230 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py2# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -start-before=greedy,0 -stop-after=virtregrewriter,0 -o - %s | FileCheck %s3 4# This testcase used to fail due to introducing a spill of an SGPR5# 1024 for every subregister use inside the loop. With overlapping6# unspillable split ranges, it wasn't able to allocate one of the7# tuples. We avoid this by ensuring wide tuples are always allocated8# first (although the allocator should probably have been smart enough9# to handle this without that hint. Ideally it would understand we10# only need to spill/restore single subregisters at a time).11 12---13name: greedy_fail_alloc_sgpr1024_spill14tracksRegLiveness: true15frameInfo:16 adjustsStack: true17 hasCalls: true18machineFunctionInfo:19 explicitKernArgSize: 1620 maxKernArgAlign: 821 isEntryFunction: true22 waveLimiter: true23 scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3'24 stackPtrOffsetReg: '$sgpr32'25 occupancy: 626body: |27 ; CHECK-LABEL: name: greedy_fail_alloc_sgpr1024_spill28 ; CHECK: bb.0:29 ; CHECK-NEXT: successors: %bb.1(0x80000000)30 ; CHECK-NEXT: liveins: $sgpr14, $sgpr15, $vgpr0, $vgpr1, $vgpr2, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr1131 ; CHECK-NEXT: {{ $}}32 ; CHECK-NEXT: renamable $sgpr34_sgpr35 = COPY $sgpr8_sgpr933 ; CHECK-NEXT: renamable $sgpr33 = COPY $sgpr1534 ; CHECK-NEXT: renamable $sgpr50 = COPY $sgpr1435 ; CHECK-NEXT: renamable $sgpr36_sgpr37 = COPY $sgpr10_sgpr1136 ; CHECK-NEXT: renamable $sgpr38_sgpr39 = COPY $sgpr6_sgpr737 ; CHECK-NEXT: renamable $sgpr48_sgpr49 = COPY $sgpr4_sgpr538 ; CHECK-NEXT: renamable $sgpr64_sgpr65 = S_LOAD_DWORDX2_IMM renamable $sgpr34_sgpr35, 0, 0 :: (dereferenceable invariant load (s64), align 16, addrspace 4)39 ; CHECK-NEXT: renamable $sgpr68 = S_MOV_B32 040 ; CHECK-NEXT: renamable $sgpr69 = S_MOV_B32 041 ; CHECK-NEXT: renamable $sgpr70 = S_MOV_B32 042 ; CHECK-NEXT: renamable $sgpr71 = S_MOV_B32 043 ; CHECK-NEXT: renamable $sgpr72 = S_MOV_B32 044 ; CHECK-NEXT: renamable $sgpr73 = S_MOV_B32 045 ; CHECK-NEXT: renamable $sgpr74 = S_MOV_B32 046 ; CHECK-NEXT: renamable $sgpr75 = S_MOV_B32 047 ; CHECK-NEXT: renamable $sgpr76 = S_MOV_B32 048 ; CHECK-NEXT: renamable $sgpr77 = S_MOV_B32 049 ; CHECK-NEXT: renamable $sgpr78 = S_MOV_B32 050 ; CHECK-NEXT: renamable $sgpr79 = S_MOV_B32 051 ; CHECK-NEXT: renamable $sgpr80 = S_MOV_B32 052 ; CHECK-NEXT: renamable $sgpr81 = S_MOV_B32 053 ; CHECK-NEXT: renamable $sgpr82 = S_MOV_B32 054 ; CHECK-NEXT: renamable $sgpr83 = S_MOV_B32 055 ; CHECK-NEXT: renamable $sgpr84 = S_MOV_B32 056 ; CHECK-NEXT: renamable $sgpr85 = S_MOV_B32 057 ; CHECK-NEXT: renamable $sgpr86 = S_MOV_B32 058 ; CHECK-NEXT: renamable $sgpr87 = S_MOV_B32 059 ; CHECK-NEXT: renamable $sgpr88 = S_MOV_B32 060 ; CHECK-NEXT: SI_SPILL_S1024_SAVE renamable $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95_sgpr96_sgpr97_sgpr98_sgpr99, %stack.0, implicit $exec, implicit $sgpr32 :: (store (s1024) into %stack.0, align 4, addrspace 5)61 ; CHECK-NEXT: renamable $sgpr52_sgpr53 = IMPLICIT_DEF62 ; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr3263 ; CHECK-NEXT: dead $sgpr30_sgpr31 = SI_CALL renamable $sgpr52_sgpr53, 0, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr364 ; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr3265 ; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr3266 ; CHECK-NEXT: $sgpr4_sgpr5 = COPY killed renamable $sgpr48_sgpr4967 ; CHECK-NEXT: $sgpr6_sgpr7 = COPY killed renamable $sgpr38_sgpr3968 ; CHECK-NEXT: $sgpr8_sgpr9 = COPY killed renamable $sgpr34_sgpr3569 ; CHECK-NEXT: $sgpr10_sgpr11 = COPY killed renamable $sgpr36_sgpr3770 ; CHECK-NEXT: $sgpr12 = COPY killed renamable $sgpr5071 ; CHECK-NEXT: $sgpr13 = COPY killed renamable $sgpr3372 ; CHECK-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr52_sgpr53, 0, csr_amdgpu, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr0_sgpr1_sgpr2_sgpr373 ; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr3274 ; CHECK-NEXT: renamable $sgpr4_sgpr5 = COPY $exec, implicit-def $exec75 ; CHECK-NEXT: dead renamable $sgpr6_sgpr7 = IMPLICIT_DEF76 ; CHECK-NEXT: {{ $}}77 ; CHECK-NEXT: bb.1:78 ; CHECK-NEXT: successors: %bb.2(0x40000000), %bb.4(0x40000000)79 ; CHECK-NEXT: liveins: $sgpr4_sgpr5, $sgpr64_sgpr65:0x000000000000000F80 ; CHECK-NEXT: {{ $}}81 ; CHECK-NEXT: renamable $sgpr6_sgpr7 = COPY $exec, implicit-def $exec82 ; CHECK-NEXT: S_CBRANCH_EXECZ %bb.4, implicit $exec83 ; CHECK-NEXT: {{ $}}84 ; CHECK-NEXT: bb.2:85 ; CHECK-NEXT: successors: %bb.3(0x80000000)86 ; CHECK-NEXT: liveins: $sgpr4_sgpr5, $sgpr64_sgpr65:0x000000000000000F87 ; CHECK-NEXT: {{ $}}88 ; CHECK-NEXT: renamable $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95_sgpr96_sgpr97_sgpr98_sgpr99 = SI_SPILL_S1024_RESTORE %stack.0, implicit $exec, implicit $sgpr32 :: (load (s1024) from %stack.0, align 4, addrspace 5)89 ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_1024 = COPY renamable $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95_sgpr96_sgpr97_sgpr98_sgpr9990 ; CHECK-NEXT: renamable $sgpr6 = S_LSHL_B32 renamable $sgpr65, 1, implicit-def dead $scc91 ; CHECK-NEXT: dead [[COPY:%[0-9]+]]:vreg_1024 = V_INDIRECT_REG_WRITE_GPR_IDX_B32_V32 [[COPY]], 0, killed $sgpr6, 3, implicit-def $m0, implicit $m0, implicit $exec92 ; CHECK-NEXT: {{ $}}93 ; CHECK-NEXT: bb.3:94 ; CHECK-NEXT: successors: %bb.5(0x40000000), %bb.1(0x40000000)95 ; CHECK-NEXT: liveins: $sgpr4_sgpr5, $sgpr64_sgpr65:0x000000000000000F, $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95_sgpr96_sgpr97_sgpr98_sgpr99:0x000000000000000396 ; CHECK-NEXT: {{ $}}97 ; CHECK-NEXT: renamable $sgpr6_sgpr7 = S_OR_SAVEEXEC_B64 renamable $sgpr4_sgpr5, implicit-def $exec, implicit-def $scc, implicit $exec98 ; CHECK-NEXT: renamable $sgpr36 = COPY renamable $sgpr6899 ; CHECK-NEXT: renamable $sgpr37 = COPY renamable $sgpr68100 ; CHECK-NEXT: renamable $sgpr38 = COPY renamable $sgpr68101 ; CHECK-NEXT: renamable $sgpr39 = COPY renamable $sgpr68102 ; CHECK-NEXT: renamable $sgpr40 = COPY renamable $sgpr68103 ; CHECK-NEXT: renamable $sgpr41 = COPY renamable $sgpr68104 ; CHECK-NEXT: renamable $sgpr42 = COPY renamable $sgpr68105 ; CHECK-NEXT: renamable $sgpr43 = COPY renamable $sgpr68106 ; CHECK-NEXT: renamable $sgpr44 = COPY renamable $sgpr68107 ; CHECK-NEXT: renamable $sgpr45 = COPY renamable $sgpr68108 ; CHECK-NEXT: renamable $sgpr46 = COPY renamable $sgpr68109 ; CHECK-NEXT: renamable $sgpr47 = COPY renamable $sgpr68110 ; CHECK-NEXT: renamable $sgpr48 = COPY renamable $sgpr68111 ; CHECK-NEXT: renamable $sgpr49 = COPY renamable $sgpr68112 ; CHECK-NEXT: renamable $sgpr50 = COPY renamable $sgpr68113 ; CHECK-NEXT: renamable $sgpr51 = COPY renamable $sgpr68114 ; CHECK-NEXT: renamable $sgpr52 = COPY renamable $sgpr68115 ; CHECK-NEXT: renamable $sgpr53 = COPY renamable $sgpr68116 ; CHECK-NEXT: renamable $sgpr54 = COPY renamable $sgpr68117 ; CHECK-NEXT: renamable $sgpr55 = COPY renamable $sgpr68118 ; CHECK-NEXT: renamable $sgpr56 = COPY renamable $sgpr68119 ; CHECK-NEXT: renamable $sgpr57 = COPY killed renamable $sgpr68120 ; CHECK-NEXT: dead [[COPY1:%[0-9]+]]:vreg_1024 = COPY renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51_sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, implicit $exec121 ; CHECK-NEXT: $exec = S_XOR_B64_term $exec, killed renamable $sgpr6_sgpr7, implicit-def $scc122 ; CHECK-NEXT: S_CBRANCH_EXECZ %bb.5, implicit $exec123 ; CHECK-NEXT: S_BRANCH %bb.1124 ; CHECK-NEXT: {{ $}}125 ; CHECK-NEXT: bb.4:126 ; CHECK-NEXT: successors: %bb.5(0x80000000)127 ; CHECK-NEXT: liveins: $sgpr6_sgpr7, $sgpr64_sgpr65:0x0000000000000003128 ; CHECK-NEXT: {{ $}}129 ; CHECK-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr6_sgpr7, implicit-def $scc130 ; CHECK-NEXT: dead renamable $sgpr4 = S_LSHL_B32 killed renamable $sgpr64, 1, implicit-def dead $scc131 ; CHECK-NEXT: renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51_sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67 = SI_SPILL_S1024_RESTORE %stack.0, implicit $exec, implicit $sgpr32 :: (load (s1024) from %stack.0, align 4, addrspace 5)132 ; CHECK-NEXT: dead [[COPY2:%[0-9]+]]:vreg_1024 = COPY killed renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43_sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51_sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59_sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67133 ; CHECK-NEXT: {{ $}}134 ; CHECK-NEXT: bb.5:135 bb.0:136 liveins: $vgpr0, $vgpr1, $vgpr2, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11, $sgpr14, $sgpr15137 138 %0:sgpr_64 = COPY $sgpr8_sgpr9139 %1:sgpr_32 = COPY $sgpr15140 %2:sgpr_32 = COPY $sgpr14141 %3:sgpr_64 = COPY $sgpr10_sgpr11142 %4:sgpr_64 = COPY $sgpr6_sgpr7143 %5:sgpr_64 = COPY $sgpr4_sgpr5144 %6:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0, 0, 0 :: (dereferenceable invariant load (s64), align 16, addrspace 4)145 undef %7.sub0:sgpr_1024 = S_MOV_B32 0146 %7.sub1:sgpr_1024 = S_MOV_B32 0147 %7.sub2:sgpr_1024 = S_MOV_B32 0148 %7.sub3:sgpr_1024 = S_MOV_B32 0149 %7.sub4:sgpr_1024 = S_MOV_B32 0150 %7.sub5:sgpr_1024 = S_MOV_B32 0151 %7.sub6:sgpr_1024 = S_MOV_B32 0152 %7.sub7:sgpr_1024 = S_MOV_B32 0153 %7.sub8:sgpr_1024 = S_MOV_B32 0154 %7.sub9:sgpr_1024 = S_MOV_B32 0155 %7.sub10:sgpr_1024 = S_MOV_B32 0156 %7.sub11:sgpr_1024 = S_MOV_B32 0157 %7.sub12:sgpr_1024 = S_MOV_B32 0158 %7.sub13:sgpr_1024 = S_MOV_B32 0159 %7.sub14:sgpr_1024 = S_MOV_B32 0160 %7.sub15:sgpr_1024 = S_MOV_B32 0161 %7.sub16:sgpr_1024 = S_MOV_B32 0162 %7.sub17:sgpr_1024 = S_MOV_B32 0163 %7.sub18:sgpr_1024 = S_MOV_B32 0164 %7.sub19:sgpr_1024 = S_MOV_B32 0165 %7.sub20:sgpr_1024 = S_MOV_B32 0166 %8:sreg_64 = IMPLICIT_DEF167 ADJCALLSTACKUP 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32168 dead $sgpr30_sgpr31 = SI_CALL %8, 0, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3169 ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32170 ADJCALLSTACKUP 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32171 $sgpr4_sgpr5 = COPY %5172 $sgpr6_sgpr7 = COPY %4173 $sgpr8_sgpr9 = COPY %0174 $sgpr10_sgpr11 = COPY %3175 $sgpr12 = COPY %2176 $sgpr13 = COPY %1177 dead $sgpr30_sgpr31 = SI_CALL %8, 0, csr_amdgpu, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit killed $sgpr12, implicit killed $sgpr13, implicit $sgpr0_sgpr1_sgpr2_sgpr3178 ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32179 %9:sreg_64 = COPY $exec, implicit-def $exec180 %10:sreg_64 = IMPLICIT_DEF181 182 bb.1:183 successors: %bb.2, %bb.4184 185 %11:sreg_64 = COPY $exec, implicit-def $exec186 S_CBRANCH_EXECZ %bb.4, implicit $exec187 188 bb.2:189 %12:vreg_1024 = COPY %7190 %13:sreg_32 = S_LSHL_B32 %6.sub1, 1, implicit-def dead $scc191 %12:vreg_1024 = V_INDIRECT_REG_WRITE_GPR_IDX_B32_V32 %12, 0, %13, 3, implicit-def $m0, implicit $m0, implicit $exec192 193 bb.3:194 %14:sreg_64 = S_OR_SAVEEXEC_B64 %9, implicit-def $exec, implicit-def $scc, implicit $exec195 undef %15.sub0:sgpr_1024 = COPY %7.sub0196 %15.sub1:sgpr_1024 = COPY %7.sub0197 %15.sub2:sgpr_1024 = COPY %7.sub0198 %15.sub3:sgpr_1024 = COPY %7.sub0199 %15.sub4:sgpr_1024 = COPY %7.sub0200 %15.sub5:sgpr_1024 = COPY %7.sub0201 %15.sub6:sgpr_1024 = COPY %7.sub0202 %15.sub7:sgpr_1024 = COPY %7.sub0203 %15.sub8:sgpr_1024 = COPY %7.sub0204 %15.sub9:sgpr_1024 = COPY %7.sub0205 %15.sub10:sgpr_1024 = COPY %7.sub0206 %15.sub11:sgpr_1024 = COPY %7.sub0207 %15.sub12:sgpr_1024 = COPY %7.sub0208 %15.sub13:sgpr_1024 = COPY %7.sub0209 %15.sub14:sgpr_1024 = COPY %7.sub0210 %15.sub15:sgpr_1024 = COPY %7.sub0211 %15.sub16:sgpr_1024 = COPY %7.sub0212 %15.sub17:sgpr_1024 = COPY %7.sub0213 %15.sub18:sgpr_1024 = COPY %7.sub0214 %15.sub19:sgpr_1024 = COPY %7.sub0215 %15.sub20:sgpr_1024 = COPY %7.sub0216 %15.sub21:sgpr_1024 = COPY %7.sub0217 %16:vreg_1024 = COPY %15, implicit $exec218 $exec = S_XOR_B64_term $exec, %14, implicit-def $scc219 S_CBRANCH_EXECZ %bb.5, implicit $exec220 S_BRANCH %bb.1221 222 bb.4:223 $exec = S_OR_B64 $exec, %11, implicit-def $scc224 %17:sreg_32 = S_LSHL_B32 %6.sub0, 1, implicit-def dead $scc225 %16:vreg_1024 = COPY %7226 227 bb.5:228 229...230