221 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py2# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=regbankselect -regbankselect-fast -verify-machineinstrs %s -o - | FileCheck %s -check-prefix=FAST3# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -run-pass=regbankselect -regbankselect-greedy -verify-machineinstrs %s -o - | FileCheck %s -check-prefix=GREEDY4 5---6name: mfma_f32_32x32x4bf16_1k_vva7legalized: true8tracksRegLiveness: true9body: |10 bb.0:11 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr3112 13 ; FAST-LABEL: name: mfma_f32_32x32x4bf16_1k_vva14 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr3115 ; FAST-NEXT: {{ $}}16 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr117 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr318 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr3119 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<32 x s32>), 0, 0, 020 ; FAST-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY [[INT]](<32 x s32>)21 ; GREEDY-LABEL: name: mfma_f32_32x32x4bf16_1k_vva22 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr3123 ; GREEDY-NEXT: {{ $}}24 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr125 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr326 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr3127 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<32 x s32>), 0, 0, 028 ; GREEDY-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY [[INT]](<32 x s32>)29 %0:_(s64) = COPY $vgpr0_vgpr130 %1:_(s64) = COPY $vgpr2_vgpr331 %2:_(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr3132 %3:_(<32 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16.1k), %0, %1, %2, 0, 0, 033 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY %334...35 36---37name: mfma_f32_16x16x4bf16_1k_vva38legalized: true39tracksRegLiveness: true40body: |41 bb.0:42 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr1543 44 ; FAST-LABEL: name: mfma_f32_16x16x4bf16_1k_vva45 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr1546 ; FAST-NEXT: {{ $}}47 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr148 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr349 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr1550 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.16x16x4bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<16 x s32>), 0, 0, 051 ; FAST-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)52 ; GREEDY-LABEL: name: mfma_f32_16x16x4bf16_1k_vva53 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr1554 ; GREEDY-NEXT: {{ $}}55 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr156 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr357 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr1558 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.16x16x4bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<16 x s32>), 0, 0, 059 ; GREEDY-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)60 %0:_(s64) = COPY $vgpr0_vgpr161 %1:_(s64) = COPY $vgpr2_vgpr362 %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr1563 %3:_(<16 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.16x16x4bf16.1k), %0, %1, %2, 0, 0, 064 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %365...66 67---68name: mfma_f32_4x4x4bf16_1k_vva69legalized: true70tracksRegLiveness: true71body: |72 bb.0:73 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr374 75 ; FAST-LABEL: name: mfma_f32_4x4x4bf16_1k_vva76 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr377 ; FAST-NEXT: {{ $}}78 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr179 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr380 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr381 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.4x4x4bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<4 x s32>), 0, 0, 082 ; FAST-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)83 ; GREEDY-LABEL: name: mfma_f32_4x4x4bf16_1k_vva84 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr385 ; GREEDY-NEXT: {{ $}}86 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr187 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr388 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr389 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.4x4x4bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<4 x s32>), 0, 0, 090 ; GREEDY-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)91 %0:_(s64) = COPY $vgpr0_vgpr192 %1:_(s64) = COPY $vgpr2_vgpr393 %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr394 %3:_(<4 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.4x4x4bf16.1k), %0, %1, %2, 0, 0, 095 $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %396...97 98---99name: mfma_f32_32x32x8bf16_1k_vva100legalized: true101tracksRegLiveness: true102body: |103 bb.0:104 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31105 106 ; FAST-LABEL: name: mfma_f32_32x32x8bf16_1k_vva107 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31108 ; FAST-NEXT: {{ $}}109 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1110 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3111 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31112 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.32x32x8bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<32 x s32>), 0, 0, 0113 ; FAST-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY [[INT]](<32 x s32>)114 ; GREEDY-LABEL: name: mfma_f32_32x32x8bf16_1k_vva115 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31116 ; GREEDY-NEXT: {{ $}}117 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1118 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3119 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31120 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.32x32x8bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<32 x s32>), 0, 0, 0121 ; GREEDY-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY [[INT]](<32 x s32>)122 %0:_(s64) = COPY $vgpr0_vgpr1123 %1:_(s64) = COPY $vgpr2_vgpr3124 %2:_(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31125 %3:_(<32 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.32x32x8bf16.1k), %0, %1, %2, 0, 0, 0126 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY %3127...128 129---130name: mfma_f32_16x16x16bf16_1k_vva131legalized: true132tracksRegLiveness: true133body: |134 bb.0:135 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3136 137 ; FAST-LABEL: name: mfma_f32_16x16x16bf16_1k_vva138 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3139 ; FAST-NEXT: {{ $}}140 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1141 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3142 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3143 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.16x16x16bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<4 x s32>), 0, 0, 0144 ; FAST-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)145 ; GREEDY-LABEL: name: mfma_f32_16x16x16bf16_1k_vva146 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3147 ; GREEDY-NEXT: {{ $}}148 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1149 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3150 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3151 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.16x16x16bf16.1k), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<4 x s32>), 0, 0, 0152 ; GREEDY-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)153 %0:_(s64) = COPY $vgpr0_vgpr1154 %1:_(s64) = COPY $vgpr2_vgpr3155 %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3156 %3:_(<4 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f32.16x16x16bf16.1k), %0, %1, %2, 0, 0, 0157 $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3158...159 160---161name: mfma_f64_16x16x4f64_vva162legalized: true163tracksRegLiveness: true164body: |165 bb.0:166 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7167 168 ; FAST-LABEL: name: mfma_f64_16x16x4f64_vva169 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7170 ; FAST-NEXT: {{ $}}171 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1172 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3173 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<8 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7174 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<8 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f64.16x16x4f64), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<8 x s32>), 0, 0, 0175 ; FAST-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INT]](<8 x s32>)176 ; GREEDY-LABEL: name: mfma_f64_16x16x4f64_vva177 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7178 ; GREEDY-NEXT: {{ $}}179 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1180 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3181 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<8 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7182 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<8 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f64.16x16x4f64), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<8 x s32>), 0, 0, 0183 ; GREEDY-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[INT]](<8 x s32>)184 %0:_(s64) = COPY $vgpr0_vgpr1185 %1:_(s64) = COPY $vgpr2_vgpr3186 %2:_(<8 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7187 %3:_(<8 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f64.16x16x4f64), %0, %1, %2, 0, 0, 0188 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %3189...190 191---192name: mfma_f64_4x4x4f64_vva193legalized: true194tracksRegLiveness: true195body: |196 bb.0:197 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1198 199 ; FAST-LABEL: name: mfma_f64_4x4x4f64_vva200 ; FAST: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1201 ; FAST-NEXT: {{ $}}202 ; FAST-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1203 ; FAST-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3204 ; FAST-NEXT: [[COPY2:%[0-9]+]]:agpr(<2 x s32>) = COPY $agpr0_agpr1205 ; FAST-NEXT: [[INT:%[0-9]+]]:agpr(<2 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f64.4x4x4f64), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<2 x s32>), 0, 0, 0206 ; FAST-NEXT: $vgpr0_vgpr1 = COPY [[INT]](<2 x s32>)207 ; GREEDY-LABEL: name: mfma_f64_4x4x4f64_vva208 ; GREEDY: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1209 ; GREEDY-NEXT: {{ $}}210 ; GREEDY-NEXT: [[COPY:%[0-9]+]]:vgpr(s64) = COPY $vgpr0_vgpr1211 ; GREEDY-NEXT: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3212 ; GREEDY-NEXT: [[COPY2:%[0-9]+]]:agpr(<2 x s32>) = COPY $agpr0_agpr1213 ; GREEDY-NEXT: [[INT:%[0-9]+]]:agpr(<2 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f64.4x4x4f64), [[COPY]](s64), [[COPY1]](s64), [[COPY2]](<2 x s32>), 0, 0, 0214 ; GREEDY-NEXT: $vgpr0_vgpr1 = COPY [[INT]](<2 x s32>)215 %0:_(s64) = COPY $vgpr0_vgpr1216 %1:_(s64) = COPY $vgpr2_vgpr3217 %2:_(<2 x s32>) = COPY $agpr0_agpr1218 %3:_(<2 x s32>) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.mfma.f64.4x4x4f64), %0, %1, %2, 0, 0, 0219 $vgpr0_vgpr1 = COPY %3220...221