test/CodeGen/AMDGPU/legalize-amdgcn.raw.tbuffer.store.f16.ll

; 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 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.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 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.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 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.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 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.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 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.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 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.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 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.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 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.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 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.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 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.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)