165 lines · cpp
1// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+"2// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-offload-mandatory -emit-llvm %s -o - | FileCheck %s --check-prefix=MANDATORY3// expected-no-diagnostics4 5void foo() {}6#pragma omp declare target(foo)7 8void bar() {}9#pragma omp declare target device_type(nohost) to(bar)10 11void host() {12#pragma omp target13 { bar(); }14}15 16void host_if(bool cond) {17#pragma omp target if(cond)18 { bar(); }19}20 21void host_dev(int device) {22#pragma omp target device(device)23 { bar(); }24}25// MANDATORY-LABEL: define {{[^@]+}}@_Z3foov26// MANDATORY-SAME: () #[[ATTR0:[0-9]+]] {27// MANDATORY-NEXT: entry:28// MANDATORY-NEXT: ret void29//30//31// MANDATORY-LABEL: define {{[^@]+}}@_Z4hostv32// MANDATORY-SAME: () #[[ATTR0]] {33// MANDATORY-NEXT: entry:34// MANDATORY-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 835// MANDATORY-NEXT: [[TMP0:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 036// MANDATORY-NEXT: store i32 3, ptr [[TMP0]], align 437// MANDATORY-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 138// MANDATORY-NEXT: store i32 0, ptr [[TMP1]], align 439// MANDATORY-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 240// MANDATORY-NEXT: store ptr null, ptr [[TMP2]], align 841// MANDATORY-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 342// MANDATORY-NEXT: store ptr null, ptr [[TMP3]], align 843// MANDATORY-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 444// MANDATORY-NEXT: store ptr null, ptr [[TMP4]], align 845// MANDATORY-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 546// MANDATORY-NEXT: store ptr null, ptr [[TMP5]], align 847// MANDATORY-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 648// MANDATORY-NEXT: store ptr null, ptr [[TMP6]], align 849// MANDATORY-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 750// MANDATORY-NEXT: store ptr null, ptr [[TMP7]], align 851// MANDATORY-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 852// MANDATORY-NEXT: store i64 0, ptr [[TMP8]], align 853// MANDATORY-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 954// MANDATORY-NEXT: store i64 0, ptr [[TMP9]], align 855// MANDATORY-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1056// MANDATORY-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP10]], align 457// MANDATORY-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1158// MANDATORY-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP11]], align 459// MANDATORY-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1260// MANDATORY-NEXT: store i32 0, ptr [[TMP12]], align 461// MANDATORY-NEXT: [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4hostv_l12.region_id, ptr [[KERNEL_ARGS]])62// MANDATORY-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 063// MANDATORY-NEXT: br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]64// MANDATORY: omp_offload.failed:65// MANDATORY-NEXT: unreachable66// MANDATORY: omp_offload.cont:67// MANDATORY-NEXT: ret void68//69//70// MANDATORY-LABEL: define {{[^@]+}}@_Z7host_ifb71// MANDATORY-SAME: (i1 noundef zeroext [[COND:%.*]]) #[[ATTR0]] {72// MANDATORY-NEXT: entry:73// MANDATORY-NEXT: [[COND_ADDR:%.*]] = alloca i8, align 174// MANDATORY-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 875// MANDATORY-NEXT: [[STOREDV:%.*]] = zext i1 [[COND]] to i876// MANDATORY-NEXT: store i8 [[STOREDV]], ptr [[COND_ADDR]], align 177// MANDATORY-NEXT: [[TMP0:%.*]] = load i8, ptr [[COND_ADDR]], align 178// MANDATORY-NEXT: [[LOADEDV:%.*]] = trunc i8 [[TMP0]] to i179// MANDATORY-NEXT: br i1 [[LOADEDV]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]80// MANDATORY: omp_if.then:81// MANDATORY-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 082// MANDATORY-NEXT: store i32 3, ptr [[TMP1]], align 483// MANDATORY-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 184// MANDATORY-NEXT: store i32 0, ptr [[TMP2]], align 485// MANDATORY-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 286// MANDATORY-NEXT: store ptr null, ptr [[TMP3]], align 887// MANDATORY-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 388// MANDATORY-NEXT: store ptr null, ptr [[TMP4]], align 889// MANDATORY-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 490// MANDATORY-NEXT: store ptr null, ptr [[TMP5]], align 891// MANDATORY-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 592// MANDATORY-NEXT: store ptr null, ptr [[TMP6]], align 893// MANDATORY-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 694// MANDATORY-NEXT: store ptr null, ptr [[TMP7]], align 895// MANDATORY-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 796// MANDATORY-NEXT: store ptr null, ptr [[TMP8]], align 897// MANDATORY-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 898// MANDATORY-NEXT: store i64 0, ptr [[TMP9]], align 899// MANDATORY-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9100// MANDATORY-NEXT: store i64 0, ptr [[TMP10]], align 8101// MANDATORY-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10102// MANDATORY-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP11]], align 4103// MANDATORY-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11104// MANDATORY-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP12]], align 4105// MANDATORY-NEXT: [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12106// MANDATORY-NEXT: store i32 0, ptr [[TMP13]], align 4107// MANDATORY-NEXT: [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7host_ifb_l17.region_id, ptr [[KERNEL_ARGS]])108// MANDATORY-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0109// MANDATORY-NEXT: br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]110// MANDATORY: omp_offload.failed:111// MANDATORY-NEXT: unreachable112// MANDATORY: omp_offload.cont:113// MANDATORY-NEXT: br label [[OMP_IF_END:%.*]]114// MANDATORY: omp_if.else:115// MANDATORY-NEXT: unreachable116// MANDATORY: omp_if.end:117// MANDATORY-NEXT: ret void118//119//120// MANDATORY-LABEL: define {{[^@]+}}@_Z8host_devi121// MANDATORY-SAME: (i32 noundef signext [[DEVICE:%.*]]) #[[ATTR0]] {122// MANDATORY-NEXT: entry:123// MANDATORY-NEXT: [[DEVICE_ADDR:%.*]] = alloca i32, align 4124// MANDATORY-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4125// MANDATORY-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8126// MANDATORY-NEXT: store i32 [[DEVICE]], ptr [[DEVICE_ADDR]], align 4127// MANDATORY-NEXT: [[TMP0:%.*]] = load i32, ptr [[DEVICE_ADDR]], align 4128// MANDATORY-NEXT: store i32 [[TMP0]], ptr [[DOTCAPTURE_EXPR_]], align 4129// MANDATORY-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4130// MANDATORY-NEXT: [[TMP2:%.*]] = sext i32 [[TMP1]] to i64131// MANDATORY-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0132// MANDATORY-NEXT: store i32 3, ptr [[TMP3]], align 4133// MANDATORY-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1134// MANDATORY-NEXT: store i32 0, ptr [[TMP4]], align 4135// MANDATORY-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2136// MANDATORY-NEXT: store ptr null, ptr [[TMP5]], align 8137// MANDATORY-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3138// MANDATORY-NEXT: store ptr null, ptr [[TMP6]], align 8139// MANDATORY-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4140// MANDATORY-NEXT: store ptr null, ptr [[TMP7]], align 8141// MANDATORY-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5142// MANDATORY-NEXT: store ptr null, ptr [[TMP8]], align 8143// MANDATORY-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6144// MANDATORY-NEXT: store ptr null, ptr [[TMP9]], align 8145// MANDATORY-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7146// MANDATORY-NEXT: store ptr null, ptr [[TMP10]], align 8147// MANDATORY-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8148// MANDATORY-NEXT: store i64 0, ptr [[TMP11]], align 8149// MANDATORY-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9150// MANDATORY-NEXT: store i64 0, ptr [[TMP12]], align 8151// MANDATORY-NEXT: [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10152// MANDATORY-NEXT: store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP13]], align 4153// MANDATORY-NEXT: [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11154// MANDATORY-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP14]], align 4155// MANDATORY-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12156// MANDATORY-NEXT: store i32 0, ptr [[TMP15]], align 4157// MANDATORY-NEXT: [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 [[TMP2]], i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8host_devi_l22.region_id, ptr [[KERNEL_ARGS]])158// MANDATORY-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0159// MANDATORY-NEXT: br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]160// MANDATORY: omp_offload.failed:161// MANDATORY-NEXT: unreachable162// MANDATORY: omp_offload.cont:163// MANDATORY-NEXT: ret void164//165