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
   6 ---
   7 name: atomic_cmpswap_i32_1d
   8 legalized: true
   9 regBankSelected: true
  10 tracksRegLiveness: true
  11
  12 body: |
  13   bb.0:
  14     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  15
  16     ; GFX6-LABEL: name: atomic_cmpswap_i32_1d
  17     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  18     ; GFX6: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  19     ; GFX6: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  20     ; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  21     ; GFX6: [[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) on custom "ImageResource")
  22     ; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_si]].sub0
  23     ; GFX6: $vgpr0 = COPY [[COPY3]]
  24     ; GFX6: SI_RETURN_TO_EPILOG implicit $vgpr0
  25     ; GFX8-LABEL: name: atomic_cmpswap_i32_1d
  26     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  27     ; GFX8: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  28     ; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  29     ; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  30     ; GFX8: [[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) on custom "ImageResource")
  31     ; GFX8: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_vi]].sub0
  32     ; GFX8: $vgpr0 = COPY [[COPY3]]
  33     ; GFX8: SI_RETURN_TO_EPILOG implicit $vgpr0
  34     ; GFX10-LABEL: name: atomic_cmpswap_i32_1d
  35     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  36     ; GFX10: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  37     ; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  38     ; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  39     ; GFX10: [[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) on custom "ImageResource")
  40     ; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V1_V1_gfx10_]].sub0
  41     ; GFX10: $vgpr0 = COPY [[COPY3]]
  42     ; GFX10: SI_RETURN_TO_EPILOG implicit $vgpr0
  43     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  44     %1:vgpr(<2 x s32>) = COPY $vgpr0_vgpr1
  45     %2:vgpr(s32) = COPY $vgpr2
  46     %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) on custom "ImageResource")
  47     $vgpr0 = COPY %3(s32)
  48     SI_RETURN_TO_EPILOG implicit $vgpr0
  49 ...
  50
  51 ---
  52 name: atomic_cmpswap_i32_1d_no_return
  53 legalized: true
  54 regBankSelected: true
  55 tracksRegLiveness: true
  56
  57 body: |
  58   bb.0:
  59     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  60
  61     ; GFX6-LABEL: name: atomic_cmpswap_i32_1d_no_return
  62     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  63     ; GFX6: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  64     ; GFX6: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  65     ; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  66     ; GFX6: [[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) on custom "ImageResource")
  67     ; GFX6: S_ENDPGM 0
  68     ; GFX8-LABEL: name: atomic_cmpswap_i32_1d_no_return
  69     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  70     ; GFX8: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  71     ; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  72     ; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  73     ; GFX8: [[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) on custom "ImageResource")
  74     ; GFX8: S_ENDPGM 0
  75     ; GFX10-LABEL: name: atomic_cmpswap_i32_1d_no_return
  76     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1, $vgpr2
  77     ; GFX10: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  78     ; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
  79     ; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
  80     ; GFX10: [[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) on custom "ImageResource")
  81     ; GFX10: S_ENDPGM 0
  82     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
  83     %1:vgpr(<2 x s32>) = COPY $vgpr0_vgpr1
  84     %2:vgpr(s32) = COPY $vgpr2
  85     %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) on custom "ImageResource")
  86     S_ENDPGM 0
  87 ...
  88
  89 ---
  90 name: atomic_cmpswap_i64_1d
  91 legalized: true
  92 regBankSelected: true
  93 tracksRegLiveness: true
  94
  95 body: |
  96   bb.0:
  97     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
  98
  99     ; GFX6-LABEL: name: atomic_cmpswap_i64_1d
 100     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 101     ; GFX6: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 102     ; GFX6: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 103     ; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 104     ; GFX6: [[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) on custom "ImageResource")
 105     ; GFX6: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_si]].sub0_sub1
 106     ; GFX6: $vgpr0_vgpr1 = COPY [[COPY3]]
 107     ; GFX6: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 108     ; GFX8-LABEL: name: atomic_cmpswap_i64_1d
 109     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 110     ; GFX8: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 111     ; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 112     ; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 113     ; GFX8: [[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) on custom "ImageResource")
 114     ; GFX8: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_vi]].sub0_sub1
 115     ; GFX8: $vgpr0_vgpr1 = COPY [[COPY3]]
 116     ; GFX8: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 117     ; GFX10-LABEL: name: atomic_cmpswap_i64_1d
 118     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 119     ; GFX10: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 120     ; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 121     ; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 122     ; GFX10: [[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) on custom "ImageResource")
 123     ; GFX10: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[IMAGE_ATOMIC_CMPSWAP_V2_V1_gfx10_]].sub0_sub1
 124     ; GFX10: $vgpr0_vgpr1 = COPY [[COPY3]]
 125     ; GFX10: SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 126     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 127     %1:vgpr(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 128     %2:vgpr(s32) = COPY $vgpr4
 129     %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) on custom "ImageResource")
 130     $vgpr0_vgpr1 = COPY %3(s64)
 131     SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1
 132 ...
 133
 134 ---
 135 name: atomic_cmpswap_i64_1d_no_return
 136 legalized: true
 137 regBankSelected: true
 138 tracksRegLiveness: true
 139
 140 body: |
 141   bb.0:
 142     liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 143
 144     ; GFX6-LABEL: name: atomic_cmpswap_i64_1d_no_return
 145     ; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 146     ; GFX6: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 147     ; GFX6: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 148     ; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 149     ; GFX6: [[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) on custom "ImageResource")
 150     ; GFX6: S_ENDPGM 0
 151     ; GFX8-LABEL: name: atomic_cmpswap_i64_1d_no_return
 152     ; GFX8: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 153     ; GFX8: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 154     ; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 155     ; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 156     ; GFX8: [[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) on custom "ImageResource")
 157     ; GFX8: S_ENDPGM 0
 158     ; GFX10-LABEL: name: atomic_cmpswap_i64_1d_no_return
 159     ; GFX10: liveins: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4
 160     ; GFX10: [[COPY:%[0-9]+]]:sgpr_256 = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 161     ; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 162     ; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr4
 163     ; GFX10: [[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) on custom "ImageResource")
 164     ; GFX10: S_ENDPGM 0
 165     %0:sgpr(<8 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7
 166     %1:vgpr(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
 167     %2:vgpr(s32) = COPY $vgpr4
 168     %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) on custom "ImageResource")
 169     S_ENDPGM 0
 170 ...