diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index 055441d2488b..a6e11e3194e4 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -5599,13 +5599,12 @@ void SIInstrInfo::legalizeGenericOperand(MachineBasicBlock &InsertMBB,
 }
 
 // Emit the actual waterfall loop, executing the wrapped instruction for each
-// unique value of \p Rsrc across all lanes. In the best case we execute 1
+// unique value of \p ScalarOps across all lanes. In the best case we execute 1
 // iteration, in the worst case we execute 64 (once per lane).
-static void
-emitLoadSRsrcFromVGPRLoop(const SIInstrInfo &TII, MachineRegisterInfo &MRI,
-                          MachineBasicBlock &OrigBB, MachineBasicBlock &LoopBB,
-                          MachineBasicBlock &BodyBB, const DebugLoc &DL,
-                          MachineOperand &Rsrc) {
+static void emitLoadScalarOpsFromVGPRLoop(
+    const SIInstrInfo &TII, MachineRegisterInfo &MRI, MachineBasicBlock &OrigBB,
+    MachineBasicBlock &LoopBB, MachineBasicBlock &BodyBB, const DebugLoc &DL,
+    ArrayRef<MachineOperand *> ScalarOps) {
   MachineFunction &MF = *OrigBB.getParent();
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   const SIRegisterInfo *TRI = ST.getRegisterInfo();
@@ -5623,73 +5622,106 @@ emitLoadSRsrcFromVGPRLoop(const SIInstrInfo &TII, MachineRegisterInfo &MRI,
   SmallVector<Register, 8> ReadlanePieces;
   Register CondReg;
 
-  Register VRsrc = Rsrc.getReg();
-  unsigned VRsrcUndef = getUndefRegState(Rsrc.isUndef());
+  for (MachineOperand *ScalarOp : ScalarOps) {
+    unsigned RegSize = TRI->getRegSizeInBits(ScalarOp->getReg(), MRI);
+    unsigned NumSubRegs = RegSize / 32;
+    Register VScalarOp = ScalarOp->getReg();
 
-  unsigned RegSize = TRI->getRegSizeInBits(Rsrc.getReg(), MRI);
-  unsigned NumSubRegs =  RegSize / 32;
-  assert(NumSubRegs % 2 == 0 && NumSubRegs <= 32 && "Unhandled register size");
+    if (NumSubRegs == 1) {
+      Register CurReg = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
 
-  for (unsigned Idx = 0; Idx < NumSubRegs; Idx += 2) {
+      BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_READFIRSTLANE_B32), CurReg)
+          .addReg(VScalarOp);
 
-    Register CurRegLo = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
-    Register CurRegHi = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+      Register NewCondReg = MRI.createVirtualRegister(BoolXExecRC);
 
-    // Read the next variant <- also loop target.
-    BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_READFIRSTLANE_B32), CurRegLo)
-            .addReg(VRsrc, VRsrcUndef, TRI->getSubRegFromChannel(Idx));
+      BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_CMP_EQ_U32_e64), NewCondReg)
+          .addReg(CurReg)
+          .addReg(VScalarOp);
 
-    // Read the next variant <- also loop target.
-    BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_READFIRSTLANE_B32), CurRegHi)
-            .addReg(VRsrc, VRsrcUndef, TRI->getSubRegFromChannel(Idx + 1));
+      // Combine the comparison results with AND.
+      if (!CondReg) // First.
+        CondReg = NewCondReg;
+      else { // If not the first, we create an AND.
+        Register AndReg = MRI.createVirtualRegister(BoolXExecRC);
+        BuildMI(LoopBB, I, DL, TII.get(AndOpc), AndReg)
+            .addReg(CondReg)
+            .addReg(NewCondReg);
+        CondReg = AndReg;
+      }
 
-    ReadlanePieces.push_back(CurRegLo);
-    ReadlanePieces.push_back(CurRegHi);
+      // Update ScalarOp operand to use the SGPR ScalarOp.
+      ScalarOp->setReg(CurReg);
+      ScalarOp->setIsKill();
+    } else {
+      unsigned VScalarOpUndef = getUndefRegState(ScalarOp->isUndef());
+      assert(NumSubRegs % 2 == 0 && NumSubRegs <= 32 &&
+             "Unhandled register size");
 
-    // Comparison is to be done as 64-bit.
-    Register CurReg = MRI.createVirtualRegister(&AMDGPU::SGPR_64RegClass);
-    BuildMI(LoopBB, I, DL, TII.get(AMDGPU::REG_SEQUENCE), CurReg)
+      for (unsigned Idx = 0; Idx < NumSubRegs; Idx += 2) {
+        Register CurRegLo = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+        Register CurRegHi = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+
+        // Read the next variant <- also loop target.
+        BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_READFIRSTLANE_B32), CurRegLo)
+            .addReg(VScalarOp, VScalarOpUndef, TRI->getSubRegFromChannel(Idx));
+
+        // Read the next variant <- also loop target.
+        BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_READFIRSTLANE_B32), CurRegHi)
+            .addReg(VScalarOp, VScalarOpUndef,
+                    TRI->getSubRegFromChannel(Idx + 1));
+
+        ReadlanePieces.push_back(CurRegLo);
+        ReadlanePieces.push_back(CurRegHi);
+
+        // Comparison is to be done as 64-bit.
+        Register CurReg = MRI.createVirtualRegister(&AMDGPU::SGPR_64RegClass);
+        BuildMI(LoopBB, I, DL, TII.get(AMDGPU::REG_SEQUENCE), CurReg)
             .addReg(CurRegLo)
             .addImm(AMDGPU::sub0)
             .addReg(CurRegHi)
             .addImm(AMDGPU::sub1);
 
-    Register NewCondReg = MRI.createVirtualRegister(BoolXExecRC);
-    auto Cmp =
-        BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_CMP_EQ_U64_e64), NewCondReg)
-            .addReg(CurReg);
-    if (NumSubRegs <= 2)
-      Cmp.addReg(VRsrc);
-    else
-      Cmp.addReg(VRsrc, VRsrcUndef, TRI->getSubRegFromChannel(Idx, 2));
+        Register NewCondReg = MRI.createVirtualRegister(BoolXExecRC);
+        auto Cmp = BuildMI(LoopBB, I, DL, TII.get(AMDGPU::V_CMP_EQ_U64_e64),
+                           NewCondReg)
+                       .addReg(CurReg);
+        if (NumSubRegs <= 2)
+          Cmp.addReg(VScalarOp);
+        else
+          Cmp.addReg(VScalarOp, VScalarOpUndef,
+                     TRI->getSubRegFromChannel(Idx, 2));
 
-    // Combine the comparison results with AND.
-    if (!CondReg) // First.
-      CondReg = NewCondReg;
-    else { // If not the first, we create an AND.
-      Register AndReg = MRI.createVirtualRegister(BoolXExecRC);
-      BuildMI(LoopBB, I, DL, TII.get(AndOpc), AndReg)
+        // Combine the comparison results with AND.
+        if (!CondReg) // First.
+          CondReg = NewCondReg;
+        else { // If not the first, we create an AND.
+          Register AndReg = MRI.createVirtualRegister(BoolXExecRC);
+          BuildMI(LoopBB, I, DL, TII.get(AndOpc), AndReg)
               .addReg(CondReg)
               .addReg(NewCondReg);
-      CondReg = AndReg;
+          CondReg = AndReg;
+        }
+      } // End for loop.
+
+      auto SScalarOpRC =
+          TRI->getEquivalentSGPRClass(MRI.getRegClass(VScalarOp));
+      Register SScalarOp = MRI.createVirtualRegister(SScalarOpRC);
+
+      // Build scalar ScalarOp.
+      auto Merge =
+          BuildMI(LoopBB, I, DL, TII.get(AMDGPU::REG_SEQUENCE), SScalarOp);
+      unsigned Channel = 0;
+      for (Register Piece : ReadlanePieces) {
+        Merge.addReg(Piece).addImm(TRI->getSubRegFromChannel(Channel++));
+      }
+
+      // Update ScalarOp operand to use the SGPR ScalarOp.
+      ScalarOp->setReg(SScalarOp);
+      ScalarOp->setIsKill();
     }
-  } // End for loop.
-
-  auto SRsrcRC = TRI->getEquivalentSGPRClass(MRI.getRegClass(VRsrc));
-  Register SRsrc = MRI.createVirtualRegister(SRsrcRC);
-
-  // Build scalar Rsrc.
-  auto Merge = BuildMI(LoopBB, I, DL, TII.get(AMDGPU::REG_SEQUENCE), SRsrc);
-  unsigned Channel = 0;
-  for (Register Piece : ReadlanePieces) {
-    Merge.addReg(Piece)
-         .addImm(TRI->getSubRegFromChannel(Channel++));
   }
 
-  // Update Rsrc operand to use the SGPR Rsrc.
-  Rsrc.setReg(SRsrc);
-  Rsrc.setIsKill();
-
   Register SaveExec = MRI.createVirtualRegister(BoolXExecRC);
   MRI.setSimpleHint(SaveExec, CondReg);
 
@@ -5708,14 +5740,15 @@ emitLoadSRsrcFromVGPRLoop(const SIInstrInfo &TII, MachineRegisterInfo &MRI,
   BuildMI(BodyBB, I, DL, TII.get(AMDGPU::SI_WATERFALL_LOOP)).addMBB(&LoopBB);
 }
 
-// Build a waterfall loop around \p MI, replacing the VGPR \p Rsrc register
+// Build a waterfall loop around \p MI, replacing the VGPR \p ScalarOp register
 // with SGPRs by iterating over all unique values across all lanes.
 // Returns the loop basic block that now contains \p MI.
 static MachineBasicBlock *
-loadSRsrcFromVGPR(const SIInstrInfo &TII, MachineInstr &MI,
-                  MachineOperand &Rsrc, MachineDominatorTree *MDT,
-                  MachineBasicBlock::iterator Begin = nullptr,
-                  MachineBasicBlock::iterator End = nullptr) {
+loadMBUFScalarOperandsFromVGPR(const SIInstrInfo &TII, MachineInstr &MI,
+                               ArrayRef<MachineOperand *> ScalarOps,
+                               MachineDominatorTree *MDT,
+                               MachineBasicBlock::iterator Begin = nullptr,
+                               MachineBasicBlock::iterator End = nullptr) {
   MachineBasicBlock &MBB = *MI.getParent();
   MachineFunction &MF = *MBB.getParent();
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
@@ -5788,7 +5821,7 @@ loadSRsrcFromVGPR(const SIInstrInfo &TII, MachineInstr &MI,
     }
   }
 
-  emitLoadSRsrcFromVGPRLoop(TII, MRI, MBB, *LoopBB, *BodyBB, DL, Rsrc);
+  emitLoadScalarOpsFromVGPRLoop(TII, MRI, MBB, *LoopBB, *BodyBB, DL, ScalarOps);
 
   // Restore the EXEC mask
   MachineBasicBlock::iterator First = RemainderBB->begin();
@@ -5985,11 +6018,11 @@ SIInstrInfo::legalizeOperands(MachineInstr &MI,
                      (isMUBUF(MI) || isMTBUF(MI)))) {
     MachineOperand *SRsrc = getNamedOperand(MI, AMDGPU::OpName::srsrc);
     if (SRsrc && !RI.isSGPRClass(MRI.getRegClass(SRsrc->getReg())))
-      CreatedBB = loadSRsrcFromVGPR(*this, MI, *SRsrc, MDT);
+      CreatedBB = loadMBUFScalarOperandsFromVGPR(*this, MI, {SRsrc}, MDT);
 
     MachineOperand *SSamp = getNamedOperand(MI, AMDGPU::OpName::ssamp);
     if (SSamp && !RI.isSGPRClass(MRI.getRegClass(SSamp->getReg())))
-      CreatedBB = loadSRsrcFromVGPR(*this, MI, *SSamp, MDT);
+      CreatedBB = loadMBUFScalarOperandsFromVGPR(*this, MI, {SSamp}, MDT);
 
     return CreatedBB;
   }
@@ -6017,25 +6050,39 @@ SIInstrInfo::legalizeOperands(MachineInstr &MI,
       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);
+      CreatedBB =
+          loadMBUFScalarOperandsFromVGPR(*this, MI, {Dest}, MDT, Start, End);
     }
   }
 
-  // Legalize MUBUF* instructions.
+  // Legalize MUBUF instructions.
+  bool isSoffsetLegal = true;
+  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()))) {
+      isSoffsetLegal = false;
+    }
+  }
+
+  bool isRsrcLegal = true;
   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()).operands()[RsrcIdx].RegClass;
-    if (RI.getCommonSubClass(MRI.getRegClass(Rsrc->getReg()),
-                             RI.getRegClass(RsrcRC))) {
-      // The operands are legal.
-      // FIXME: We may need to legalize operands besides srsrc.
-      return CreatedBB;
+    if (Rsrc->isReg() && !RI.isSGPRClass(MRI.getRegClass(Rsrc->getReg()))) {
+      isRsrcLegal = false;
     }
+  }
 
-    // Legalize a VGPR Rsrc.
+  // The operands are legal.
+  if (isRsrcLegal && isSoffsetLegal)
+    return CreatedBB;
+
+  if (!isRsrcLegal) {
+    // Legalize a VGPR Rsrc
     //
     // If the instruction is _ADDR64, we can avoid a waterfall by extracting
     // the base pointer from the VGPR Rsrc, adding it to the VAddr, then using
@@ -6048,6 +6095,7 @@ SIInstrInfo::legalizeOperands(MachineInstr &MI,
     // Otherwise we are on non-ADDR64 hardware, and/or we have
     // idxen/offen/bothen and we fall back to a waterfall loop.
 
+    MachineOperand *Rsrc = &MI.getOperand(RsrcIdx);
     MachineBasicBlock &MBB = *MI.getParent();
 
     MachineOperand *VAddr = getNamedOperand(MI, AMDGPU::OpName::vaddr);
@@ -6157,12 +6205,24 @@ SIInstrInfo::legalizeOperands(MachineInstr &MI,
           .addReg(RsrcPtr, 0, AMDGPU::sub1)
           .addImm(AMDGPU::sub1);
     } else {
-      // This is another variant; legalize Rsrc with waterfall loop from VGPRs
-      // to SGPRs.
-      CreatedBB = loadSRsrcFromVGPR(*this, MI, *Rsrc, MDT);
+      // Legalize a VGPR Rsrc and soffset together.
+      if (!isSoffsetLegal) {
+        MachineOperand *Soffset = getNamedOperand(MI, AMDGPU::OpName::soffset);
+        CreatedBB =
+            loadMBUFScalarOperandsFromVGPR(*this, MI, {Rsrc, Soffset}, MDT);
+        return CreatedBB;
+      }
+      CreatedBB = loadMBUFScalarOperandsFromVGPR(*this, MI, {Rsrc}, MDT);
       return CreatedBB;
     }
   }
+
+  // Legalize a VGPR soffset.
+  if (!isSoffsetLegal) {
+    MachineOperand *Soffset = getNamedOperand(MI, AMDGPU::OpName::soffset);
+    CreatedBB = loadMBUFScalarOperandsFromVGPR(*this, MI, {Soffset}, MDT);
+    return CreatedBB;
+  }
   return CreatedBB;
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.format.f16.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.format.f16.ll
new file mode 100644
index 000000000000..a02f95020ac7
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.format.f16.ll
@@ -0,0 +1,283 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+; Natural mapping
+define half @raw_buffer_load_format_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.buffer.load.format.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret half %val
+}
+
+define <2 x half> @raw_buffer_load_format_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XY_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_XY_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_XY_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call <2 x half> @llvm.amdgcn.raw.buffer.load.format.v2f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <2 x half> %val
+}
+
+define <4 x half> @raw_buffer_load_format_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <4 x half> @llvm.amdgcn.raw.buffer.load.format.v4f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x half> %val
+}
+
+; Waterfall for rsrc and soffset, copy for voffset
+define half @raw_buffer_load_format_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.buffer.load.format.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret half %val
+}
+
+define <4 x half> @raw_buffer_load_format_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add_4095(<4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add_4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %voffset = add i32 %voffset.base, 4095
+  %val = call <4 x half> @llvm.amdgcn.raw.buffer.load.format.v4f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x half> %val
+}
+
+declare half @llvm.amdgcn.raw.buffer.load.format.f16(<4 x i32>, i32, i32, i32 immarg) #0
+declare <2 x half> @llvm.amdgcn.raw.buffer.load.format.v2f16(<4 x i32>, i32, i32, i32 immarg) #0
+declare <3 x half> @llvm.amdgcn.raw.buffer.load.format.v3f16(<4 x i32>, i32, i32, i32 immarg) #0
+declare <4 x half> @llvm.amdgcn.raw.buffer.load.format.v4f16(<4 x i32>, i32, i32, i32 immarg) #0
+
+attributes #0 = { nounwind readonly }
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.format.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.format.ll
new file mode 100644
index 000000000000..df91ca39b0e5
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.format.ll
@@ -0,0 +1,352 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+; Natural mapping
+define float @raw_buffer_load_format_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.format.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define <2 x float> @raw_buffer_load_format_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_FORMAT_XY_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <2 x float> @llvm.amdgcn.raw.buffer.load.format.v2f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <2 x float> %val
+}
+
+define <3 x float> @raw_buffer_load_format_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = BUFFER_LOAD_FORMAT_XYZ_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+  %val = call <3 x float> @llvm.amdgcn.raw.buffer.load.format.v3f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <3 x float> %val
+}
+
+define <4 x float> @raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   $vgpr3 = COPY [[COPY9]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  %val = call <4 x float> @llvm.amdgcn.raw.buffer.load.format.v4f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x float> %val
+}
+
+; Waterfall for rsrc and soffset, copy for voffset
+define float @raw_buffer_load_format_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.format.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define <4 x float> @raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add_4095(<4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_format_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add_4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   $vgpr3 = COPY [[COPY9]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  %voffset = add i32 %voffset.base, 4095
+  %val = call <4 x float> @llvm.amdgcn.raw.buffer.load.format.v4f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x float> %val
+}
+
+declare float @llvm.amdgcn.raw.buffer.load.format.f32(<4 x i32>, i32, i32, i32 immarg) #0
+declare <2 x float> @llvm.amdgcn.raw.buffer.load.format.v2f32(<4 x i32>, i32, i32, i32 immarg) #0
+declare <3 x float> @llvm.amdgcn.raw.buffer.load.format.v3f32(<4 x i32>, i32, i32, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.raw.buffer.load.format.v4f32(<4 x i32>, i32, i32, i32 immarg) #0
+
+attributes #0 = { nounwind readonly }
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.ll
new file mode 100644
index 000000000000..343e9ce1cdff
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.load.ll
@@ -0,0 +1,1925 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+; Natural mapping
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+; Copies for VGPR arguments
+define float @raw_buffer_load_f32__sgpr_rsrc__sgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__sgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+; Waterfall for rsrc
+define float @raw_buffer_load_f32__vgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__vgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+; Waterfall for rsrc and soffset
+define float @raw_buffer_load_f32__vgpr_rsrc__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__vgpr_rsrc__vgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+; Natural mapping + glc
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 1, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 1)
+  ret float %val
+}
+
+; Natural mapping + slc
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 2, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 2)
+  ret float %val
+}
+
+; Natural mapping + dlc
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 4, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 4)
+  ret float %val
+}
+
+; Natural mapping + slc + dlc
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_dlc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 6, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 6)
+  ret float %val
+}
+
+; Natural mapping + glc + dlc
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc_dlc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 5, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 5)
+  ret float %val
+}
+
+; Natural mapping + glc + slc + dlc
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc_slc_dlc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc_slc_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 7, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 7)
+  ret float %val
+}
+
+; Natural mapping
+define <2 x float> @raw_buffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORDX2_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <2 x float> %val
+}
+
+define <3 x float> @raw_buffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORDX3_OFFEN:%[0-9]+]]:vreg_96 = BUFFER_LOAD_DWORDX3_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub2
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+  %val = call <3 x float> @llvm.amdgcn.raw.buffer.load.v3f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <3 x float> %val
+}
+
+define <4 x float> @raw_buffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORDX4_OFFEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub3
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   $vgpr3 = COPY [[COPY9]]
+  ; GFX908-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> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x float> %val
+}
+
+define half @raw_buffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_USHORT_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_LOAD_USHORT_OFFEN]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.buffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret half %val
+}
+
+define <2 x half> @raw_buffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call <2 x half> @llvm.amdgcn.raw.buffer.load.v2f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <2 x half> %val
+}
+
+define <4 x half> @raw_buffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORDX2_OFFEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <4 x half> @llvm.amdgcn.raw.buffer.load.v4f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret <4 x half> %val
+}
+
+define float @raw_buffer_load_i8__sgpr_rsrc__vgpr_voffset__sgpr_soffset_zext(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_i8__sgpr_rsrc__vgpr_voffset__sgpr_soffset_zext
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s8), addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  %zext = zext i8 %val to i32
+  %cast = bitcast i32 %zext to float
+  ret float %cast
+}
+
+define float @raw_buffer_load_i8__sgpr_rsrc__vgpr_voffset__sgpr_soffset_sext(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_i8__sgpr_rsrc__vgpr_voffset__sgpr_soffset_sext
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_SBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_SBYTE_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s8), addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_SBYTE_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  %zext = sext i8 %val to i32
+  %cast = bitcast i32 %zext to float
+  ret float %cast
+}
+
+define float @raw_buffer_load_i16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_zext(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_i16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_zext
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_USHORT_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_USHORT_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call i16 @llvm.amdgcn.raw.buffer.load.i16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  %zext = zext i16 %val to i32
+  %cast = bitcast i32 %zext to float
+  ret float %cast
+}
+
+define float @raw_buffer_load_i16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_sext(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_i16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_sext
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_SSHORT_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_SSHORT_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_SSHORT_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call i16 @llvm.amdgcn.raw.buffer.load.i16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  %sext = sext i16 %val to i32
+  %cast = bitcast i32 %sext to float
+  ret float %cast
+}
+
+; Waterfall for rsrc
+define half @raw_buffer_load_f16__vgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f16__vgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_USHORT_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_LOAD_USHORT_OFFEN]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.buffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret half %val
+}
+
+; Waterfall for rsrc
+define float @raw_buffer_load_i8__vgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_i8__vgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s8), addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  %zext = zext i8 %val to i32
+  %cast = bitcast i32 %zext to float
+  ret float %cast
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vdpr_voffset__sgpr_soffset__voffset0(<4 x i32> %rsrc, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vdpr_voffset__sgpr_soffset__voffset0
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 0, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4095(<4 x i32> %rsrc, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 4095, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4096(<4 x i32> %rsrc, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 4096, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add16(<4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 16, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %voffset = add i32 %voffset.base, 16
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4095(<4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %voffset = add i32 %voffset.base, 4095
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4096(<4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %voffset = add i32 %voffset.base, 4096
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4095(<4 x i32> %rsrc, i32 %voffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 4095, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4096(<4 x i32> %rsrc, i32 %voffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 4096, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 16
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %soffset = add i32 %soffset.base, 16
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4095(<4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %soffset = add i32 %soffset.base, 4095
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4096(<4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %soffset = add i32 %soffset.base, 4096
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+; An add of the offset is necessary, with a waterfall loop. Make sure the add is done outside of the waterfall loop.
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add5000(<4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add5000
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 5000
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %soffset = add i32 %soffset.base, 5000
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+; An add of the offset is necessary, with a waterfall loop. Make sure the add is done outside of the waterfall loop.
+define float @raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add5000(<4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add5000
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 904, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %voffset = add i32 %voffset.base, 5000
+  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret float %val
+}
+
+declare float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32>, i32, i32, i32 immarg)
+declare <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32>, i32, i32, i32 immarg)
+declare <3 x float> @llvm.amdgcn.raw.buffer.load.v3f32(<4 x i32>, i32, i32, i32 immarg)
+declare <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32>, i32, i32, i32 immarg)
+
+declare half @llvm.amdgcn.raw.buffer.load.f16(<4 x i32>, i32, i32, i32 immarg)
+declare <2 x half> @llvm.amdgcn.raw.buffer.load.v2f16(<4 x i32>, i32, i32, i32 immarg)
+declare <3 x half> @llvm.amdgcn.raw.buffer.load.v3f16(<4 x i32>, i32, i32, i32 immarg)
+declare <4 x half> @llvm.amdgcn.raw.buffer.load.v4f16(<4 x i32>, i32, i32, i32 immarg)
+
+declare i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32>, i32, i32, i32 immarg)
+declare i16 @llvm.amdgcn.raw.buffer.load.i16(<4 x i32>, i32, i32, i32 immarg)
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.format.f16.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.format.f16.ll
new file mode 100644
index 000000000000..8cdfb6467fa3
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.format.f16.ll
@@ -0,0 +1,606 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f16(<4 x i32> %rsrc, half %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__voffset_4095__sgpr_soffset_f16(<4 x i32> %rsrc, half %val, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__voffset_4095__sgpr_soffset_f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_X_OFFSET_exact [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.f16(half %val, <4 x i32> %rsrc, i32 4095, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16(<4 x i32> %rsrc, <4 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XYZW_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v4f16(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+; Make sure unpack code is emitted outside of loop
+define void @raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16(<4 x i32> %rsrc, <4 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XYZW_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v4f16(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4095(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY1]], [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 4095, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4096(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY1]], [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 4096, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_voffset_add_16(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_voffset_add_16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 16, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 16
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_voffset_add_4095(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_voffset_add_4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 4095
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_voffset_add_4096(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_voffset_add_4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY2]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 4096
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+
+; Check what happens with offset add inside a waterfall loop
+define void @raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16_add_4096(<4 x i32> %rsrc, <4 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16_add_4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_D16_XYZW_OFFEN_exact [[COPY8]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 4096
+  call void @llvm.amdgcn.raw.buffer.store.format.v4f16(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+declare void @llvm.amdgcn.raw.buffer.store.format.f16(half, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.format.v2f16(<2 x half>, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.format.v4f16(<4 x half>, <4 x i32>, i32, i32, i32 immarg)
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.format.f32.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.format.f32.ll
new file mode 100644
index 000000000000..34fdbed11f13
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.format.f32.ll
@@ -0,0 +1,708 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__voffset_4095__sgpr_soffset_f32(<4 x i32> %rsrc, float %val, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__voffset_4095__sgpr_soffset_f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.f32(float %val, <4 x i32> %rsrc, i32 4095, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v3f32(<4 x i32> %rsrc, <3 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v3f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY5]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XYZ_OFFEN_exact [[COPY9]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v3f32(<3 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32(<4 x i32> %rsrc, <4 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr9
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY8]], %subreg.sub1, [[COPY7]], %subreg.sub2, [[COPY6]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF8:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XYZW_OFFEN_exact [[COPY10]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v4f32(<4 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32(<4 x i32> %rsrc, <4 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr9
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY8]], %subreg.sub1, [[COPY7]], %subreg.sub2, [[COPY6]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF8:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XYZW_OFFEN_exact [[COPY10]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v4f32(<4 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_soffset4095(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_soffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY7]], [[COPY]], killed [[REG_SEQUENCE4]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 4095, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_soffset4096(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_soffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY7]], [[COPY]], killed [[REG_SEQUENCE4]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 4096, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_voffset_add_16(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_voffset_add_16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 16, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 16
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_voffset_add_4095(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_voffset_add_4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 4095
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_voffset_add_4096(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32_voffset_add_4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY8]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 4096
+  call void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+
+; Check what happens with offset add inside a waterfall loop
+define void @raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32_add_4096(<4 x i32> %rsrc, <4 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store_format__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32_add_4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr9
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY8]], %subreg.sub1, [[COPY7]], %subreg.sub2, [[COPY6]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF8:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_FORMAT_XYZW_OFFEN_exact [[COPY10]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset.add = add i32 %voffset, 4096
+  call void @llvm.amdgcn.raw.buffer.store.format.v4f32(<4 x float> %val, <4 x i32> %rsrc, i32 %voffset.add, i32 %soffset, i32 0)
+  ret void
+}
+
+declare void @llvm.amdgcn.raw.buffer.store.format.f32(float, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.format.v2f32(<2 x float>, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.format.v3f32(<3 x float>, <4 x i32>, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.raw.buffer.store.format.v4f32(<4 x float>, <4 x i32>, i32, i32, i32 immarg)
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.ll
new file mode 100644
index 000000000000..bbfc03a41ece
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.buffer.store.ll
@@ -0,0 +1,1808 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+; Natural mapping
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__sgpr_val__sgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 1, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 1)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 2, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 2)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 3, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 3)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 4, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 4)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc_dlc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_slc_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 6, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 6)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_dlc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 5, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 5)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc_dlc(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_glc_slc_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 7, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 7)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32(<4 x i32> %rsrc, <2 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX2_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v3f32(<4 x i32> %rsrc, <3 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v3f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY5]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX3_OFFEN_exact [[COPY9]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32(<4 x i32> %rsrc, <4 x float> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr9
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY8]], %subreg.sub1, [[COPY7]], %subreg.sub2, [[COPY6]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF8:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX4_OFFEN_exact [[COPY10]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i8(<4 x i32> %rsrc, i32 %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i8
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_BYTE_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s8), addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i16(<4 x i32> %rsrc, i32 %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_i16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_SHORT_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f16(<4 x i32> %rsrc, half %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY2]]
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_SHORT_OFFEN_exact [[COPY7]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16(<4 x i32> %rsrc, <4 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX2_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16(<4 x i32> %rsrc, <4 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v4f16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1, [[COPY5]], %subreg.sub2, [[COPY4]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX2_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4095(<4 x i32> %rsrc, float %val, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFSET_exact [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 4095, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4096(<4 x i32> %rsrc, float %val, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__sgpr_soffset_f32_voffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY1]], [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.buffer.store.f32(float %val, <4 x i32> %rsrc, i32 4096, i32 %soffset, i32 0)
+  ret void
+}
+
+define void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_16(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 16, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4095(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4096(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_f32_voffset_add_4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4095(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY1]], [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4096(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY1]], [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_16(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 16, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4095(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4096(<4 x i32> %rsrc, <2 x half> %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_v2f16_soffset_add_4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_offset_add_5000(<4 x i32> %rsrc, float %val, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__vgpr_voffset__sgpr_soffset_offset_add_5000
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY2]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 904, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %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 void @raw_buffer_store__vgpr_rsrc__vgpr_val__5000_voffset__sgpr_soffset_offset(<4 x i32> %rsrc, float %val, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_buffer_store__vgpr_rsrc__vgpr_val__5000_voffset__sgpr_soffset_offset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORD_OFFEN_exact [[COPY1]], [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 904, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  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)
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.load.f16.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.load.f16.ll
new file mode 100644
index 000000000000..0e485a523e48
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.load.f16.ll
@@ -0,0 +1,436 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+define half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret half %val
+}
+
+define <2 x half> @raw_tbuffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_XY_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_XY_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_XY_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call <2 x half> @llvm.amdgcn.raw.tbuffer.load.v2f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <2 x half> %val
+}
+
+define <4 x half> @raw_tbuffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <4 x half> @llvm.amdgcn.raw.tbuffer.load.v4f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <4 x half> %val
+}
+
+define half @raw_tbuffer_load_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret half %val
+}
+
+define half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 1, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
+  ret half %val
+}
+
+define half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 2, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
+  ret half %val
+}
+
+define half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 3, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
+  ret half %val
+}
+
+define half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 4, 0, implicit $exec :: (dereferenceable load (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
+  ret half %val
+}
+
+declare half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <2 x half> @llvm.amdgcn.raw.tbuffer.load.v2f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <3 x half> @llvm.amdgcn.raw.tbuffer.load.v3f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <4 x half> @llvm.amdgcn.raw.tbuffer.load.v4f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+
+attributes #0 = { nounwind readonly }
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.load.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.load.ll
new file mode 100644
index 000000000000..1aa216b41cb9
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.load.ll
@@ -0,0 +1,501 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+define float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret float %val
+}
+
+define <2 x float> @raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1
+  %val = call <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <2 x float> %val
+}
+
+define <3 x float> @raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+  %val = call <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <3 x float> %val
+}
+
+define <4 x float> @raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY6]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY7]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY8]]
+  ; GFX908-NEXT:   $vgpr3 = COPY [[COPY9]]
+  ; GFX908-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> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <4 x float> %val
+}
+
+define float @raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret float %val
+}
+
+define float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 1, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
+  ret float %val
+}
+
+define float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 2, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
+  ret float %val
+}
+
+define float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 3, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
+  ret float %val
+}
+
+define float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 4, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-NEXT:   SI_RETURN implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
+  ret float %val
+}
+
+declare float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+
+attributes #0 = { nounwind readonly }
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.f16.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.f16.ll
new file mode 100644
index 000000000000..8e6ef12c814b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.f16.ll
@@ -0,0 +1,545 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+define void @raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_XY_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_XYZW_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v4f16(<4 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc
+define void @raw_tbuffer_store_f16__vgpr_rsrc__vgpr_voffset__sgpr_soffset(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__vgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc and soffset
+define void @raw_tbuffer_store_f16__vgpr_rsrc__vgpr_voffset__vgpr_soffset(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__vgpr_rsrc__vgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc and soffset, copy for voffset
+define void @raw_tbuffer_store_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 1, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
+  ret void
+}
+
+define void @raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 2, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
+  ret void
+}
+
+define void @raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 3, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
+  ret void
+}
+
+define void @raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_D16_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 4, 0, implicit $exec :: (dereferenceable store (s16), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f16(half %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
+  ret void
+}
+
+declare void @llvm.amdgcn.raw.tbuffer.store.f16(half, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
+declare void @llvm.amdgcn.raw.tbuffer.store.v2f16(<2 x half>, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
+declare void @llvm.amdgcn.raw.tbuffer.store.v3f16(<3 x half>, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
+declare void @llvm.amdgcn.raw.tbuffer.store.v4f16(<4 x half>, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.ll b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.ll
new file mode 100644
index 000000000000..0753ddea5133
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.ll
@@ -0,0 +1,1377 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+; Natural mapping
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Natural mapping
+define void @raw_tbuffer_store_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v2f32(<2 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Natural mapping
+define void @raw_tbuffer_store_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<3 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1, [[COPY6]], %subreg.sub2
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_XYZ_OFFEN_exact [[COPY9]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v3f32(<3 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Natural mapping
+define void @raw_tbuffer_store_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr9
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr8
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr7
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY8]], %subreg.sub1, [[COPY7]], %subreg.sub2, [[COPY6]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE1]]
+  ; GFX908-NEXT:   [[DEF8:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE3]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_XYZW_OFFEN_exact [[COPY10]], [[COPY1]], killed [[REG_SEQUENCE4]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Copies for VGPR arguments
+define void @raw_tbuffer_store_f32__sgpr_rsrc__sgpr_voffset__sgpr_soffset(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__sgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc
+define void @raw_tbuffer_store_f32__vgpr_rsrc__vgpr_voffset__sgpr_soffset(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__vgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 1, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 1)
+  ret void
+}
+
+; Waterfall for rsrc and soffset
+define void @raw_tbuffer_store_f32__vgpr_rsrc__vgpr_voffset__vgpr_soffset(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__vgpr_rsrc__vgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+; Waterfall for rsrc and soffset, copy for voffset
+define void @raw_tbuffer_store_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret void
+}
+
+; Natural mapping + glc
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 1, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
+  ret void
+}
+
+; Natural mapping + slc
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 2, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
+  ret void
+}
+
+; Natural mapping + glc + slc
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 3, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
+  ret void
+}
+
+; Natural mapping + dlc
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 78, 4, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vdpr_voffset__sgpr_soffset__voffset0(float %val, <4 x i32> %rsrc, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vdpr_voffset__sgpr_soffset__voffset0
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 0, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4095(float %val, <4 x i32> %rsrc, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFSET_exact [[COPY5]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 4095, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4096(float %val, <4 x i32> %rsrc, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY5]], [[V_MOV_B32_e32_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 4096, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add16(float %val, <4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 16, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset = add i32 %voffset.base, 16
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4095(float %val, <4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 4095, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset = add i32 %voffset.base, 4095
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4096(float %val, <4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset__voffset_add4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset = add i32 %voffset.base, 4096
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4095(float %val, <4 x i32> %rsrc, i32 %voffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY5]], [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 4095, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4096(float %val, <4 x i32> %rsrc, i32 %voffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY5]], [[COPY]], killed [[REG_SEQUENCE3]], [[S_MOV_B32_]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 4096, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add16(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add16
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 16
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %soffset = add i32 %soffset.base, 16
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4095(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4095
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %soffset = add i32 %soffset.base, 4095
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+define void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4096(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add4096
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %soffset = add i32 %soffset.base, 4096
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, 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 void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add5000(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset.base) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_soffset_add5000
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 5000
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[V_ADD_U32_e64_]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 0, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %soffset = add i32 %soffset.base, 5000
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, 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 void @raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add5000(float %val, <4 x i32> %rsrc, i32 %voffset.base, i32 %soffset) {
+  ; GFX908-LABEL: name: raw_tbuffer_store_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_voffset_add5000
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr6
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
+  ; GFX908-NEXT:   [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY1]], killed [[S_MOV_B32_]], 0, implicit $exec
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub0, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub1, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[REG_SEQUENCE]].sub0_sub1, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub2, implicit $exec
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE]].sub3, implicit $exec
+  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[REG_SEQUENCE]].sub2_sub3, implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_]], [[V_CMP_EQ_U64_e64_1]], implicit-def $scc
+  ; GFX908-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_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
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def $scc
+  ; GFX908-NEXT:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_X_OFFEN_exact [[COPY6]], [[V_ADD_U32_e64_]], killed [[REG_SEQUENCE3]], killed [[V_READFIRSTLANE_B32_4]], 904, 94, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   SI_RETURN
+  %voffset = add i32 %voffset.base, 5000
+  call void @llvm.amdgcn.raw.tbuffer.store.f32(float %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 94, i32 0)
+  ret void
+}
+
+declare void @llvm.amdgcn.raw.tbuffer.store.f32(float, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
+declare void @llvm.amdgcn.raw.tbuffer.store.v2f32(<2 x float>, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
+declare void @llvm.amdgcn.raw.tbuffer.store.v3f32(<3 x float>, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
+declare void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32 immarg, i32 immarg)
diff --git a/llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll b/llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll
new file mode 100644
index 000000000000..1f5072483611
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll
@@ -0,0 +1,635 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -march=amdgcn -mcpu=gfx908 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck -check-prefix=GFX908 %s
+
+define float @llvm_amdgcn_raw_buffer_load_f32(i32 %voffset, i32 %soffset) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_load_f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_load_f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_load_v2f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_load_v2f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_load_v3f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_OFFEN]].sub2
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_load_v3f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_load_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_OFFEN]].sub3
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; GFX908-NEXT:   $vgpr3 = COPY [[COPY5]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_load_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
+  ; GFX908-NEXT:   $vgpr0 = COPY [[COPY2]]
+  ; GFX908-NEXT:   $vgpr1 = COPY [[COPY3]]
+  ; GFX908-NEXT:   $vgpr2 = COPY [[COPY4]]
+  ; GFX908-NEXT:   $vgpr3 = COPY [[COPY5]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_store_f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_store_f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-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)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_store_v2f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX2_OFFEN_exact [[COPY4]], [[COPY1]], [[DEF2]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_store_v2f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_XY_OFFEN_exact [[COPY4]], [[COPY1]], [[DEF2]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s64), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_store_v3f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE]]
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX3_OFFEN_exact [[COPY5]], [[COPY1]], [[DEF3]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_store_v3f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_96 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vreg_96 = COPY [[REG_SEQUENCE]]
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_XYZ_OFFEN_exact [[COPY5]], [[COPY1]], [[DEF3]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s96), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_buffer_store_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE]]
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   BUFFER_STORE_DWORDX4_OFFEN_exact [[COPY6]], [[COPY1]], [[DEF4]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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) {
+  ; GFX908-LABEL: name: llvm_amdgcn_raw_tbuffer_store_v4f32
+  ; GFX908: bb.0 (%ir-block.0):
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr5
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; GFX908-NEXT:   [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
+  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vreg_128 = COPY [[REG_SEQUENCE]]
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[COPY]], implicit $exec
+  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_]], [[COPY]], implicit $exec
+  ; GFX908-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
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.2:
+  ; GFX908-NEXT:   successors: %bb.1(0x40000000), %bb.3(0x40000000)
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   TBUFFER_STORE_FORMAT_XYZW_OFFEN_exact [[COPY6]], [[COPY1]], [[DEF4]], killed [[V_READFIRSTLANE_B32_]], 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s128), align 1, addrspace 7)
+  ; GFX908-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GFX908-NEXT:   SI_WATERFALL_LOOP %bb.1, implicit $exec
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.3:
+  ; GFX908-NEXT:   $exec = S_MOV_B64 [[S_MOV_B64_]]
+  ; GFX908-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)