126 lines · plain
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py2# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -verify-machineinstrs -run-pass=regbankselect -o - %s | FileCheck %s3 4# Make sure that an arbitrary AGPR is treated as a divergent value5# that needs to be copied to VGPR, and then waterfalled6 7# 32-bit case8---9name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__agpr_soffset10legalized: true11tracksRegLiveness: true12body: |13 bb.0:14 liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $vgpr0, $vgpr1, $agpr015 16 ; CHECK-LABEL: name: raw_buffer_store__sgpr_rsrc__vgpr_val__vgpr_voffset__agpr_soffset17 ; CHECK: successors: %bb.1(0x80000000)18 ; CHECK-NEXT: liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $vgpr0, $vgpr1, $agpr019 ; CHECK-NEXT: {{ $}}20 ; CHECK-NEXT: %val:vgpr(s32) = COPY $vgpr021 ; CHECK-NEXT: %rsrc:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr322 ; CHECK-NEXT: %agpr:agpr(s32) = COPY $agpr023 ; CHECK-NEXT: %voffset:vgpr(s32) = COPY $vgpr124 ; CHECK-NEXT: %zero:sgpr(s32) = G_CONSTANT i32 025 ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY %zero(s32)26 ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF27 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY %agpr(s32)28 ; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec29 ; CHECK-NEXT: {{ $}}30 ; CHECK-NEXT: .1:31 ; CHECK-NEXT: successors: %bb.2(0x80000000)32 ; CHECK-NEXT: {{ $}}33 ; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF]], %bb.0, %9, %bb.234 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[COPY1]](s32), implicit $exec35 ; CHECK-NEXT: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[V_READFIRSTLANE_B32_]](s32), [[COPY1]]36 ; CHECK-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_64_xexec(s64) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.ballot), [[ICMP]](s1)37 ; CHECK-NEXT: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[INTRINSIC_CONVERGENT]](s64), implicit-def $exec, implicit-def $scc, implicit $exec38 ; CHECK-NEXT: {{ $}}39 ; CHECK-NEXT: .2:40 ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.1(0x40000000)41 ; CHECK-NEXT: {{ $}}42 ; CHECK-NEXT: G_AMDGPU_BUFFER_STORE %val(s32), %rsrc(<4 x s32>), [[COPY]](s32), %voffset, [[V_READFIRSTLANE_B32_]], 0, 0, 0 :: (dereferenceable store (s32), addrspace 4)43 ; CHECK-NEXT: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc44 ; CHECK-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec45 ; CHECK-NEXT: {{ $}}46 ; CHECK-NEXT: .3:47 ; CHECK-NEXT: successors: %bb.4(0x80000000)48 ; CHECK-NEXT: {{ $}}49 ; CHECK-NEXT: $exec = S_MOV_B64_term [[S_MOV_B64_]]50 ; CHECK-NEXT: {{ $}}51 ; CHECK-NEXT: .4:52 ; CHECK-NEXT: S_ENDPGM 053 %val:_(s32) = COPY $vgpr054 %rsrc:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr355 %agpr:_(s32) = COPY $agpr056 %voffset:_(s32) = COPY $vgpr157 %zero:_(s32) = G_CONSTANT i32 058 G_AMDGPU_BUFFER_STORE %val, %rsrc, %zero, %voffset, %agpr, 0, 0, 0 :: (dereferenceable store (s32), addrspace 4)59 S_ENDPGM 060 61...62 63# Register tuple case64---65name: load_1d_vgpr_vaddr__agpr_srsrc66legalized: true67tracksRegLiveness: true68body: |69 bb.0:70 liveins: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, $vgpr071 ; CHECK-LABEL: name: load_1d_vgpr_vaddr__agpr_srsrc72 ; CHECK: successors: %bb.1(0x80000000)73 ; CHECK-NEXT: liveins: $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7, $vgpr074 ; CHECK-NEXT: {{ $}}75 ; CHECK-NEXT: [[COPY:%[0-9]+]]:agpr(<8 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr776 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr077 ; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF78 ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr(<8 x s32>) = COPY [[COPY]](<8 x s32>)79 ; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64_xexec = S_MOV_B64 $exec80 ; CHECK-NEXT: {{ $}}81 ; CHECK-NEXT: .1:82 ; CHECK-NEXT: successors: %bb.2(0x80000000)83 ; CHECK-NEXT: {{ $}}84 ; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF]], %bb.0, %6, %bb.285 ; CHECK-NEXT: [[UV:%[0-9]+]]:vgpr_32(s32), [[UV1:%[0-9]+]]:vgpr_32(s32), [[UV2:%[0-9]+]]:vgpr_32(s32), [[UV3:%[0-9]+]]:vgpr_32(s32), [[UV4:%[0-9]+]]:vgpr_32(s32), [[UV5:%[0-9]+]]:vgpr_32(s32), [[UV6:%[0-9]+]]:vgpr_32(s32), [[UV7:%[0-9]+]]:vgpr_32(s32) = G_UNMERGE_VALUES [[COPY2]](<8 x s32>)86 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]](s32), implicit $exec87 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]](s32), implicit $exec88 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]](s32), implicit $exec89 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]](s32), implicit $exec90 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]](s32), implicit $exec91 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]](s32), implicit $exec92 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV6]](s32), implicit $exec93 ; CHECK-NEXT: [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV7]](s32), implicit $exec94 ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)95 ; CHECK-NEXT: [[UV8:%[0-9]+]]:vgpr(s64), [[UV9:%[0-9]+]]:vgpr(s64), [[UV10:%[0-9]+]]:vgpr(s64), [[UV11:%[0-9]+]]:vgpr(s64) = G_UNMERGE_VALUES [[COPY2]](<8 x s32>)96 ; CHECK-NEXT: [[UV12:%[0-9]+]]:sgpr(s64), [[UV13:%[0-9]+]]:sgpr(s64), [[UV14:%[0-9]+]]:sgpr(s64), [[UV15:%[0-9]+]]:sgpr(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)97 ; CHECK-NEXT: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[UV12]](s64), [[UV8]]98 ; CHECK-NEXT: [[ICMP1:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[UV13]](s64), [[UV9]]99 ; CHECK-NEXT: [[AND:%[0-9]+]]:vcc(s1) = G_AND [[ICMP]], [[ICMP1]]100 ; CHECK-NEXT: [[ICMP2:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[UV14]](s64), [[UV10]]101 ; CHECK-NEXT: [[AND1:%[0-9]+]]:vcc(s1) = G_AND [[AND]], [[ICMP2]]102 ; CHECK-NEXT: [[ICMP3:%[0-9]+]]:vcc(s1) = G_ICMP intpred(eq), [[UV15]](s64), [[UV11]]103 ; CHECK-NEXT: [[AND2:%[0-9]+]]:vcc(s1) = G_AND [[AND1]], [[ICMP3]]104 ; CHECK-NEXT: [[INTRINSIC_CONVERGENT:%[0-9]+]]:sreg_64_xexec(s64) = G_INTRINSIC_CONVERGENT intrinsic(@llvm.amdgcn.ballot), [[AND2]](s1)105 ; CHECK-NEXT: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[INTRINSIC_CONVERGENT]](s64), implicit-def $exec, implicit-def $scc, implicit $exec106 ; CHECK-NEXT: {{ $}}107 ; CHECK-NEXT: .2:108 ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.1(0x40000000)109 ; CHECK-NEXT: {{ $}}110 ; CHECK-NEXT: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY1]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load (<4 x s32>))111 ; CHECK-NEXT: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc112 ; CHECK-NEXT: SI_WATERFALL_LOOP %bb.1, implicit $exec113 ; CHECK-NEXT: {{ $}}114 ; CHECK-NEXT: .3:115 ; CHECK-NEXT: successors: %bb.4(0x80000000)116 ; CHECK-NEXT: {{ $}}117 ; CHECK-NEXT: $exec = S_MOV_B64_term [[S_MOV_B64_]]118 ; CHECK-NEXT: {{ $}}119 ; CHECK-NEXT: .4:120 ; CHECK-NEXT: S_ENDPGM 0, implicit [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)121 %0:_(<8 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7122 %1:_(s32) = COPY $vgpr0123 %2:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, %1(s32), %0(<8 x s32>), 0, 0, 0 :: (dereferenceable load (<4 x s32>))124 S_ENDPGM 0, implicit %2125...126