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
   7 ---
   8 name: atomic_cmpswap_i32_1d
   9 legalized: true
  10 regBankSelected: true
  11 tracksRegLiveness: true
  12
  13 body: |
  14   bb.0:
  15     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  16
  17     ; GFX6-LABEL: name: atomic_cmpswap_i32_1d
  18     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  19     ; GFX6-NEXT: {{  $}}
  20     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  21     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  22     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  23     ; 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)
  24     ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_si]].sub0
  25     ; GFX6-NEXT: $vgpr0 = COPY [[COPY3]]
  26     ; GFX6-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  27     ; GFX8-LABEL: name: atomic_cmpswap_i32_1d
  28     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  29     ; GFX8-NEXT: {{  $}}
  30     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  31     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  32     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  33     ; 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)
  34     ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_vi]].sub0
  35     ; GFX8-NEXT: $vgpr0 = COPY [[COPY3]]
  36     ; GFX8-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  37     ; GFX10-LABEL: name: atomic_cmpswap_i32_1d
  38     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  39     ; GFX10-NEXT: {{  $}}
  40     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  41     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  42     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  43     ; 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)
  44     ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10_]].sub0
  45     ; GFX10-NEXT: $vgpr0 = COPY [[COPY3]]
  46     ; GFX10-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  47     ; GFX11-LABEL: name: atomic_cmpswap_i32_1d
  48     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  49     ; GFX11-NEXT: {{  $}}
  50     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  51     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  52     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  53     ; 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)
  54     ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx11_]].sub0
  55     ; GFX11-NEXT: $vgpr0 = COPY [[COPY3]]
  56     ; GFX11-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0
  57     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  58     %1:vgpr(<2 x s32>) = COPY $vgpr0_vgpr1
  59     %2:vgpr(s32) = COPY $vgpr2
  60     %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)
  61     $vgpr0 = COPY %3(s32)
  62     SI_RETURN_TO_EPILOG implicit $vgpr0
  63 ...
  64
  65 ---
  66 name: atomic_cmpswap_i32_1d_no_return
  67 legalized: true
  68 regBankSelected: true
  69 tracksRegLiveness: true
  70
  71 body: |
  72   bb.0:
  73     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  74
  75     ; GFX6-LABEL: name: atomic_cmpswap_i32_1d_no_return
  76     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  77     ; GFX6-NEXT: {{  $}}
  78     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  79     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  80     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  81     ; 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)
  82     ; GFX6-NEXT: S_ENDPGM 0
  83     ; GFX8-LABEL: name: atomic_cmpswap_i32_1d_no_return
  84     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  85     ; GFX8-NEXT: {{  $}}
  86     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  87     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  88     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  89     ; 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)
  90     ; GFX8-NEXT: S_ENDPGM 0
  91     ; GFX10-LABEL: name: atomic_cmpswap_i32_1d_no_return
  92     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  93     ; GFX10-NEXT: {{  $}}
  94     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  95     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  96     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  97     ; 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)
  98     ; GFX10-NEXT: S_ENDPGM 0
  99     ; GFX11-LABEL: name: atomic_cmpswap_i32_1d_no_return
 100     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
 101     ; GFX11-NEXT: {{  $}}
 102     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 103     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
 104     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
 105     ; 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)
 106     ; GFX11-NEXT: S_ENDPGM 0
 107     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 108     %1:vgpr(<2 x s32>) = COPY $vgpr0_vgpr1
 109     %2:vgpr(s32) = COPY $vgpr2
 110     %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)
 111     S_ENDPGM 0
 112 ...
 113
 114 ---
 115 name: atomic_cmpswap_i64_1d
 116 legalized: true
 117 regBankSelected: true
 118 tracksRegLiveness: true
 119
 120 body: |
 121   bb.0:
 122     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 123
 124     ; GFX6-LABEL: name: atomic_cmpswap_i64_1d
 125     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 126     ; GFX6-NEXT: {{  $}}
 127     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 128     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 129     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 130     ; 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)
 131     ; GFX6-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_si]].sub0_sub1
 132     ; GFX6-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 133     ; GFX6-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 134     ; GFX8-LABEL: name: atomic_cmpswap_i64_1d
 135     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 136     ; GFX8-NEXT: {{  $}}
 137     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 138     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 139     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 140     ; 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)
 141     ; GFX8-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_vi]].sub0_sub1
 142     ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 143     ; GFX8-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 144     ; GFX10-LABEL: name: atomic_cmpswap_i64_1d
 145     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 146     ; GFX10-NEXT: {{  $}}
 147     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 148     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 149     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 150     ; 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)
 151     ; GFX10-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10_]].sub0_sub1
 152     ; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 153     ; GFX10-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 154     ; GFX11-LABEL: name: atomic_cmpswap_i64_1d
 155     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 156     ; GFX11-NEXT: {{  $}}
 157     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 158     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 159     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 160     ; 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)
 161     ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx11_]].sub0_sub1
 162     ; GFX11-NEXT: $vgpr0_vgpr1 = COPY [[COPY3]]
 163     ; GFX11-NEXT: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 164     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 165     %1:vgpr(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 166     %2:vgpr(s32) = COPY $vgpr4
 167     %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)
 168     $vgpr0_vgpr1 = COPY %3(s64)
 169     SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 170 ...
 171
 172 ---
 173 name: atomic_cmpswap_i64_1d_no_return
 174 legalized: true
 175 regBankSelected: true
 176 tracksRegLiveness: true
 177
 178 body: |
 179   bb.0:
 180     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 181
 182     ; GFX6-LABEL: name: atomic_cmpswap_i64_1d_no_return
 183     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 184     ; GFX6-NEXT: {{  $}}
 185     ; GFX6-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 186     ; GFX6-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 187     ; GFX6-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 188     ; 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)
 189     ; GFX6-NEXT: S_ENDPGM 0
 190     ; GFX8-LABEL: name: atomic_cmpswap_i64_1d_no_return
 191     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 192     ; GFX8-NEXT: {{  $}}
 193     ; GFX8-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 194     ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 195     ; GFX8-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 196     ; 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)
 197     ; GFX8-NEXT: S_ENDPGM 0
 198     ; GFX10-LABEL: name: atomic_cmpswap_i64_1d_no_return
 199     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 200     ; GFX10-NEXT: {{  $}}
 201     ; GFX10-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 202     ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 203     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 204     ; 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)
 205     ; GFX10-NEXT: S_ENDPGM 0
 206     ; GFX11-LABEL: name: atomic_cmpswap_i64_1d_no_return
 207     ; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 208     ; GFX11-NEXT: {{  $}}
 209     ; GFX11-NEXT: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 210     ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 211     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 212     ; 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)
 213     ; GFX11-NEXT: S_ENDPGM 0
 214     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 215     %1:vgpr(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 216     %2:vgpr(s32) = COPY $vgpr4
 217     %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)
 218     S_ENDPGM 0
 219 ...