160 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 42; RUN: opt < %s -passes=amdgpu-sw-lower-lds -S -mtriple=amdgcn-amd-amdhsa | FileCheck %s3 4; Test to check if static LDS accesses in kernel are lowered correctly.5@lds_1 = internal addrspace(3) global [1 x i8] poison, align 46@lds_2 = internal addrspace(3) global [1 x i32] poison, align 87 8;.9; CHECK: @llvm.amdgcn.sw.lds.k0 = internal addrspace(3) global ptr poison, no_sanitize_address, align 8, !absolute_symbol [[META0:![0-9]+]]10; CHECK: @llvm.amdgcn.sw.lds.k0.md = internal addrspace(1) global %llvm.amdgcn.sw.lds.k0.md.type { %llvm.amdgcn.sw.lds.k0.md.item { i32 0, i32 8, i32 32 }, %llvm.amdgcn.sw.lds.k0.md.item { i32 32, i32 1, i32 32 }, %llvm.amdgcn.sw.lds.k0.md.item { i32 64, i32 4, i32 32 } }, no_sanitize_address11;.12define amdgpu_kernel void @k0() sanitize_address {13; CHECK-LABEL: define amdgpu_kernel void @k0(14; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {15; CHECK-NEXT: WId:16; CHECK-NEXT: [[TMP0:%.*]] = call i32 @llvm.amdgcn.workitem.id.x()17; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.amdgcn.workitem.id.y()18; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.amdgcn.workitem.id.z()19; CHECK-NEXT: [[TMP3:%.*]] = or i32 [[TMP0]], [[TMP1]]20; CHECK-NEXT: [[TMP4:%.*]] = or i32 [[TMP3]], [[TMP2]]21; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i32 [[TMP4]], 022; CHECK-NEXT: br i1 [[TMP5]], label [[MALLOC:%.*]], label [[TMP7:%.*]]23; CHECK: Malloc:24; CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE:%.*]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 2, i32 0), align 425; CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 2, i32 2), align 426; CHECK-NEXT: [[TMP16:%.*]] = add i32 [[TMP13]], [[TMP14]]27; CHECK-NEXT: [[TMP15:%.*]] = zext i32 [[TMP16]] to i6428; CHECK-NEXT: [[TMP23:%.*]] = call ptr @llvm.returnaddress(i32 0)29; CHECK-NEXT: [[TMP11:%.*]] = ptrtoint ptr [[TMP23]] to i6430; CHECK-NEXT: [[TMP12:%.*]] = call i64 @__asan_malloc_impl(i64 [[TMP15]], i64 [[TMP11]])31; CHECK-NEXT: [[TMP6:%.*]] = inttoptr i64 [[TMP12]] to ptr addrspace(1)32; CHECK-NEXT: store ptr addrspace(1) [[TMP6]], ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, align 833; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP6]], i64 834; CHECK-NEXT: [[TMP41:%.*]] = ptrtoint ptr addrspace(1) [[TMP25]] to i6435; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP41]], i64 24)36; CHECK-NEXT: [[TMP61:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP6]], i64 3337; CHECK-NEXT: [[TMP62:%.*]] = ptrtoint ptr addrspace(1) [[TMP61]] to i6438; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP62]], i64 31)39; CHECK-NEXT: [[TMP63:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP6]], i64 6840; CHECK-NEXT: [[TMP64:%.*]] = ptrtoint ptr addrspace(1) [[TMP63]] to i6441; CHECK-NEXT: call void @__asan_poison_region(i64 [[TMP64]], i64 28)42; CHECK-NEXT: br label [[TMP7]]43; CHECK: 20:44; CHECK-NEXT: [[XYZCOND:%.*]] = phi i1 [ false, [[WID:%.*]] ], [ true, [[MALLOC]] ]45; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()46; CHECK-NEXT: [[TMP19:%.*]] = load ptr addrspace(1), ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, align 847; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 1, i32 0), align 448; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, i32 [[TMP10]]49; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds ([[LLVM_AMDGCN_SW_LDS_K0_MD_TYPE]], ptr addrspace(1) @llvm.amdgcn.sw.lds.k0.md, i32 0, i32 2, i32 0), align 450; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i8, ptr addrspace(3) @llvm.amdgcn.sw.lds.k0, i32 [[TMP17]]51; CHECK-NEXT: [[TMP26:%.*]] = ptrtoint ptr addrspace(3) [[TMP18]] to i3252; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP19]], i32 [[TMP26]]53; CHECK-NEXT: [[TMP28:%.*]] = ptrtoint ptr addrspace(1) [[TMP27]] to i6454; CHECK-NEXT: [[TMP29:%.*]] = lshr i64 [[TMP28]], 355; CHECK-NEXT: [[TMP30:%.*]] = add i64 [[TMP29]], 214745088056; CHECK-NEXT: [[TMP31:%.*]] = inttoptr i64 [[TMP30]] to ptr57; CHECK-NEXT: [[TMP32:%.*]] = load i8, ptr [[TMP31]], align 158; CHECK-NEXT: [[TMP33:%.*]] = icmp ne i8 [[TMP32]], 059; CHECK-NEXT: [[TMP34:%.*]] = and i64 [[TMP28]], 760; CHECK-NEXT: [[TMP35:%.*]] = trunc i64 [[TMP34]] to i861; CHECK-NEXT: [[TMP36:%.*]] = icmp sge i8 [[TMP35]], [[TMP32]]62; CHECK-NEXT: [[TMP37:%.*]] = and i1 [[TMP33]], [[TMP36]]63; CHECK-NEXT: [[TMP38:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 [[TMP37]])64; CHECK-NEXT: [[TMP39:%.*]] = icmp ne i64 [[TMP38]], 065; CHECK-NEXT: br i1 [[TMP39]], label [[ASAN_REPORT:%.*]], label [[TMP42:%.*]], !prof [[PROF2:![0-9]+]]66; CHECK: asan.report:67; CHECK-NEXT: br i1 [[TMP37]], label [[TMP40:%.*]], label [[CONDFREE:%.*]]68; CHECK: 40:69; CHECK-NEXT: call void @__asan_report_store1(i64 [[TMP28]]) #[[ATTR6:[0-9]+]]70; CHECK-NEXT: call void @llvm.amdgcn.unreachable()71; CHECK-NEXT: br label [[CONDFREE]]72; CHECK: 41:73; CHECK-NEXT: br label [[TMP42]]74; CHECK: 42:75; CHECK-NEXT: store i8 7, ptr addrspace(1) [[TMP27]], align 476; CHECK-NEXT: [[TMP43:%.*]] = ptrtoint ptr addrspace(3) [[TMP24]] to i3277; CHECK-NEXT: [[TMP44:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[TMP19]], i32 [[TMP43]]78; CHECK-NEXT: [[TMP45:%.*]] = ptrtoint ptr addrspace(1) [[TMP44]] to i6479; CHECK-NEXT: [[TMP51:%.*]] = add i64 [[TMP45]], 380; CHECK-NEXT: [[TMP78:%.*]] = inttoptr i64 [[TMP51]] to ptr addrspace(1)81; CHECK-NEXT: [[TMP79:%.*]] = ptrtoint ptr addrspace(1) [[TMP44]] to i6482; CHECK-NEXT: [[TMP46:%.*]] = lshr i64 [[TMP79]], 383; CHECK-NEXT: [[TMP47:%.*]] = add i64 [[TMP46]], 214745088084; CHECK-NEXT: [[TMP48:%.*]] = inttoptr i64 [[TMP47]] to ptr85; CHECK-NEXT: [[TMP49:%.*]] = load i8, ptr [[TMP48]], align 186; CHECK-NEXT: [[TMP50:%.*]] = icmp ne i8 [[TMP49]], 087; CHECK-NEXT: [[TMP52:%.*]] = and i64 [[TMP79]], 788; CHECK-NEXT: [[TMP53:%.*]] = trunc i64 [[TMP52]] to i889; CHECK-NEXT: [[TMP54:%.*]] = icmp sge i8 [[TMP53]], [[TMP49]]90; CHECK-NEXT: [[TMP55:%.*]] = and i1 [[TMP50]], [[TMP54]]91; CHECK-NEXT: [[TMP56:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 [[TMP55]])92; CHECK-NEXT: [[TMP57:%.*]] = icmp ne i64 [[TMP56]], 093; CHECK-NEXT: br i1 [[TMP57]], label [[ASAN_REPORT1:%.*]], label [[TMP60:%.*]], !prof [[PROF2]]94; CHECK: asan.report1:95; CHECK-NEXT: br i1 [[TMP55]], label [[TMP58:%.*]], label [[TMP59:%.*]]96; CHECK: 60:97; CHECK-NEXT: call void @__asan_report_store1(i64 [[TMP79]]) #[[ATTR6]]98; CHECK-NEXT: call void @llvm.amdgcn.unreachable()99; CHECK-NEXT: br label [[TMP59]]100; CHECK: 61:101; CHECK-NEXT: br label [[TMP60]]102; CHECK: 62:103; CHECK-NEXT: [[TMP80:%.*]] = ptrtoint ptr addrspace(1) [[TMP78]] to i64104; CHECK-NEXT: [[TMP81:%.*]] = lshr i64 [[TMP80]], 3105; CHECK-NEXT: [[TMP65:%.*]] = add i64 [[TMP81]], 2147450880106; CHECK-NEXT: [[TMP66:%.*]] = inttoptr i64 [[TMP65]] to ptr107; CHECK-NEXT: [[TMP67:%.*]] = load i8, ptr [[TMP66]], align 1108; CHECK-NEXT: [[TMP68:%.*]] = icmp ne i8 [[TMP67]], 0109; CHECK-NEXT: [[TMP69:%.*]] = and i64 [[TMP80]], 7110; CHECK-NEXT: [[TMP70:%.*]] = trunc i64 [[TMP69]] to i8111; CHECK-NEXT: [[TMP71:%.*]] = icmp sge i8 [[TMP70]], [[TMP67]]112; CHECK-NEXT: [[TMP72:%.*]] = and i1 [[TMP68]], [[TMP71]]113; CHECK-NEXT: [[TMP73:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 [[TMP72]])114; CHECK-NEXT: [[TMP74:%.*]] = icmp ne i64 [[TMP73]], 0115; CHECK-NEXT: br i1 [[TMP74]], label [[ASAN_REPORT2:%.*]], label [[TMP77:%.*]], !prof [[PROF2]]116; CHECK: asan.report2:117; CHECK-NEXT: br i1 [[TMP72]], label [[TMP75:%.*]], label [[TMP76:%.*]]118; CHECK: 75:119; CHECK-NEXT: call void @__asan_report_store1(i64 [[TMP80]]) #[[ATTR6]]120; CHECK-NEXT: call void @llvm.amdgcn.unreachable()121; CHECK-NEXT: br label [[TMP76]]122; CHECK: 76:123; CHECK-NEXT: br label [[TMP77]]124; CHECK: 77:125; CHECK-NEXT: store i32 8, ptr addrspace(1) [[TMP44]], align 2126; CHECK-NEXT: br label [[CONDFREE1:%.*]]127; CHECK: CondFree:128; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()129; CHECK-NEXT: br i1 [[XYZCOND]], label [[FREE:%.*]], label [[END:%.*]]130; CHECK: Free:131; CHECK-NEXT: [[TMP20:%.*]] = call ptr @llvm.returnaddress(i32 0)132; CHECK-NEXT: [[TMP21:%.*]] = ptrtoint ptr [[TMP20]] to i64133; CHECK-NEXT: [[TMP22:%.*]] = ptrtoint ptr addrspace(1) [[TMP19]] to i64134; CHECK-NEXT: call void @__asan_free_impl(i64 [[TMP22]], i64 [[TMP21]])135; CHECK-NEXT: br label [[END]]136; CHECK: End:137; CHECK-NEXT: ret void138;139 store i8 7, ptr addrspace(3) @lds_1, align 4140 store i32 8, ptr addrspace(3) @lds_2, align 2141 ret void142}143 144!llvm.module.flags = !{!0}145!0 = !{i32 4, !"nosanitize_address", i32 1}146 147;.148; CHECK: attributes #[[ATTR0]] = { sanitize_address "amdgpu-lds-size"="8" }149; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }150; CHECK: attributes #[[ATTR2:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) }151; CHECK: attributes #[[ATTR3:[0-9]+]] = { convergent nocallback nofree nounwind willreturn }152; CHECK: attributes #[[ATTR4:[0-9]+]] = { convergent nocallback nofree nounwind willreturn memory(none) }153; CHECK: attributes #[[ATTR5:[0-9]+]] = { convergent nocallback nofree nounwind }154; CHECK: attributes #[[ATTR6]] = { nomerge }155;.156; CHECK: [[META0]] = !{i32 0, i32 1}157; CHECK: [[META1:![0-9]+]] = !{i32 4, !"nosanitize_address", i32 1}158; CHECK: [[PROF2]] = !{!"branch_weights", i32 1, i32 1048575}159;.160