llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.atomic.dim.mir

   1 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
   2 # RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -run-pass=instruction-select -mcpu=tahiti -o - %s | FileCheck -check-prefix=GFX6 %s
   3 # RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -run-pass=instruction-select -mcpu=fiji -o - %s | FileCheck -check-prefix=GFX8 %s
   4 # RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -run-pass=instruction-select -mcpu=gfx1010 -o - %s | FileCheck -check-prefix=GFX10 %s
   5 # RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -run-pass=instruction-select -mcpu=gfx1100 -o - %s | FileCheck -check-prefix=GFX11 %s
   6 # RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -run-pass=instruction-select -mcpu=gfx1200 -o - %s | FileCheck -check-prefix=GFX12 %s
   7
   8 ---
   9 name: atomic_cmpswap_i32_1d
  10 legalized: true
  11 regBankSelected: true
  12 tracksRegLiveness: true
  13
  14 body: |
  15   bb.0:
  16     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  17
  18     ; GFX6-LABEL: name: atomic_cmpswap_i32_1d
  19     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  20     ; GFX6-NEXT: {{  $}}
  21     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  22     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  23     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  24     ; GFX6-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_si:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_si [[COPY1]], [[COPY2]], [[COPY]], 3, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
  25     ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_si]].sub0
  26     ; GFX6-NEXT: $vgpr0 = COPY [[COPY3]]
  27     ; GFX6-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  28     ; GFX8-LABEL: name: atomic_cmpswap_i32_1d
  29     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  30     ; GFX8-NEXT: {{  $}}
  31     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  32     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  33     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  34     ; GFX8-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_vi:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_vi [[COPY1]], [[COPY2]], [[COPY]], 3, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
  35     ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_vi]].sub0
  36     ; GFX8-NEXT: $vgpr0 = COPY [[COPY3]]
  37     ; GFX8-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  38     ; GFX10-LABEL: name: atomic_cmpswap_i32_1d
  39     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  40     ; GFX10-NEXT: {{  $}}
  41     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  42     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  43     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  44     ; GFX10-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10_:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10 [[COPY1]], [[COPY2]], [[COPY]], 3, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
  45     ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10_]].sub0
  46     ; GFX10-NEXT: $vgpr0 = COPY [[COPY3]]
  47     ; GFX10-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  48     ; GFX11-LABEL: name: atomic_cmpswap_i32_1d
  49     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  50     ; GFX11-NEXT: {{  $}}
  51     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  52     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  53     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  54     ; GFX11-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx11_:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx11 [[COPY1]], [[COPY2]], [[COPY]], 3, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
  55     ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx11_]].sub0
  56     ; GFX11-NEXT: $vgpr0 = COPY [[COPY3]]
  57     ; GFX11-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  58     ; GFX12-LABEL: name: atomic_cmpswap_i32_1d
  59     ; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  60     ; GFX12-NEXT: {{  $}}
  61     ; GFX12-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  62     ; GFX12-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  63     ; GFX12-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  64     ; GFX12-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx12_:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx12 [[COPY1]], [[COPY2]], [[COPY]], 3, 0, 1, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
  65     ; GFX12-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx12_]].sub0
  66     ; GFX12-NEXT: $vgpr0 = COPY [[COPY3]]
  67     ; GFX12-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  68     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  69     %1:vgpr(<2 x s32>) = COPY $vgpr0_vgpr1
  70     %2:vgpr(s32) = COPY $vgpr2
  71     %3:vgpr(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), %1(<2 x s32>), $noreg, %2(s32), %0(<8 x s32>), 0, 0, 0 :: (volatile dereferenceable load store (s32), addrspace 8)
  72     $vgpr0 = COPY %3(s32)
  73     SI_RETURN_TO_EPILOG implicit $vgpr0
  74 ...
  75
  76 ---
  77 name: atomic_cmpswap_i32_1d_no_return
  78 legalized: true
  79 regBankSelected: true
  80 tracksRegLiveness: true
  81
  82 body: |
  83   bb.0:
  84     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  85
  86     ; GFX6-LABEL: name: atomic_cmpswap_i32_1d_no_return
  87     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  88     ; GFX6-NEXT: {{  $}}
  89     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  90     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  91     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  92     ; GFX6-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_si:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_si [[COPY1]], [[COPY2]], [[COPY]], 3, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
  93     ; GFX6-NEXT: S_ENDPGM 0
  94     ; GFX8-LABEL: name: atomic_cmpswap_i32_1d_no_return
  95     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  96     ; GFX8-NEXT: {{  $}}
  97     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  98     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  99     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
 100     ; GFX8-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_vi:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_vi [[COPY1]], [[COPY2]], [[COPY]], 3, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
 101     ; GFX8-NEXT: S_ENDPGM 0
 102     ; GFX10-LABEL: name: atomic_cmpswap_i32_1d_no_return
 103     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
 104     ; GFX10-NEXT: {{  $}}
 105     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 106     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
 107     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
 108     ; GFX10-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10_:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10 [[COPY1]], [[COPY2]], [[COPY]], 3, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
 109     ; GFX10-NEXT: S_ENDPGM 0
 110     ; GFX11-LABEL: name: atomic_cmpswap_i32_1d_no_return
 111     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
 112     ; GFX11-NEXT: {{  $}}
 113     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 114     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
 115     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
 116     ; GFX11-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx11_:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx11 [[COPY1]], [[COPY2]], [[COPY]], 3, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
 117     ; GFX11-NEXT: S_ENDPGM 0
 118     ; GFX12-LABEL: name: atomic_cmpswap_i32_1d_no_return
 119     ; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
 120     ; GFX12-NEXT: {{  $}}
 121     ; GFX12-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 122     ; GFX12-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
 123     ; GFX12-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
 124     ; GFX12-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx12_:%[0-9]+]]:vreg_64 = IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx12 [[COPY1]], [[COPY2]], [[COPY]], 3, 0, 1, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), addrspace 8)
 125     ; GFX12-NEXT: S_ENDPGM 0
 126     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 127     %1:vgpr(<2 x s32>) = COPY $vgpr0_vgpr1
 128     %2:vgpr(s32) = COPY $vgpr2
 129     %3:vgpr(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), %1(<2 x s32>), $noreg, %2(s32), %0(<8 x s32>), 0, 0, 0 :: (volatile dereferenceable load store (s32), addrspace 8)
 130     S_ENDPGM 0
 131 ...
 132
 133 ---
 134 name: atomic_cmpswap_i64_1d
 135 legalized: true
 136 regBankSelected: true
 137 tracksRegLiveness: true
 138
 139 body: |
 140   bb.0:
 141     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 142
 143     ; GFX6-LABEL: name: atomic_cmpswap_i64_1d
 144     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 145     ; GFX6-NEXT: {{  $}}
 146     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 147     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 148     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 149     ; GFX6-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_si:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_si [[COPY1]], [[COPY2]], [[COPY]], 15, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 150     ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_si]].sub0_sub1
 151     ; GFX6-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 152     ; GFX6-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 153     ; GFX8-LABEL: name: atomic_cmpswap_i64_1d
 154     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 155     ; GFX8-NEXT: {{  $}}
 156     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 157     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 158     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 159     ; GFX8-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_vi:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_vi [[COPY1]], [[COPY2]], [[COPY]], 15, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 160     ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_vi]].sub0_sub1
 161     ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 162     ; GFX8-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 163     ; GFX10-LABEL: name: atomic_cmpswap_i64_1d
 164     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 165     ; GFX10-NEXT: {{  $}}
 166     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 167     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 168     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 169     ; GFX10-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10_:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10 [[COPY1]], [[COPY2]], [[COPY]], 15, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 170     ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10_]].sub0_sub1
 171     ; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 172     ; GFX10-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 173     ; GFX11-LABEL: name: atomic_cmpswap_i64_1d
 174     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 175     ; GFX11-NEXT: {{  $}}
 176     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 177     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 178     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 179     ; GFX11-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx11_:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx11 [[COPY1]], [[COPY2]], [[COPY]], 15, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 180     ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx11_]].sub0_sub1
 181     ; GFX11-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 182     ; GFX11-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 183     ; GFX12-LABEL: name: atomic_cmpswap_i64_1d
 184     ; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 185     ; GFX12-NEXT: {{  $}}
 186     ; GFX12-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 187     ; GFX12-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 188     ; GFX12-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 189     ; GFX12-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx12_:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx12 [[COPY1]], [[COPY2]], [[COPY]], 15, 0, 1, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 190     ; GFX12-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx12_]].sub0_sub1
 191     ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 192     ; GFX12-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 193     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 194     %1:vgpr(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 195     %2:vgpr(s32) = COPY $vgpr4
 196     %3:vgpr(s64) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), %1(<2 x s64>), $noreg, %2(s32), %0(<8 x s32>), 0, 0, 0 :: (volatile dereferenceable load store (s64), addrspace 8)
 197     $vgpr0_vgpr1 = COPY %3(s64)
 198     SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 199 ...
 200
 201 ---
 202 name: atomic_cmpswap_i64_1d_no_return
 203 legalized: true
 204 regBankSelected: true
 205 tracksRegLiveness: true
 206
 207 body: |
 208   bb.0:
 209     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 210
 211     ; GFX6-LABEL: name: atomic_cmpswap_i64_1d_no_return
 212     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 213     ; GFX6-NEXT: {{  $}}
 214     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 215     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 216     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 217     ; GFX6-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_si:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_si [[COPY1]], [[COPY2]], [[COPY]], 15, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 218     ; GFX6-NEXT: S_ENDPGM 0
 219     ; GFX8-LABEL: name: atomic_cmpswap_i64_1d_no_return
 220     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 221     ; GFX8-NEXT: {{  $}}
 222     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 223     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 224     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 225     ; GFX8-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_vi:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_vi [[COPY1]], [[COPY2]], [[COPY]], 15, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 226     ; GFX8-NEXT: S_ENDPGM 0
 227     ; GFX10-LABEL: name: atomic_cmpswap_i64_1d_no_return
 228     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 229     ; GFX10-NEXT: {{  $}}
 230     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 231     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 232     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 233     ; GFX10-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10_:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10 [[COPY1]], [[COPY2]], [[COPY]], 15, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 234     ; GFX10-NEXT: S_ENDPGM 0
 235     ; GFX11-LABEL: name: atomic_cmpswap_i64_1d_no_return
 236     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 237     ; GFX11-NEXT: {{  $}}
 238     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 239     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 240     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 241     ; GFX11-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx11_:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx11 [[COPY1]], [[COPY2]], [[COPY]], 15, 0, 1, 1, 0, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 242     ; GFX11-NEXT: S_ENDPGM 0
 243     ; GFX12-LABEL: name: atomic_cmpswap_i64_1d_no_return
 244     ; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 245     ; GFX12-NEXT: {{  $}}
 246     ; GFX12-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 247     ; GFX12-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 248     ; GFX12-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 249     ; GFX12-NEXT: [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx12_:%[0-9]+]]:vreg_128 = IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx12 [[COPY1]], [[COPY2]], [[COPY]], 15, 0, 1, 0, 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), addrspace 8)
 250     ; GFX12-NEXT: S_ENDPGM 0
 251     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 252     %1:vgpr(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 253     %2:vgpr(s32) = COPY $vgpr4
 254     %3:vgpr(s64) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), %1(<2 x s64>), $noreg, %2(s32), %0(<8 x s32>), 0, 0, 0 :: (volatile dereferenceable load store (s64), addrspace 8)
 255     S_ENDPGM 0
 256 ...