198 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py2# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -verify-machineinstrs -run-pass=prologepilog %s -o - | FileCheck %s3 4# Test what happens when an SGPR is unavailable for the unused add5# carry out when materializing the frame index.6 7 8# There are truly no free SGPRs, so the entire frame index expansion9# needs to be inverted to restore the original frame register.10 11---12name: scavenge_sgpr_pei_no_sgprs13tracksRegLiveness: true14 15stack:16 - { id: 0, type: default, offset: 0, size: 4, alignment: 8192 }17 - { id: 1, type: default, offset: 0, size: 4, alignment: 8192 }18 19machineFunctionInfo:20 isEntryFunction: false21 scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr322 frameOffsetReg: $sgpr3323 stackPtrOffsetReg: $sgpr3224 25body: |26 bb.0:27 liveins: $vgpr128 29 ; CHECK-LABEL: name: scavenge_sgpr_pei_no_sgprs30 ; CHECK: liveins: $sgpr40, $sgpr41, $vgpr131 ; CHECK-NEXT: {{ $}}32 ; CHECK-NEXT: $sgpr40 = frame-setup COPY $sgpr3333 ; CHECK-NEXT: $sgpr33 = frame-setup S_ADD_I32 $sgpr32, 524224, implicit-def $scc34 ; CHECK-NEXT: $sgpr33 = frame-setup S_AND_B32 killed $sgpr33, 4294443008, implicit-def dead $scc35 ; CHECK-NEXT: $sgpr41 = frame-setup COPY $sgpr3436 ; CHECK-NEXT: $sgpr34 = frame-setup COPY $sgpr3237 ; CHECK-NEXT: $sgpr32 = frame-setup S_ADD_I32 $sgpr32, 2097152, implicit-def dead $scc38 ; CHECK-NEXT: S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr28, implicit-def $sgpr29, implicit-def $sgpr30, implicit-def $sgpr31, implicit-def $vcc39 ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec40 ; CHECK-NEXT: $sgpr42 = S_MOV_B32 819241 ; CHECK-NEXT: $vgpr0, dead $sgpr42_sgpr43 = V_ADD_CO_U32_e64 killed $sgpr42, killed $vgpr0, 0, implicit $exec42 ; CHECK-NEXT: $vgpr2 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec43 ; CHECK-NEXT: $sgpr42 = S_MOV_B32 1638444 ; CHECK-NEXT: $vgpr2, dead $sgpr42_sgpr43 = V_ADD_CO_U32_e64 killed $sgpr42, killed $vgpr2, 0, implicit $exec45 ; CHECK-NEXT: $vgpr0 = V_OR_B32_e32 killed $vgpr2, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr28, implicit $sgpr29, implicit $sgpr30, implicit $sgpr3146 ; CHECK-NEXT: $sgpr32 = frame-destroy COPY $sgpr3447 ; CHECK-NEXT: $sgpr34 = frame-destroy COPY $sgpr4148 ; CHECK-NEXT: $sgpr33 = frame-destroy COPY $sgpr4049 ; CHECK-NEXT: S_ENDPGM 0, implicit $vcc50 S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr28, implicit-def $sgpr29, implicit-def $sgpr30, implicit-def $sgpr31, implicit-def $vcc51 $vgpr0 = V_MOV_B32_e32 %stack.0, implicit $exec52 $vgpr0 = V_OR_B32_e32 %stack.1, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr28, implicit $sgpr29, implicit $sgpr30, implicit $sgpr3153 S_ENDPGM 0, implicit $vcc54...55 56# One 32-bit SGPR is available for the intermediate scale computation,57# so only an extra copy to VALU is necessary.58 59---60name: scavenge_sgpr_pei_one_sgpr61tracksRegLiveness: true62 63stack:64 - { id: 0, type: default, offset: 0, size: 4, alignment: 8192 }65 - { id: 1, type: default, offset: 0, size: 4, alignment: 8192 }66 67machineFunctionInfo:68 isEntryFunction: false69 scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr370 frameOffsetReg: $sgpr3371 stackPtrOffsetReg: $sgpr3272 73body: |74 bb.0:75 liveins: $vgpr176 77 ; CHECK-LABEL: name: scavenge_sgpr_pei_one_sgpr78 ; CHECK: liveins: $sgpr29, $sgpr40, $vgpr179 ; CHECK-NEXT: {{ $}}80 ; CHECK-NEXT: $sgpr29 = frame-setup COPY $sgpr3381 ; CHECK-NEXT: $sgpr33 = frame-setup S_ADD_I32 $sgpr32, 524224, implicit-def $scc82 ; CHECK-NEXT: $sgpr33 = frame-setup S_AND_B32 killed $sgpr33, 4294443008, implicit-def dead $scc83 ; CHECK-NEXT: $sgpr40 = frame-setup COPY $sgpr3484 ; CHECK-NEXT: $sgpr34 = frame-setup COPY $sgpr3285 ; CHECK-NEXT: $sgpr32 = frame-setup S_ADD_I32 $sgpr32, 2097152, implicit-def dead $scc86 ; CHECK-NEXT: S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr28, implicit-def $sgpr30, implicit-def $sgpr31, implicit-def $vcc87 ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec88 ; CHECK-NEXT: $sgpr42 = S_MOV_B32 819289 ; CHECK-NEXT: $vgpr0, dead $sgpr42_sgpr43 = V_ADD_CO_U32_e64 killed $sgpr42, killed $vgpr0, 0, implicit $exec90 ; CHECK-NEXT: $vgpr2 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec91 ; CHECK-NEXT: $sgpr42 = S_MOV_B32 1638492 ; CHECK-NEXT: $vgpr2, dead $sgpr42_sgpr43 = V_ADD_CO_U32_e64 killed $sgpr42, killed $vgpr2, 0, implicit $exec93 ; CHECK-NEXT: $vgpr0 = V_OR_B32_e32 killed $vgpr2, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr28, implicit $sgpr3194 ; CHECK-NEXT: $sgpr32 = frame-destroy COPY $sgpr3495 ; CHECK-NEXT: $sgpr34 = frame-destroy COPY $sgpr4096 ; CHECK-NEXT: $sgpr33 = frame-destroy COPY $sgpr2997 ; CHECK-NEXT: S_ENDPGM 0, implicit $vcc98 S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr28, implicit-def $sgpr30, implicit-def $sgpr31, implicit-def $vcc99 $vgpr0 = V_MOV_B32_e32 %stack.0, implicit $exec100 $vgpr0 = V_OR_B32_e32 %stack.1, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr28, implicit $sgpr31101 S_ENDPGM 0, implicit $vcc102...103 104# When only one 64-bit SGPR is available for the unused carry out pre gfx9,105# we must reuse one of the 32-bit SGPR sub-regs to materialize the offset.106 107---108name: scavenge_sgpr_pei_one_sgpr_64109tracksRegLiveness: true110 111stack:112 - { id: 0, type: default, offset: 0, size: 4, alignment: 8192 }113 - { id: 1, type: default, offset: 0, size: 4, alignment: 8192 }114 115machineFunctionInfo:116 isEntryFunction: false117 scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3118 frameOffsetReg: $sgpr33119 stackPtrOffsetReg: $sgpr32120 121body: |122 bb.0:123 liveins: $vgpr1124 125 ; CHECK-LABEL: name: scavenge_sgpr_pei_one_sgpr_64126 ; CHECK: liveins: $sgpr28, $sgpr29, $vgpr1127 ; CHECK-NEXT: {{ $}}128 ; CHECK-NEXT: $sgpr28 = frame-setup COPY $sgpr33129 ; CHECK-NEXT: $sgpr33 = frame-setup S_ADD_I32 $sgpr32, 524224, implicit-def $scc130 ; CHECK-NEXT: $sgpr33 = frame-setup S_AND_B32 killed $sgpr33, 4294443008, implicit-def dead $scc131 ; CHECK-NEXT: $sgpr29 = frame-setup COPY $sgpr34132 ; CHECK-NEXT: $sgpr34 = frame-setup COPY $sgpr32133 ; CHECK-NEXT: $sgpr32 = frame-setup S_ADD_I32 $sgpr32, 2097152, implicit-def dead $scc134 ; CHECK-NEXT: S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr30, implicit-def $sgpr31, implicit-def $vcc135 ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec136 ; CHECK-NEXT: $sgpr40 = S_MOV_B32 8192137 ; CHECK-NEXT: $vgpr0, dead $sgpr40_sgpr41 = V_ADD_CO_U32_e64 killed $sgpr40, killed $vgpr0, 0, implicit $exec138 ; CHECK-NEXT: $vgpr2 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec139 ; CHECK-NEXT: $sgpr40 = S_MOV_B32 16384140 ; CHECK-NEXT: $vgpr2, dead $sgpr40_sgpr41 = V_ADD_CO_U32_e64 killed $sgpr40, killed $vgpr2, 0, implicit $exec141 ; CHECK-NEXT: $vgpr0 = V_OR_B32_e32 killed $vgpr2, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr31142 ; CHECK-NEXT: $sgpr32 = frame-destroy COPY $sgpr34143 ; CHECK-NEXT: $sgpr34 = frame-destroy COPY $sgpr29144 ; CHECK-NEXT: $sgpr33 = frame-destroy COPY $sgpr28145 ; CHECK-NEXT: S_ENDPGM 0, implicit $vcc146 S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr30, implicit-def $sgpr31, implicit-def $vcc147 $vgpr0 = V_MOV_B32_e32 %stack.0, implicit $exec148 $vgpr0 = V_OR_B32_e32 %stack.1, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr31149 S_ENDPGM 0, implicit $vcc150...151 152# Prefer to use vcc as unused carry out.153 154---155name: scavenge_sgpr_pei_prefer_vcc156tracksRegLiveness: true157 158stack:159 - { id: 0, type: default, offset: 0, size: 4, alignment: 8192 }160 - { id: 1, type: default, offset: 0, size: 4, alignment: 8192 }161 162machineFunctionInfo:163 isEntryFunction: false164 scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3165 frameOffsetReg: $sgpr33166 stackPtrOffsetReg: $sgpr32167 168body: |169 bb.0:170 liveins: $vgpr1171 172 ; CHECK-LABEL: name: scavenge_sgpr_pei_prefer_vcc173 ; CHECK: liveins: $sgpr28, $sgpr29, $vgpr1174 ; CHECK-NEXT: {{ $}}175 ; CHECK-NEXT: $sgpr28 = frame-setup COPY $sgpr33176 ; CHECK-NEXT: $sgpr33 = frame-setup S_ADD_I32 $sgpr32, 524224, implicit-def $scc177 ; CHECK-NEXT: $sgpr33 = frame-setup S_AND_B32 killed $sgpr33, 4294443008, implicit-def dead $scc178 ; CHECK-NEXT: $sgpr29 = frame-setup COPY $sgpr34179 ; CHECK-NEXT: $sgpr34 = frame-setup COPY $sgpr32180 ; CHECK-NEXT: $sgpr32 = frame-setup S_ADD_I32 $sgpr32, 2097152, implicit-def dead $scc181 ; CHECK-NEXT: S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr30, implicit-def $sgpr31182 ; CHECK-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec183 ; CHECK-NEXT: $vcc_lo = S_MOV_B32 8192184 ; CHECK-NEXT: $vgpr0, dead $vcc = V_ADD_CO_U32_e64 killed $vcc_lo, killed $vgpr0, 0, implicit $exec185 ; CHECK-NEXT: $vgpr2 = V_LSHRREV_B32_e64 6, $sgpr33, implicit $exec186 ; CHECK-NEXT: $vcc_lo = S_MOV_B32 16384187 ; CHECK-NEXT: $vgpr2, dead $vcc = V_ADD_CO_U32_e64 killed $vcc_lo, killed $vgpr2, 0, implicit $exec188 ; CHECK-NEXT: $vgpr0 = V_OR_B32_e32 killed $vgpr2, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr31189 ; CHECK-NEXT: $sgpr32 = frame-destroy COPY $sgpr34190 ; CHECK-NEXT: $sgpr34 = frame-destroy COPY $sgpr29191 ; CHECK-NEXT: $sgpr33 = frame-destroy COPY $sgpr28192 ; CHECK-NEXT: S_ENDPGM 0193 S_NOP 0, implicit-def $sgpr4, implicit-def $sgpr5, implicit-def $sgpr6, implicit-def $sgpr7, implicit-def $sgpr8, implicit-def $sgpr9, implicit-def $sgpr10, implicit-def $sgpr11, implicit-def $sgpr12, implicit-def $sgpr13, implicit-def $sgpr14, implicit-def $sgpr15, implicit-def $sgpr16, implicit-def $sgpr17, implicit-def $sgpr18, implicit-def $sgpr19, implicit-def $sgpr20, implicit-def $sgpr21, implicit-def $sgpr22, implicit-def $sgpr23, implicit-def $sgpr24, implicit-def $sgpr25, implicit-def $sgpr26, implicit-def $sgpr27, implicit-def $sgpr30, implicit-def $sgpr31194 $vgpr0 = V_MOV_B32_e32 %stack.0, implicit $exec195 $vgpr0 = V_OR_B32_e32 %stack.1, $vgpr1, implicit $exec, implicit $sgpr4, implicit $sgpr5, implicit $sgpr6, implicit $sgpr7, implicit $sgpr8, implicit $sgpr9, implicit $sgpr10, implicit $sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $sgpr16, implicit $sgpr17, implicit $sgpr18, implicit $sgpr19, implicit $sgpr20, implicit $sgpr21, implicit $sgpr22, implicit $sgpr23, implicit $sgpr24, implicit $sgpr25, implicit $sgpr26, implicit $sgpr27, implicit $sgpr31196 S_ENDPGM 0197...198