainderBB->splice(RemainderBB->begin(), &MBB, End, MBB.end());
+  BodyBB->splice(BodyBB->begin(), &MBB, Begin, MBB.end());
+
+  MBB.addSuccessor(LoopBB);
+
+  // Update dominators. We know that MBB immediately dominates LoopBB, that
+  // LoopBB immediately dominates BodyBB, and BodyBB immediately dominates
+  // RemainderBB. RemainderBB immediately dominates all of the successors
+  // transferred to it from MBB that MBB used to properly dominate.
+  if (MDT) {
+    MDT->addNewBlock(LoopBB, &MBB);
+    MDT->addNewBlock(BodyBB, LoopBB);
+    MDT->addNewBlock(RemainderBB, BodyBB);
+    for (auto &Succ : RemainderBB->successors()) {
+      if (MDT->properlyDominates(&MBB, Succ)) {
+        MDT->changeImmediateDominator(Succ, RemainderBB);
+      }
+    }
+  }
+
+  emitLoadSoffsetFromVGPRLoop(TII, MRI, MBB, *LoopBB, *BodyBB, DL, Rsrc);
+
+  // Restore the EXEC mask
+  MachineBasicBlock::iterator First = RemainderBB->begin();
+  BuildMI(*RemainderBB, First, DL, TII.get(MovExecOpc), Exec).addReg(SaveExec);
+  return BodyBB;
+}
+
 // Extract pointer from Rsrc and return a zero-value Rsrc replacement.
 static std::tuple<unsigned, unsigned>
 extractRsrcPtr(const SIInstrInfo &TII, MachineInstr &MI, MachineOperand &Rsrc) {
   MachineBasicBlock &MBB = *MI.getParent();
   MachineFunction &MF = *MBB.getParent();
   MachineRegisterInfo &MRI = MF.getRegInfo();
 
   // Extract the ptr from the resource descriptor.
if (!RI.isSGPRClass(MRI.getRegClass(Dest->getReg()))) {
       ++End;
       while (End != MBB.end() && End->isCopy() && End->getOperand(1).isReg() &&
              MI.definesRegister(End->getOperand(1).getReg()))
         ++End;
       CreatedBB = loadSRsrcFromVGPR(*this, MI, *Dest, MDT, Start, End);
     }
   }
 
-  // Legalize MUBUF* instructions.
+  // Legalize soffset of MUBUF instructions.
+  int SoffsetIdx =
+      AMDGPU::getNamedOperandIdx(MI.getOpcode(), AMDGPU::OpName::soffset);
+  if (SoffsetIdx != -1) {
+    MachineOperand *Soffset = &MI.getOperand(SoffsetIdx);
+    if (Soffset->isReg() && !RI.isSGPRClass(MRI.getRegClass(Soffset->getReg()))) {
+      CreatedBB = loadSoffsetFromVGPR(*this, MI, *Soffset, MDT);
+    }
+  }
+
   int RsrcIdx =
       AMDGPU::getNamedOperandIdx(MI.getOpcode(), AMDGPU::OpName::srsrc);
   if (RsrcIdx != -1) {
     // We have an MUBUF instruction
     MachineOperand *Rsrc = &MI.getOperand(RsrcIdx);
     unsigned RsrcRC = get(MI.getOpcode()).OpInfo[RsrcIdx].RegClass;
     if (RI.getCommonSubClass(MRI.getRegClass(Rsrc->getReg()),
                              RI.getRegClass(RsrcRC))) {

llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc < %s -march=amdgcn -mcpu=gfx906 -stop-after=si-fix-sgpr-copies -verify-machineinstrs | FileCheck %s
+
+define float @llvm_amdgcn_raw_buffer_load_f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_load_f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @llvm_amdgcn_raw_tbuffer_load_f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_load_f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret float %val
+}
+
+define <2 x float> @llvm_amdgcn_raw_buffer_load_v2f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_load_v2f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[BUFFER_LOAD_DWORDX2_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub0
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub1
+  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret <2 x float> %val
+}
+
+define <2 x float> @llvm_amdgcn_raw_tbuffer_load_v2f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_load_v2f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
+  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret <2 x float> %val
+}
+
+define <3 x float> @llvm_amdgcn_raw_buffer_load_v3f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_load_v3f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[BUFFER_LOAD_DWORDX3_OFFEN:%[0-9]+]]:vreg_96 = BUFFER_LOAD_DWORDX3_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub0
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub1
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub2
+  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+  %val = call <3 x float> @llvm.amdgcn.raw.buffer.load.v3f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret <3 x float> %val
+}
+
+define <3 x float> @llvm_amdgcn_raw_tbuffer_load_v3f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_load_v3f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
+  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+  %val = call <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret <3 x float> %val
+}
+
+define <4 x float> @llvm_amdgcn_raw_buffer_load_v4f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_load_v4f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub0
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub1
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub2
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub3
+  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; CHECK-NEXT:   $vgpr3 = COPY [[COPY5]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  %val = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x float> %val
+}
+
+define <4 x float> @llvm_amdgcn_raw_tbuffer_load_v4f32(i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_load_v4f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
+  ; CHECK-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; CHECK-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; CHECK-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; CHECK-NEXT:   $vgpr3 = COPY [[COPY5]]
+  ; CHECK-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  %val = call <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret <4 x float> %val
+}
+
+define void @llvm_amdgcn_raw_buffer_store_f32(float %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_store_f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_tbuffer_store_f32(float %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_store_f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY2]], [[COPY1]], [[DEF]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_buffer_store_v2f32(<2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_store_v2f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY3]], implicit $exec
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[COPY4]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
+  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF3:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   BUFFER_STORE_DWORDX2_OFFEN_exact [[COPY5]], [[COPY1]], [[DEF2]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.v2f32(<2 x float> %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_tbuffer_store_v2f32(<2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_store_v2f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY3]], implicit $exec
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[COPY4]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
+  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF3:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   TBUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY5]], [[COPY1]], [[DEF2]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v2f32(<2 x float> %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_buffer_store_v3f32(<3 x float> %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_store_v3f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY4]], implicit $exec
+  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY3]], implicit $exec
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[COPY5]], %subreg.sub0, killed [[COPY6]], %subreg.sub1, [[COPY2]], %subreg.sub2
+  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE]]
+  ; CHECK-NEXT:   [[DEF3:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   BUFFER_STORE_DWORDX3_OFFEN_exact [[COPY7]], [[COPY1]], [[DEF3]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.v3f32(<3 x float> %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_tbuffer_store_v3f32(<3 x float> %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_store_v3f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY4]], implicit $exec
+  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY2]], implicit $exec
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_96 = REG_SEQUENCE killed [[COPY5]], %subreg.sub0, [[COPY3]], %subreg.sub1, killed [[COPY6]], %subreg.sub2
+  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE]]
+  ; CHECK-NEXT:   [[DEF3:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   TBUFFER_STORE_FORMAT_XYZ_OFFEN_exact [[COPY7]], [[COPY1]], [[DEF3]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v3f32(<3 x float> %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_buffer_store_v4f32(<4 x float> %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_store_v4f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY5]], implicit $exec
+  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]], implicit $exec
+  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY2]], implicit $exec
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[COPY6]], %subreg.sub0, killed [[COPY7]], %subreg.sub1, [[COPY3]], %subreg.sub2, killed [[COPY8]], %subreg.sub3
+  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE]]
+  ; CHECK-NEXT:   [[DEF4:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF5:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   BUFFER_STORE_DWORDX4_OFFEN_exact [[COPY9]], [[COPY1]], [[DEF4]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.v4f32(<4 x float> %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @llvm_amdgcn_raw_tbuffer_store_v4f32(<4 x float> %val, i32 %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_tbuffer_store_v4f32
+  ; CHECK: bb.0 (%ir-block.0):
+  ; CHECK-NEXT:   successors: %bb.1(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY5]], implicit $exec
+  ; CHECK-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY3]], implicit $exec
+  ; CHECK-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY2]], implicit $exec
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE killed [[COPY6]], %subreg.sub0, [[COPY4]], %subreg.sub1, killed [[COPY7]], %subreg.sub2, killed [[COPY8]], %subreg.sub3
+  ; CHECK-NEXT:   [[COPY9:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE]]
+  ; CHECK-NEXT:   [[DEF4:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[DEF5:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; CHECK-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[V_CMP_EQ_U32_e64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   TBUFFER_STORE_FORMAT_XYZW_OFFEN_exact [[COPY9]], [[COPY1]], [[DEF4]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; CHECK-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; CHECK-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; CHECK-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %val, <4 x i32> poison, i32 %voffset, i32 %soffset, i32 0, i32 0)
+  ret void
+}
+
+declare float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32>, i32, i32, i32 )
+declare float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32>, i32, i32, i32, i32)
+declare <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32>, i32, i32, i32)
+declare <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32>, i32, i32, i32, i32)
+declare <3 x float> @llvm.amdgcn.raw.buffer.load.v3f32(<4 x i32>, i32, i32, i32)
+declare <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32>, i32, i32, i32, i32)
+declare <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32>, i32, i32, i32)
+declare <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32, i32)
+declare void @llvm.amdgcn.raw.buffer.store.f32(float, <4 x i32>, i32, i32, i32)
+declare void @llvm.amdgcn.raw.tbuffer.store.f32(float, <4 x i32>, i32, i32, i32, i32)
+declare void @llvm.amdgcn.raw.buffer.store.v2f32(<2 x float>, <4 x i32>, i32, i32, i32)
+declare void @llvm.amdgcn.raw.tbuffer.store.v2f32(<2 x float>, <4 x i32>, i32, i32, i32, i32)
+declare void @llvm.amdgcn.raw.buffer.store.v3f32(<3 x float>, <4 x i32>, i32, i32, i32)
+declare void @llvm.amdgcn.raw.tbuffer.store.v3f32(<3 x float>, <4 x i32>, i32, i32, i32, i32)
+declare void @llvm.amdgcn.raw.buffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32)
+declare void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32, i32)