test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll - llvm-project/llvm - Git at Google

 ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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 8)
   ; 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)