xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.struct.ptr.buffer.atomic.fadd.ll (revision c3cfbbc4160c3e0284034c98b332b468328458e3)

; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx908 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s -check-prefix=GFX908
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx90a -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s -check-prefix=GFX90A

; Natural mapping
define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX90A-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
  ret void
}

define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_plus4095__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_plus4095__sgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 4095, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_plus4095__sgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX90A-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 4095, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %voffset.add = add i32 %voffset, 4095
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset.add, i32 %soffset, i32 0)
  ret void
}

define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__4095_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__4095_voffset__sgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 4095, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__4095_voffset__sgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 4095, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 4095, i32 %soffset, i32 0)
  ret void
}

; Natural mapping, no voffset
define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
  ret void
}

; All register operands need legalization
define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__sgpr_val__vgpr_rsrc__sgpr_voffset__vgpr_soffset(float inreg %val, ptr addrspace(8) %rsrc, i32 inreg %vindex, i32 inreg %voffset, i32 %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__sgpr_val__vgpr_rsrc__sgpr_voffset__vgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr4
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY]]
  ; GFX908-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY6]]
  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT: bb.2:
  ; GFX908-NEXT:   successors: %bb.3(0x80000000)
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY1]], implicit $exec
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY4]], implicit $exec
  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[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:   [[COPY11:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
  ; GFX908-NEXT:   [[COPY12:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
  ; GFX908-NEXT:   [[COPY13:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
  ; GFX908-NEXT:   [[COPY14:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY13]], [[COPY11]], implicit $exec
  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY14]], [[COPY12]], 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 dead $scc
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY7]], implicit $exec
  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY7]], 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 dead $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.3:
  ; GFX908-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY10]], %subreg.sub1
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY8]], [[REG_SEQUENCE2]], [[REG_SEQUENCE1]], [[V_READFIRSTLANE_B32_4]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, 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.2, implicit $exec
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT: bb.4:
  ; GFX908-NEXT:   successors: %bb.5(0x80000000)
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT: bb.5:
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__sgpr_val__vgpr_rsrc__sgpr_voffset__vgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   successors: %bb.2(0x80000000)
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr4
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY]]
  ; GFX90A-NEXT:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
  ; GFX90A-NEXT:   [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[COPY6]]
  ; GFX90A-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.2:
  ; GFX90A-NEXT:   successors: %bb.3(0x80000000)
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY1]], implicit $exec
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY4]], implicit $exec
  ; GFX90A-NEXT:   [[REG_SEQUENCE1:%[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
  ; GFX90A-NEXT:   [[COPY11:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]].sub0_sub1
  ; GFX90A-NEXT:   [[COPY12:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]].sub2_sub3
  ; GFX90A-NEXT:   [[COPY13:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
  ; GFX90A-NEXT:   [[COPY14:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
  ; GFX90A-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY13]], [[COPY11]], implicit $exec
  ; GFX90A-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY14]], [[COPY12]], implicit $exec
  ; GFX90A-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 dead $scc
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY7]], implicit $exec
  ; GFX90A-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY7]], implicit $exec
  ; GFX90A-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def dead $scc
  ; GFX90A-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
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.3:
  ; GFX90A-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY9]], %subreg.sub0, [[COPY10]], %subreg.sub1
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY8]], [[REG_SEQUENCE2]], [[REG_SEQUENCE1]], [[V_READFIRSTLANE_B32_4]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
  ; GFX90A-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.4:
  ; GFX90A-NEXT:   successors: %bb.5(0x80000000)
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.5:
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
  ret void
}

; All register operands need legalization, no voffset
define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__sgpr_val__vgpr_rsrc__0_voffset__vgpr_soffset(float inreg %val, ptr addrspace(8) %rsrc, i32 inreg %vindex, i32 %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__sgpr_val__vgpr_rsrc__0_voffset__vgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   successors: %bb.2(0x80000000)
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY]]
  ; GFX908-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
  ; GFX908-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT: bb.2:
  ; GFX908-NEXT:   successors: %bb.3(0x80000000)
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY1]], implicit $exec
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY4]], implicit $exec
  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[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:   [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
  ; GFX908-NEXT:   [[COPY10:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
  ; GFX908-NEXT:   [[COPY11:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
  ; GFX908-NEXT:   [[COPY12:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY11]], [[COPY9]], implicit $exec
  ; GFX908-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY12]], [[COPY10]], 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 dead $scc
  ; GFX908-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
  ; GFX908-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], 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 dead $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.3:
  ; GFX908-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY7]], [[COPY8]], [[REG_SEQUENCE1]], [[V_READFIRSTLANE_B32_4]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, 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.2, implicit $exec
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT: bb.4:
  ; GFX908-NEXT:   successors: %bb.5(0x80000000)
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT: bb.5:
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__sgpr_val__vgpr_rsrc__0_voffset__vgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   successors: %bb.2(0x80000000)
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr3
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY]]
  ; GFX90A-NEXT:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]]
  ; GFX90A-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.2:
  ; GFX90A-NEXT:   successors: %bb.3(0x80000000)
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY1]], implicit $exec
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY2]], implicit $exec
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY3]], implicit $exec
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY4]], implicit $exec
  ; GFX90A-NEXT:   [[REG_SEQUENCE1:%[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
  ; GFX90A-NEXT:   [[COPY9:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]].sub0_sub1
  ; GFX90A-NEXT:   [[COPY10:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]].sub2_sub3
  ; GFX90A-NEXT:   [[COPY11:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub0_sub1
  ; GFX90A-NEXT:   [[COPY12:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE1]].sub2_sub3
  ; GFX90A-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY11]], [[COPY9]], implicit $exec
  ; GFX90A-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[COPY12]], [[COPY10]], implicit $exec
  ; GFX90A-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 dead $scc
  ; GFX90A-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec
  ; GFX90A-NEXT:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec
  ; GFX90A-NEXT:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[S_AND_B64_]], [[V_CMP_EQ_U32_e64_]], implicit-def dead $scc
  ; GFX90A-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
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.3:
  ; GFX90A-NEXT:   successors: %bb.4(0x40000000), %bb.2(0x40000000)
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY7]], [[COPY8]], [[REG_SEQUENCE1]], [[V_READFIRSTLANE_B32_4]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
  ; GFX90A-NEXT:   SI_WATERFALL_LOOP %bb.2, implicit $exec
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.4:
  ; GFX90A-NEXT:   successors: %bb.5(0x80000000)
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   $exec = S_MOV_B64_term [[S_MOV_B64_]]
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT: bb.5:
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
  ret void
}

; Natural mapping + slc
define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX90A-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
  ret void
}

define amdgpu_ps void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset_slc(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset_slc
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset_slc
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   BUFFER_ATOMIC_ADD_F32_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 2)
  ret void
}

define amdgpu_ps void @struct_ptr_buffer_atomic_add_v2f16_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<2 x half> %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_v2f16_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX908-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX908-NEXT:   BUFFER_ATOMIC_PK_ADD_F16_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 0, 0, implicit $exec :: (volatile dereferenceable load store (<2 x s16>) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_v2f16_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1, $vgpr2
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  ; GFX90A-NEXT:   [[COPY7:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY6]], %subreg.sub1
  ; GFX90A-NEXT:   BUFFER_ATOMIC_PK_ADD_F16_BOTHEN [[COPY]], [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY7]], 0, 0, implicit $exec :: (volatile dereferenceable load store (<2 x s16>) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call <2 x half> @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
  ret void
}

define amdgpu_ps void @struct_ptr_buffer_atomic_add_v2f16_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset(<2 x half> %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
  ; GFX908-LABEL: name: struct_ptr_buffer_atomic_add_v2f16_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset
  ; GFX908: bb.1 (%ir-block.0):
  ; GFX908-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX908-NEXT: {{  $}}
  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX908-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX908-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX908-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX908-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX908-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX908-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX908-NEXT:   BUFFER_ATOMIC_PK_ADD_F16_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, implicit $exec :: (volatile dereferenceable load store (<2 x s16>) on %ir.rsrc, align 1, addrspace 8)
  ; GFX908-NEXT:   S_ENDPGM 0
  ;
  ; GFX90A-LABEL: name: struct_ptr_buffer_atomic_add_v2f16_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset
  ; GFX90A: bb.1 (%ir-block.0):
  ; GFX90A-NEXT:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1
  ; GFX90A-NEXT: {{  $}}
  ; GFX90A-NEXT:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
  ; GFX90A-NEXT:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr2
  ; GFX90A-NEXT:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr3
  ; GFX90A-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr4
  ; GFX90A-NEXT:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr5
  ; GFX90A-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
  ; GFX90A-NEXT:   [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6
  ; GFX90A-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY4]], %subreg.sub3
  ; GFX90A-NEXT:   BUFFER_ATOMIC_PK_ADD_F16_IDXEN [[COPY]], [[COPY5]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, implicit $exec :: (volatile dereferenceable load store (<2 x s16>) on %ir.rsrc, align 1, addrspace 8)
  ; GFX90A-NEXT:   S_ENDPGM 0
  %ret = call <2 x half> @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
  ret void
}

declare float @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f32(float, ptr addrspace(8), i32, i32, i32, i32 immarg) #0
declare <2 x half> @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.v2f16(<2 x half>, ptr addrspace(8), i32, i32, i32, i32 immarg) #0

attributes #0 = { nounwind }