222 lines · plain
1# RUN: llc -mtriple=amdgcn -mcpu=gfx90a -verify-machineinstrs -run-pass si-load-store-opt -o - %s | FileCheck -check-prefix=GFX9 %s2 3# GFX9-LABEL: name: diffoporder_add4 5# GFX9: %{{[0-9]+}}:vreg_64_align2 = REG_SEQUENCE6 7# GFX9: S_MOV_B32 61448# GFX9-NEXT: V_ADD_CO_U329# GFX9-NEXT: V_ADDC_U3210# GFX9-NEXT: [[PTR0:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE11# GFX9-NEXT: %{{[0-9]+}}:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[PTR0]], -2048, 012# GFX9: %{{[0-9]+}}:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[PTR0]], 0, 013 14name: diffoporder_add15body: |16 bb.0.entry:17 %0:sgpr_64 = COPY $sgpr0_sgpr118 %1:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0, 36, 019 %3:sgpr_128 = COPY $sgpr96_sgpr97_sgpr98_sgpr9920 %4:sreg_32_xm0 = COPY $sgpr10121 %5:sreg_32_xm0 = S_MOV_B32 022 $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %323 $sgpr4 = COPY %424 $vgpr0 = V_MOV_B32_e32 0, implicit $exec25 %6:vreg_64_align2 = COPY $vgpr0_vgpr126 %7:vgpr_32 = V_AND_B32_e32 255, %6.sub0, implicit $exec27 %8:vgpr_32 = V_MOV_B32_e32 0, implicit $exec28 %9:vreg_64_align2 = REG_SEQUENCE killed %7, %subreg.sub0, %8, %subreg.sub129 %10:vgpr_32 = V_LSHLREV_B32_e64 7, %6.sub0, implicit $exec30 %11:vgpr_32 = V_AND_B32_e32 -32768, killed %10, implicit $exec31 %12:sgpr_32 = COPY %1.sub132 %13:vgpr_32 = COPY %533 %14:vgpr_32, %15:sreg_64_xexec = V_ADD_CO_U32_e64 %1.sub0, %11, 0, implicit $exec34 %16:vgpr_32 = COPY %1235 %17:vgpr_32, dead %18:sreg_64_xexec = V_ADDC_U32_e64 %16, %13, killed %15, 0, implicit $exec36 %19:vreg_64_align2 = REG_SEQUENCE %14, %subreg.sub0, %17, %subreg.sub137 %20:vreg_64_align2 = V_LSHLREV_B64_e64 3, %9, implicit $exec38 %21:vgpr_32, %22:sreg_64_xexec = V_ADD_CO_U32_e64 %14, %20.sub0, 0, implicit $exec39 %23:vgpr_32, dead %24:sreg_64_xexec = V_ADDC_U32_e64 %17, %20.sub1, killed %22, 0, implicit $exec40 %25:sgpr_32 = S_MOV_B32 409641 %26:vgpr_32, %27:sreg_64_xexec = V_ADD_CO_U32_e64 %25, %21, 0, implicit $exec42 %28:vgpr_32, dead %29:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %27, 0, implicit $exec43 %30:vreg_64_align2 = REG_SEQUENCE %26, %subreg.sub0, %28, %subreg.sub144 %31:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %30, 0, 0, implicit $exec45 %32:sgpr_32 = S_MOV_B32 614446 %33:vgpr_32, %34:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %32, 0, implicit $exec47 %35:vgpr_32, dead %36:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %34, 0, implicit $exec48 %37:vreg_64_align2 = REG_SEQUENCE %33, %subreg.sub0, %35, %subreg.sub149 %38:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %37, 0, 0, implicit $exec50...51---52 53# GFX9-LABEL: name: LowestInMiddle54# GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 1120055# GFX9: [[BASE_LO:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_5:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 %{{[0-9]+}}, [[S_MOV_B32_1]]56# GFX9: [[BASE_HI:%[0-9]+]]:vgpr_32, dead %{{[0-9]+}}:sreg_64_xexec = V_ADDC_U32_e64 %{{[0-9]+}}, 0, killed [[V_ADD_CO_U32_e64_5]]57# GFX9: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[BASE_LO]], %subreg.sub0, [[BASE_HI]], %subreg.sub158# GFX9: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE2]], -3200, 059#60# GFX9: [[S_MOV_B32_2:%[0-9]+]]:sgpr_32 = S_MOV_B32 640061# GFX9: [[BASE1_LO:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_7:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 %{{[0-9]+}}, [[S_MOV_B32_2]]62# GFX9: [[BASE1_HI:%[0-9]+]]:vgpr_32, dead %{{[0-9]+}}:sreg_64_xexec = V_ADDC_U32_e64 %{{[0-9]+}}, 0, killed [[V_ADD_CO_U32_e64_7]]63# GFX9: [[REG_SEQUENCE3:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[BASE1_LO]], %subreg.sub0, [[BASE1_HI]], %subreg.sub164# GFX9: [[GLOBAL_LOAD_DWORDX2_1:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE3]], 0, 0,65# GFX9: [[GLOBAL_LOAD_DWORDX2_2:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE2]], 0, 0,66 67name: LowestInMiddle68body: |69 bb.0.entry:70 %0:sgpr_64 = COPY $sgpr0_sgpr171 %1:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0, 36, 072 %3:sgpr_128 = COPY $sgpr96_sgpr97_sgpr98_sgpr9973 %4:sreg_32_xm0 = COPY $sgpr10174 %5:sreg_32_xm0 = S_MOV_B32 075 $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %376 $sgpr4 = COPY %477 $vgpr0 = V_MOV_B32_e32 0, implicit $exec78 %6:vreg_64_align2 = COPY $vgpr0_vgpr179 %7:vgpr_32 = V_AND_B32_e32 255, %6.sub0, implicit $exec80 %8:vgpr_32 = V_MOV_B32_e32 0, implicit $exec81 %9:vreg_64_align2 = REG_SEQUENCE killed %7, %subreg.sub0, %8, %subreg.sub182 %10:vgpr_32 = V_LSHLREV_B32_e64 7, %6.sub0, implicit $exec83 %11:vgpr_32 = V_AND_B32_e32 -32768, killed %10, implicit $exec84 %12:sgpr_32 = COPY %1.sub185 %13:vgpr_32 = COPY %586 %14:vgpr_32, %15:sreg_64_xexec = V_ADD_CO_U32_e64 %1.sub0, %11, 0, implicit $exec87 %16:vgpr_32 = COPY %1288 %17:vgpr_32, dead %18:sreg_64_xexec = V_ADDC_U32_e64 %16, %13, killed %15, 0, implicit $exec89 %19:vreg_64_align2 = REG_SEQUENCE %14, %subreg.sub0, %17, %subreg.sub190 %20:vreg_64_align2 = V_LSHLREV_B64_e64 3, %9, implicit $exec91 %21:vgpr_32, %22:sreg_64_xexec = V_ADD_CO_U32_e64 %14, %20.sub0, 0, implicit $exec92 %23:vgpr_32, dead %24:sreg_64_xexec = V_ADDC_U32_e64 %17, %20.sub1, killed %22, 0, implicit $exec93 %25:sgpr_32 = S_MOV_B32 800094 %26:vgpr_32, %27:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %25, 0, implicit $exec95 %28:vgpr_32, dead %29:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %27, 0, implicit $exec96 %30:vreg_64_align2 = REG_SEQUENCE %26, %subreg.sub0, %28, %subreg.sub197 %31:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %30, 0, 0, implicit $exec98 %32:sgpr_32 = S_MOV_B32 640099 %33:vgpr_32, %34:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %32, 0, implicit $exec100 %35:vgpr_32, dead %36:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %34, 0, implicit $exec101 %37:vreg_64_align2 = REG_SEQUENCE %33, %subreg.sub0, %35, %subreg.sub1102 %38:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %37, 0, 0, implicit $exec103 %39:sgpr_32 = S_MOV_B32 11200104 %40:vgpr_32, %41:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %39, 0, implicit $exec105 %42:vgpr_32, dead %43:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %41, 0, implicit $exec106 %44:vreg_64_align2 = REG_SEQUENCE %40, %subreg.sub0, %42, %subreg.sub1107 %45:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %44, 0, 0, implicit $exec108...109---110 111# GFX9-LABEL: name: NegativeDistance112# GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 10240113# GFX9: [[V_ADD_CO_U32_e64_4:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_5:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 %{{[0-9]+}}, [[S_MOV_B32_1]]114# GFX9: [[BASE_HI:%[0-9]+]]:vgpr_32, dead %{{[0-9]+}}:sreg_64_xexec = V_ADDC_U32_e64 %{{[0-9]+}}, 0, killed [[V_ADD_CO_U32_e64_5]]115# GFX9: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[V_ADD_CO_U32_e64_4]], %subreg.sub0, [[BASE_HI]], %subreg.sub1116# GFX9: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE2]], -4096, 0117# GFX9: [[GLOBAL_LOAD_DWORDX2_1:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE2]], -2048, 0118# GFX9: [[GLOBAL_LOAD_DWORDX2_2:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE2]], 0, 0119 120name: NegativeDistance121body: |122 bb.0.entry:123 %0:sgpr_64 = COPY $sgpr0_sgpr1124 %1:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0, 36, 0125 %3:sgpr_128 = COPY $sgpr96_sgpr97_sgpr98_sgpr99126 %4:sreg_32_xm0 = COPY $sgpr10127 %5:sreg_32_xm0 = S_MOV_B32 0128 $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3129 $sgpr4 = COPY %4130 $vgpr0 = V_MOV_B32_e32 0, implicit $exec131 %6:vreg_64_align2 = COPY $vgpr0_vgpr1132 %7:vgpr_32 = V_AND_B32_e32 255, %6.sub0, implicit $exec133 %8:vgpr_32 = V_MOV_B32_e32 0, implicit $exec134 %9:vreg_64_align2 = REG_SEQUENCE killed %7, %subreg.sub0, %8, %subreg.sub1135 %10:vgpr_32 = V_LSHLREV_B32_e64 7, %6.sub0, implicit $exec136 %11:vgpr_32 = V_AND_B32_e32 -32768, killed %10, implicit $exec137 %12:sgpr_32 = COPY %1.sub1138 %13:vgpr_32 = COPY %5139 %14:vgpr_32, %15:sreg_64_xexec = V_ADD_CO_U32_e64 %1.sub0, %11, 0, implicit $exec140 %16:vgpr_32 = COPY %12141 %17:vgpr_32, dead %18:sreg_64_xexec = V_ADDC_U32_e64 %16, %13, killed %15, 0, implicit $exec142 %19:vreg_64_align2 = REG_SEQUENCE %14, %subreg.sub0, %17, %subreg.sub1143 %20:vreg_64_align2 = V_LSHLREV_B64_e64 3, %9, implicit $exec144 %21:vgpr_32, %22:sreg_64_xexec = V_ADD_CO_U32_e64 %14, %20.sub0, 0, implicit $exec145 %23:vgpr_32, dead %24:sreg_64_xexec = V_ADDC_U32_e64 %17, %20.sub1, killed %22, 0, implicit $exec146 %25:sgpr_32 = S_MOV_B32 6144147 %26:vgpr_32, %27:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %25, 0, implicit $exec148 %28:vgpr_32, dead %29:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %27, 0, implicit $exec149 %30:vreg_64_align2 = REG_SEQUENCE %26, %subreg.sub0, %28, %subreg.sub1150 %31:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %30, 0, 0, implicit $exec151 %32:sgpr_32 = S_MOV_B32 8192152 %33:vgpr_32, %34:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %32, 0, implicit $exec153 %35:vgpr_32, dead %36:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %34, 0, implicit $exec154 %37:vreg_64_align2 = REG_SEQUENCE %33, %subreg.sub0, %35, %subreg.sub1155 %38:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %37, 0, 0, implicit $exec156 %39:sgpr_32 = S_MOV_B32 10240157 %40:vgpr_32, %41:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %39, 0, implicit $exec158 %42:vgpr_32, dead %43:sreg_64_xexec = V_ADDC_U32_e64 %23, 0, killed %41, 0, implicit $exec159 %44:vreg_64_align2 = REG_SEQUENCE %40, %subreg.sub0, %42, %subreg.sub1160 %45:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %44, 0, 0, implicit $exec161...162---163 164# Tests for a successful compilation.165name: assert_hit166body: |167 bb.0.entry:168 %0:sgpr_64 = COPY $sgpr0_sgpr1169 %1:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0, 36, 0170 %3:sgpr_128 = COPY $sgpr96_sgpr97_sgpr98_sgpr99171 %4:sreg_32_xm0 = COPY $sgpr101172 %5:sreg_32_xm0 = S_MOV_B32 0173 $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3174 $sgpr4 = COPY %4175 $vgpr0 = V_MOV_B32_e32 0, implicit $exec176 %6:vreg_64_align2 = COPY $vgpr0_vgpr1177 %7:vgpr_32 = V_AND_B32_e32 255, %6.sub0, implicit $exec178 %8:vgpr_32 = V_MOV_B32_e32 0, implicit $exec179 %9:vreg_64_align2 = REG_SEQUENCE killed %7, %subreg.sub0, %8, %subreg.sub1180 %10:vgpr_32 = V_LSHLREV_B32_e64 7, %6.sub0, implicit $exec181 %11:vgpr_32 = V_AND_B32_e32 -32768, killed %10, implicit $exec182 %12:sgpr_32 = COPY %1.sub1183 %13:vgpr_32 = COPY %5184 %14:vgpr_32, %15:sreg_64_xexec = V_ADD_CO_U32_e64 %1.sub0, %11, 0, implicit $exec185 %16:vgpr_32 = COPY %12186 %17:vgpr_32, dead %18:sreg_64_xexec = V_ADDC_U32_e64 %16, %13, killed %15, 0, implicit $exec187 %19:vreg_64_align2 = REG_SEQUENCE %14, %subreg.sub0, %17, %subreg.sub1188 %20:vreg_64_align2 = V_LSHLREV_B64_e64 3, %9, implicit $exec189 %21:vgpr_32, %22:sreg_64_xexec = V_ADD_CO_U32_e64 %14, %20.sub0, 0, implicit $exec190 %23:vgpr_32, dead %24:sreg_64_xexec = V_ADDC_U32_e64 %17, %20.sub1, killed %22, 0, implicit $exec191 192 %25:sgpr_32 = S_MOV_B32 6144193 %26:vgpr_32, %27:sreg_64_xexec = V_ADD_CO_U32_e64 %21, %25, 0, implicit $exec194 %28:vgpr_32, dead %29:sreg_64_xexec = V_ADDC_U32_e64 %23, 4294967295, killed %27, 0, implicit $exec195 %30:vreg_64_align2 = REG_SEQUENCE %26, %subreg.sub0, %28, %subreg.sub1196 %31:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 %30, 0, 0, implicit $exec197...198---199 200# GFX9-LABEL: name: diffoporder_add_store201# GFX9: GLOBAL_STORE_DWORD %{{[0-9]+}}, %0.sub0, 1000, 0,202# GFX9: GLOBAL_STORE_DWORD %{{[0-9]+}}, %0.sub1, 0, 0,203 204name: diffoporder_add_store205body: |206 bb.0.entry:207 208 %0:vreg_64_align2 = COPY $vgpr0_vgpr1209 210 %1:sgpr_32 = S_MOV_B32 4000211 %2:vgpr_32, %3:sreg_64_xexec = V_ADD_CO_U32_e64 %0.sub0, %1, 0, implicit $exec212 %4:vgpr_32, dead %5:sreg_64_xexec = V_ADDC_U32_e64 %0.sub1, 0, %3, 0, implicit $exec213 %6:vreg_64_align2 = REG_SEQUENCE %2, %subreg.sub0, %4, %subreg.sub1214 GLOBAL_STORE_DWORD %6, %0.sub0, 0, 0, implicit $exec215 216 %8:sgpr_32 = S_MOV_B32 3000217 %9:vgpr_32, %10:sreg_64_xexec = V_ADD_CO_U32_e64 %0.sub0, %8, 0, implicit $exec218 %11:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 %0.sub1, 0, %10, 0, implicit $exec219 %13:vreg_64_align2 = REG_SEQUENCE %9, %subreg.sub0, %11, %subreg.sub1220 GLOBAL_STORE_DWORD %13, %0.sub1, 0, 0, implicit $exec221...222