brintos

brintos / llvm-project-archived public Read only

0
0
Text · 14.3 KiB · 01506d0 Raw
239 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 52# RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx942 -run-pass si-fold-operands %s -o - | FileCheck %s3# RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx942 -start-before=si-fold-operands -stop-after=register-coalescer %s -o - | FileCheck %s --check-prefixes=COALESCE4# RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx908 -start-before=si-fold-operands -stop-after=register-coalescer %s -o - | FileCheck %s --check-prefixes=GFX908-COALESCE5 6...7---8name:            test9tracksRegLiveness: true10body:             |11  ; CHECK-LABEL: name: test12  ; CHECK: bb.0:13  ; CHECK-NEXT:   successors: %bb.1(0x80000000)14  ; CHECK-NEXT:   liveins: $sgpr4_sgpr515  ; CHECK-NEXT: {{  $}}16  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr4_sgpr517  ; CHECK-NEXT:   [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[COPY]](p4), 0, 0 :: (dereferenceable invariant load (s32), align 16, addrspace 4)18  ; CHECK-NEXT:   S_BITCMP1_B32 killed [[S_LOAD_DWORD_IMM]], 0, implicit-def $scc19  ; CHECK-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sgpr_32 = S_MOV_B32 020  ; CHECK-NEXT:   [[S_CSELECT_B64_:%[0-9]+]]:sreg_64_xexec = S_CSELECT_B64 -1, 0, implicit $scc21  ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec22  ; CHECK-NEXT:   [[COPY1:%[0-9]+]]:agpr_32 = COPY [[V_MOV_B32_e32_]]23  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF24  ; CHECK-NEXT:   [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[S_CSELECT_B64_]], implicit $exec25  ; CHECK-NEXT:   [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_NE_U32_e64 [[V_CNDMASK_B32_e64_]], 1, implicit $exec26  ; CHECK-NEXT: {{  $}}27  ; CHECK-NEXT: bb.1:28  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)29  ; CHECK-NEXT: {{  $}}30  ; CHECK-NEXT:   [[PHI:%[0-9]+]]:agpr_32 = PHI [[COPY1]], %bb.0, %24, %bb.331  ; CHECK-NEXT:   [[PHI1:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_]], %bb.0, %11, %bb.332  ; CHECK-NEXT:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[PHI]]33  ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:agpr_32 = COPY [[PHI]]34  ; CHECK-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 -135  ; CHECK-NEXT:   $vcc = S_AND_B64 $exec, [[V_CMP_NE_U32_e64_]], implicit-def $scc36  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.3, implicit $vcc37  ; CHECK-NEXT:   S_BRANCH %bb.238  ; CHECK-NEXT: {{  $}}39  ; CHECK-NEXT: bb.2:40  ; CHECK-NEXT:   successors: %bb.3(0x80000000)41  ; CHECK-NEXT: {{  $}}42  ; CHECK-NEXT:   [[S_OR_B32_:%[0-9]+]]:sreg_32 = S_OR_B32 [[PHI1]], 1, implicit-def dead $scc43  ; CHECK-NEXT:   [[S_ASHR_I32_:%[0-9]+]]:sreg_32 = S_ASHR_I32 [[PHI1]], 31, implicit-def dead $scc44  ; CHECK-NEXT:   [[S_AND_B32_:%[0-9]+]]:sreg_32 = S_AND_B32 killed [[S_ASHR_I32_]], killed [[S_OR_B32_]], implicit-def dead $scc45  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[V_MOV_B32_e32_]], %subreg.sub1, [[V_MOV_B32_e32_]], %subreg.sub2, [[V_MOV_B32_e32_]], %subreg.sub346  ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_]], %subreg.sub147  ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]48  ; CHECK-NEXT:   [[V_ACCVGPR_WRITE_B32_e64_:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 0, implicit $exec49  ; CHECK-NEXT:   [[V_ACCVGPR_WRITE_B32_e64_1:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 0, implicit $exec50  ; CHECK-NEXT:   [[V_ACCVGPR_WRITE_B32_e64_2:%[0-9]+]]:agpr_32 = V_ACCVGPR_WRITE_B32_e64 0, implicit $exec51  ; CHECK-NEXT:   [[REG_SEQUENCE2:%[0-9]+]]:areg_128_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[V_ACCVGPR_WRITE_B32_e64_]], %subreg.sub1, [[V_ACCVGPR_WRITE_B32_e64_1]], %subreg.sub2, [[V_ACCVGPR_WRITE_B32_e64_2]], %subreg.sub352  ; CHECK-NEXT:   [[V_MFMA_F32_16X16X16F16_e64_:%[0-9]+]]:areg_128_align2 = V_MFMA_F32_16X16X16F16_e64 [[COPY4]], [[COPY4]], killed [[REG_SEQUENCE2]], 0, 0, 0, implicit $mode, implicit $exec53  ; CHECK-NEXT:   [[S_MOV_B64_1:%[0-9]+]]:sreg_64 = S_MOV_B64 054  ; CHECK-NEXT: {{  $}}55  ; CHECK-NEXT: bb.3:56  ; CHECK-NEXT:   successors: %bb.4(0x40000000), %bb.1(0x40000000)57  ; CHECK-NEXT: {{  $}}58  ; CHECK-NEXT:   [[PHI2:%[0-9]+]]:sreg_32 = PHI [[DEF]], %bb.1, [[S_AND_B32_]], %bb.259  ; CHECK-NEXT:   [[PHI3:%[0-9]+]]:agpr_32 = PHI [[COPY3]], %bb.1, [[V_MFMA_F32_16X16X16F16_e64_]].sub0, %bb.260  ; CHECK-NEXT:   [[PHI4:%[0-9]+]]:sreg_64_xexec = PHI [[S_MOV_B64_]], %bb.1, [[S_MOV_B64_1]], %bb.261  ; CHECK-NEXT:   [[V_CNDMASK_B32_e64_1:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[PHI4]], implicit $exec62  ; CHECK-NEXT:   [[V_CMP_NE_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_NE_U32_e64 [[V_CNDMASK_B32_e64_1]], 1, implicit $exec63  ; CHECK-NEXT:   $vcc = S_AND_B64 $exec, [[V_CMP_NE_U32_e64_1]], implicit-def $scc64  ; CHECK-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc65  ; CHECK-NEXT:   S_BRANCH %bb.466  ; CHECK-NEXT: {{  $}}67  ; CHECK-NEXT: bb.4:68  ; CHECK-NEXT:   successors: %bb.5(0x80000000)69  ; CHECK-NEXT: {{  $}}70  ; CHECK-NEXT: bb.5:71  ; CHECK-NEXT:   S_ENDPGM 072  ;73  ; COALESCE-LABEL: name: test74  ; COALESCE: bb.0:75  ; COALESCE-NEXT:   successors: %bb.1(0x80000000)76  ; COALESCE-NEXT:   liveins: $sgpr4_sgpr577  ; COALESCE-NEXT: {{  $}}78  ; COALESCE-NEXT:   [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr4_sgpr579  ; COALESCE-NEXT:   [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[COPY]](p4), 0, 0 :: (dereferenceable invariant load (s32), align 16, addrspace 4)80  ; COALESCE-NEXT:   S_BITCMP1_B32 [[S_LOAD_DWORD_IMM]], 0, implicit-def $scc81  ; COALESCE-NEXT:   undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_64 = S_MOV_B32 082  ; COALESCE-NEXT:   [[S_CSELECT_B64_:%[0-9]+]]:sreg_64_xexec = S_CSELECT_B64 -1, 0, implicit killed $scc83  ; COALESCE-NEXT:   undef [[AV_MOV_:%[0-9]+]].sub0:areg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec84  ; COALESCE-NEXT:   [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[S_CSELECT_B64_]], implicit $exec85  ; COALESCE-NEXT:   [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_NE_U32_e64 1, [[V_CNDMASK_B32_e64_]], implicit $exec86  ; COALESCE-NEXT:   undef [[AV_MOV_1:%[0-9]+]].sub1:areg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec87  ; COALESCE-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 088  ; COALESCE-NEXT: {{  $}}89  ; COALESCE-NEXT: bb.1:90  ; COALESCE-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)91  ; COALESCE-NEXT: {{  $}}92  ; COALESCE-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[AV_MOV_]].sub093  ; COALESCE-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 -194  ; COALESCE-NEXT:   $vcc = S_AND_B64 $exec, [[V_CMP_NE_U32_e64_]], implicit-def dead $scc95  ; COALESCE-NEXT:   S_CBRANCH_VCCNZ %bb.3, implicit killed $vcc96  ; COALESCE-NEXT:   S_BRANCH %bb.297  ; COALESCE-NEXT: {{  $}}98  ; COALESCE-NEXT: bb.2:99  ; COALESCE-NEXT:   successors: %bb.3(0x80000000)100  ; COALESCE-NEXT: {{  $}}101  ; COALESCE-NEXT:   [[S_OR_B32_:%[0-9]+]]:sreg_32 = S_OR_B32 [[S_MOV_B32_1]], 1, implicit-def dead $scc102  ; COALESCE-NEXT:   [[S_ASHR_I32_:%[0-9]+]]:sreg_32 = S_ASHR_I32 [[S_MOV_B32_1]], 31, implicit-def dead $scc103  ; COALESCE-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_AND_B32 [[S_ASHR_I32_]], [[S_OR_B32_]], implicit-def dead $scc104  ; COALESCE-NEXT:   [[S_MOV_B32_:%[0-9]+]].sub1:sgpr_64 = COPY [[S_MOV_B32_]].sub0105  ; COALESCE-NEXT:   [[COPY2:%[0-9]+]]:vreg_64_align2 = COPY [[S_MOV_B32_]]106  ; COALESCE-NEXT:   [[AV_MOV_1:%[0-9]+]].sub0:areg_128_align2 = COPY [[COPY1]]107  ; COALESCE-NEXT:   [[AV_MOV_1:%[0-9]+]].sub2:areg_128_align2 = COPY [[AV_MOV_1]].sub1108  ; COALESCE-NEXT:   [[AV_MOV_1:%[0-9]+]].sub3:areg_128_align2 = COPY [[AV_MOV_1]].sub1109  ; COALESCE-NEXT:   [[AV_MOV_:%[0-9]+]]:areg_128_align2 = V_MFMA_F32_16X16X16F16_e64 [[COPY2]], [[COPY2]], [[AV_MOV_1]], 0, 0, 0, implicit $mode, implicit $exec110  ; COALESCE-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 0111  ; COALESCE-NEXT: {{  $}}112  ; COALESCE-NEXT: bb.3:113  ; COALESCE-NEXT:   successors: %bb.4(0x40000000), %bb.1(0x40000000)114  ; COALESCE-NEXT: {{  $}}115  ; COALESCE-NEXT:   [[V_CNDMASK_B32_e64_1:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[S_MOV_B64_]], implicit $exec116  ; COALESCE-NEXT:   [[V_CMP_NE_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_NE_U32_e64 1, [[V_CNDMASK_B32_e64_1]], implicit $exec117  ; COALESCE-NEXT:   $vcc = S_AND_B64 $exec, [[V_CMP_NE_U32_e64_1]], implicit-def dead $scc118  ; COALESCE-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc119  ; COALESCE-NEXT:   S_BRANCH %bb.4120  ; COALESCE-NEXT: {{  $}}121  ; COALESCE-NEXT: bb.4:122  ; COALESCE-NEXT:   successors: %bb.5(0x80000000)123  ; COALESCE-NEXT: {{  $}}124  ; COALESCE-NEXT: bb.5:125  ; COALESCE-NEXT:   S_ENDPGM 0126  ;127  ; GFX908-COALESCE-LABEL: name: test128  ; GFX908-COALESCE: bb.0:129  ; GFX908-COALESCE-NEXT:   successors: %bb.1(0x80000000)130  ; GFX908-COALESCE-NEXT:   liveins: $sgpr4_sgpr5131  ; GFX908-COALESCE-NEXT: {{  $}}132  ; GFX908-COALESCE-NEXT:   [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr4_sgpr5133  ; GFX908-COALESCE-NEXT:   [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[COPY]](p4), 0, 0 :: (dereferenceable invariant load (s32), align 16, addrspace 4)134  ; GFX908-COALESCE-NEXT:   S_BITCMP1_B32 [[S_LOAD_DWORD_IMM]], 0, implicit-def $scc135  ; GFX908-COALESCE-NEXT:   undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_64 = S_MOV_B32 0136  ; GFX908-COALESCE-NEXT:   [[S_CSELECT_B64_:%[0-9]+]]:sreg_64_xexec = S_CSELECT_B64 -1, 0, implicit killed $scc137  ; GFX908-COALESCE-NEXT:   undef [[AV_MOV_:%[0-9]+]].sub0:areg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec138  ; GFX908-COALESCE-NEXT:   [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[S_CSELECT_B64_]], implicit $exec139  ; GFX908-COALESCE-NEXT:   [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_NE_U32_e64 1, [[V_CNDMASK_B32_e64_]], implicit $exec140  ; GFX908-COALESCE-NEXT:   undef [[AV_MOV_1:%[0-9]+]].sub1:areg_128_align2 = AV_MOV_B32_IMM_PSEUDO 0, implicit $exec141  ; GFX908-COALESCE-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0142  ; GFX908-COALESCE-NEXT: {{  $}}143  ; GFX908-COALESCE-NEXT: bb.1:144  ; GFX908-COALESCE-NEXT:   successors: %bb.2(0x40000000), %bb.3(0x40000000)145  ; GFX908-COALESCE-NEXT: {{  $}}146  ; GFX908-COALESCE-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[AV_MOV_]].sub0147  ; GFX908-COALESCE-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 -1148  ; GFX908-COALESCE-NEXT:   $vcc = S_AND_B64 $exec, [[V_CMP_NE_U32_e64_]], implicit-def dead $scc149  ; GFX908-COALESCE-NEXT:   S_CBRANCH_VCCNZ %bb.3, implicit killed $vcc150  ; GFX908-COALESCE-NEXT:   S_BRANCH %bb.2151  ; GFX908-COALESCE-NEXT: {{  $}}152  ; GFX908-COALESCE-NEXT: bb.2:153  ; GFX908-COALESCE-NEXT:   successors: %bb.3(0x80000000)154  ; GFX908-COALESCE-NEXT: {{  $}}155  ; GFX908-COALESCE-NEXT:   [[S_OR_B32_:%[0-9]+]]:sreg_32 = S_OR_B32 [[S_MOV_B32_1]], 1, implicit-def dead $scc156  ; GFX908-COALESCE-NEXT:   [[S_ASHR_I32_:%[0-9]+]]:sreg_32 = S_ASHR_I32 [[S_MOV_B32_1]], 31, implicit-def dead $scc157  ; GFX908-COALESCE-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_AND_B32 [[S_ASHR_I32_]], [[S_OR_B32_]], implicit-def dead $scc158  ; GFX908-COALESCE-NEXT:   [[S_MOV_B32_:%[0-9]+]].sub1:sgpr_64 = COPY [[S_MOV_B32_]].sub0159  ; GFX908-COALESCE-NEXT:   [[COPY2:%[0-9]+]]:vreg_64_align2 = COPY [[S_MOV_B32_]]160  ; GFX908-COALESCE-NEXT:   [[AV_MOV_1:%[0-9]+]].sub0:areg_128_align2 = COPY [[COPY1]]161  ; GFX908-COALESCE-NEXT:   [[AV_MOV_1:%[0-9]+]].sub2:areg_128_align2 = COPY [[AV_MOV_1]].sub1162  ; GFX908-COALESCE-NEXT:   [[AV_MOV_1:%[0-9]+]].sub3:areg_128_align2 = COPY [[AV_MOV_1]].sub1163  ; GFX908-COALESCE-NEXT:   [[AV_MOV_:%[0-9]+]]:areg_128_align2 = V_MFMA_F32_16X16X16F16_e64 [[COPY2]], [[COPY2]], [[AV_MOV_1]], 0, 0, 0, implicit $mode, implicit $exec164  ; GFX908-COALESCE-NEXT:   [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 0165  ; GFX908-COALESCE-NEXT: {{  $}}166  ; GFX908-COALESCE-NEXT: bb.3:167  ; GFX908-COALESCE-NEXT:   successors: %bb.4(0x40000000), %bb.1(0x40000000)168  ; GFX908-COALESCE-NEXT: {{  $}}169  ; GFX908-COALESCE-NEXT:   [[V_CNDMASK_B32_e64_1:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, [[S_MOV_B64_]], implicit $exec170  ; GFX908-COALESCE-NEXT:   [[V_CMP_NE_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_NE_U32_e64 1, [[V_CNDMASK_B32_e64_1]], implicit $exec171  ; GFX908-COALESCE-NEXT:   $vcc = S_AND_B64 $exec, [[V_CMP_NE_U32_e64_1]], implicit-def dead $scc172  ; GFX908-COALESCE-NEXT:   S_CBRANCH_VCCNZ %bb.1, implicit killed $vcc173  ; GFX908-COALESCE-NEXT:   S_BRANCH %bb.4174  ; GFX908-COALESCE-NEXT: {{  $}}175  ; GFX908-COALESCE-NEXT: bb.4:176  ; GFX908-COALESCE-NEXT:   successors: %bb.5(0x80000000)177  ; GFX908-COALESCE-NEXT: {{  $}}178  ; GFX908-COALESCE-NEXT: bb.5:179  ; GFX908-COALESCE-NEXT:   S_ENDPGM 0180  bb.0:181    successors: %bb.1182    liveins: $sgpr4_sgpr5183 184    %0:sgpr_64(p4) = COPY $sgpr4_sgpr5185    %1:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0(p4), 0, 0 :: (dereferenceable invariant load (s32), align 16, addrspace 4)186    S_BITCMP1_B32 killed %1, 0, implicit-def $scc187    %2:sgpr_32 = S_MOV_B32 0188    %3:sreg_64_xexec = S_CSELECT_B64 -1, 0, implicit $scc189    %4:vgpr_32 = V_MOV_B32_e32 0, implicit $exec190    %5:sreg_32 = IMPLICIT_DEF191    %6:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, %3, implicit $exec192    %7:sreg_64_xexec = V_CMP_NE_U32_e64 %6, 1, implicit $exec193 194  bb.1:195    successors: %bb.2, %bb.3196 197    %8:vgpr_32 = PHI %4, %bb.0, %9, %bb.3198    %10:sreg_32 = PHI %2, %bb.0, %11, %bb.3199    %12:agpr_32 = COPY %8200    %13:sreg_64 = S_MOV_B64 -1201    $vcc = S_AND_B64 $exec, %7, implicit-def $scc202    S_CBRANCH_VCCNZ %bb.3, implicit $vcc203    S_BRANCH %bb.2204 205  bb.2:206    successors: %bb.3207 208    %14:sreg_32 = S_OR_B32 %10, 1, implicit-def dead $scc209    %15:sreg_32 = S_ASHR_I32 %10, 31, implicit-def dead $scc210    %16:sreg_32 = S_AND_B32 killed %15, killed %14, implicit-def dead $scc211    %17:vreg_128_align2 = REG_SEQUENCE %8, %subreg.sub0, %4, %subreg.sub1, %4, %subreg.sub2, %4, %subreg.sub3212    %18:sreg_64 = REG_SEQUENCE %2, %subreg.sub0, %2, %subreg.sub1213    %19:vreg_64_align2 = COPY %18214    %20:areg_128_align2 = COPY %17215    %21:areg_128_align2 = V_MFMA_F32_16X16X16F16_e64 %19, %19, killed %20, 0, 0, 0, implicit $mode, implicit $exec216    %22:vgpr_32 = COPY %21.sub0217    %23:sreg_64 = S_MOV_B64 0218 219  bb.3:220    successors: %bb.4, %bb.1221 222    %11:sreg_32 = PHI %5, %bb.1, %16, %bb.2223    %24:agpr_32 = PHI %12, %bb.1, %21.sub0, %bb.2224    %25:sreg_64_xexec = PHI %13, %bb.1, %23, %bb.2225    %9:vgpr_32 = COPY %24226    %26:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, 1, %25, implicit $exec227    %27:sreg_64_xexec = V_CMP_NE_U32_e64 %26, 1, implicit $exec228    $vcc = S_AND_B64 $exec, %27, implicit-def $scc229    S_CBRANCH_VCCNZ %bb.1, implicit $vcc230    S_BRANCH %bb.4231 232  bb.4:233    successors: %bb.5234 235  bb.5:236    S_ENDPGM 0237 238...239