1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals 2; RUN: opt -S -passes=openmp-opt < %s | FileCheck %s 3 4; void foo(double x) { 5; #pragma omp target map(from:x) 6; { 7; x = sin(M_PI); 8; #pragma omp parallel 9; { } 10; } 11; } 12 13target triple = "nvptx64" 14 15%struct.ident_t = type { i32, i32, i32, i32, ptr } 16%struct.KernelEnvironmentTy = type { %struct.ConfigurationEnvironmentTy, ptr, ptr } 17%struct.ConfigurationEnvironmentTy = type { i8, i8, i8, i32, i32, i32, i32, i32, i32 } 18 19@0 = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1 20@1 = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, ptr @0 }, align 8 21@__omp_offloading_fd02_404433c2_main_l5_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 0, i8 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @1, ptr null } 22 23 24; Function Attrs: alwaysinline convergent norecurse nounwind 25;. 26; CHECK: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c" 27; CHECK: @[[GLOB1:[0-9]+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, ptr @[[GLOB0]] }, align 8 28; CHECK: @__omp_offloading_fd02_404433c2_main_l5_kernel_environment = local_unnamed_addr constant %struct.KernelEnvironmentTy { %struct.ConfigurationEnvironmentTy { i8 0, i8 0, i8 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0 }, ptr @[[GLOB1]], ptr null } 29; CHECK: @[[GLOB2:[0-9]+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 22, ptr @[[GLOB0]] }, align 8 30;. 31define weak ptx_kernel void @__omp_offloading_fd02_404433c2_main_l5(ptr %dyn, ptr nonnull align 8 dereferenceable(8) %x) local_unnamed_addr #0 { 32; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_fd02_404433c2_main_l5 33; CHECK-SAME: (ptr [[DYN:%.*]], ptr nonnull align 8 dereferenceable(8) [[X:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { 34; CHECK-NEXT: entry: 35; CHECK-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 8 36; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr nonnull @__omp_offloading_fd02_404433c2_main_l5_kernel_environment, ptr [[DYN]]) #[[ATTR3:[0-9]+]] 37; CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1 38; CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[COMMON_RET:%.*]] 39; CHECK: common.ret: 40; CHECK-NEXT: ret void 41; CHECK: user_code.entry: 42; CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr nonnull @[[GLOB1]]) #[[ATTR3]] 43; CHECK-NEXT: [[CALL_I:%.*]] = call double @__nv_sin(double 0x400921FB54442D18) #[[ATTR7:[0-9]+]] 44; CHECK-NEXT: br label [[REGION_CHECK_TID:%.*]] 45; CHECK: region.check.tid: 46; CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block() 47; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i32 [[TMP2]], 0 48; CHECK-NEXT: br i1 [[TMP3]], label [[REGION_GUARDED:%.*]], label [[REGION_BARRIER:%.*]] 49; CHECK: region.guarded: 50; CHECK-NEXT: store double [[CALL_I]], ptr [[X]], align 8, !tbaa [[TBAA7:![0-9]+]] 51; CHECK-NEXT: br label [[REGION_GUARDED_END:%.*]] 52; CHECK: region.guarded.end: 53; CHECK-NEXT: br label [[REGION_BARRIER]] 54; CHECK: region.barrier: 55; CHECK-NEXT: call void @__kmpc_barrier_simple_spmd(ptr @[[GLOB2]], i32 [[TMP2]]) 56; CHECK-NEXT: br label [[REGION_EXIT:%.*]] 57; CHECK: region.exit: 58; CHECK-NEXT: call void @__kmpc_parallel_51(ptr nonnull @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr nonnull [[CAPTURED_VARS_ADDRS]], i64 0) #[[ATTR3]] 59; CHECK-NEXT: call void @__kmpc_target_deinit() #[[ATTR3]] 60; CHECK-NEXT: br label [[COMMON_RET]] 61; 62entry: 63 %captured_vars_addrs = alloca [0 x ptr], align 8 64 %0 = call i32 @__kmpc_target_init(ptr nonnull @__omp_offloading_fd02_404433c2_main_l5_kernel_environment, ptr %dyn) #3 65 %exec_user_code = icmp eq i32 %0, -1 66 br i1 %exec_user_code, label %user_code.entry, label %common.ret 67 68common.ret: ; preds = %entry, %user_code.entry 69 ret void 70 71user_code.entry: ; preds = %entry 72 %1 = call i32 @__kmpc_global_thread_num(ptr nonnull @1) 73 %call.i = call double @__nv_sin(double 0x400921FB54442D18) #6 74 store double %call.i, ptr %x, align 8, !tbaa !8 75 call void @__kmpc_parallel_51(ptr nonnull @1, i32 %1, i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr nonnull %captured_vars_addrs, i64 0) #3 76 call void @__kmpc_target_deinit() #3 77 br label %common.ret 78} 79 80declare i32 @__kmpc_target_init(ptr, ptr) local_unnamed_addr 81 82; Function Attrs: alwaysinline mustprogress nofree norecurse nosync nounwind readnone willreturn 83define internal void @__omp_outlined__(ptr noalias nocapture %.global_tid., ptr noalias nocapture %.bound_tid.) #1 { 84; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__ 85; CHECK-SAME: (ptr noalias captures(none) [[DOTGLOBAL_TID_:%.*]], ptr noalias captures(none) [[DOTBOUND_TID_:%.*]]) #[[ATTR1:[0-9]+]] { 86; CHECK-NEXT: entry: 87; CHECK-NEXT: ret void 88; 89entry: 90 ret void 91} 92 93; Function Attrs: norecurse nounwind 94define internal void @__omp_outlined___wrapper(i16 zeroext %0, i32 %1) #2 { 95; CHECK-LABEL: define {{[^@]+}}@__omp_outlined___wrapper 96; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR2:[0-9]+]] { 97; CHECK-NEXT: entry: 98; CHECK-NEXT: [[GLOBAL_ARGS:%.*]] = alloca ptr, align 8 99; CHECK-NEXT: call void @__kmpc_get_shared_variables(ptr nonnull [[GLOBAL_ARGS]]) #[[ATTR3]] 100; CHECK-NEXT: ret void 101; 102entry: 103 %global_args = alloca ptr, align 8 104 call void @__kmpc_get_shared_variables(ptr nonnull %global_args) #3 105 ret void 106} 107 108declare void @__kmpc_get_shared_variables(ptr) local_unnamed_addr 109 110; Function Attrs: nounwind 111declare i32 @__kmpc_global_thread_num(ptr) local_unnamed_addr #3 112 113; Function Attrs: alwaysinline 114declare void @__kmpc_parallel_51(ptr, i32, i32, i32, i32, ptr, ptr, ptr, i64) local_unnamed_addr #4 115 116declare void @__kmpc_target_deinit() local_unnamed_addr 117 118; Function Attrs: convergent 119declare double @__nv_sin(double) local_unnamed_addr #5 120 121attributes #0 = { alwaysinline convergent norecurse nounwind "kernel" "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 122attributes #1 = { alwaysinline mustprogress nofree norecurse nosync nounwind readnone willreturn "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 123attributes #2 = { norecurse nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 124attributes #3 = { nounwind } 125attributes #4 = { alwaysinline } 126attributes #5 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 127attributes #6 = { convergent nounwind "llvm.assume"="ompx_spmd_amenable" } 128 129!omp_offload.info = !{!0} 130!llvm.module.flags = !{!2, !3, !4, !5, !6} 131!llvm.ident = !{!7} 132 133!0 = !{i32 0, i32 64770, i32 1078211522, !"main", i32 5, i32 0} 134!2 = !{i32 1, !"wchar_size", i32 4} 135!3 = !{i32 7, !"openmp", i32 50} 136!4 = !{i32 7, !"openmp-device", i32 50} 137!5 = !{i32 8, !"PIC Level", i32 2} 138!6 = !{i32 7, !"frame-pointer", i32 2} 139!7 = !{!"clang version 14.0.0"} 140!8 = !{!9, !9, i64 0} 141!9 = !{!"double", !10, i64 0} 142!10 = !{!"omnipotent char", !11, i64 0} 143!11 = !{!"Simple C/C++ TBAA"} 144;. 145; CHECK: attributes #[[ATTR0]] = { alwaysinline convergent norecurse nounwind "frame-pointer"="all" "kernel" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 146; CHECK: attributes #[[ATTR1]] = { alwaysinline mustprogress nofree norecurse nosync nounwind willreturn memory(none) "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 147; CHECK: attributes #[[ATTR2]] = { norecurse nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 148; CHECK: attributes #[[ATTR3]] = { nounwind } 149; CHECK: attributes #[[ATTR4:[0-9]+]] = { alwaysinline } 150; CHECK: attributes #[[ATTR5:[0-9]+]] = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" } 151; CHECK: attributes #[[ATTR6:[0-9]+]] = { convergent nounwind } 152; CHECK: attributes #[[ATTR7]] = { convergent nounwind "llvm.assume"="ompx_spmd_amenable" } 153;. 154; CHECK: [[META0:![0-9]+]] = !{i32 0, i32 64770, i32 1078211522, !"main", i32 5, i32 0} 155; CHECK: [[META1:![0-9]+]] = !{i32 1, !"wchar_size", i32 4} 156; CHECK: [[META2:![0-9]+]] = !{i32 7, !"openmp", i32 50} 157; CHECK: [[META3:![0-9]+]] = !{i32 7, !"openmp-device", i32 50} 158; CHECK: [[META4:![0-9]+]] = !{i32 8, !"PIC Level", i32 2} 159; CHECK: [[META5:![0-9]+]] = !{i32 7, !"frame-pointer", i32 2} 160; CHECK: [[META6:![0-9]+]] = !{!"{{.*}}clang version {{.*}}"} 161; CHECK: [[TBAA7]] = !{[[META8:![0-9]+]], [[META8]], i64 0} 162; CHECK: [[META8]] = !{!"double", [[META9:![0-9]+]], i64 0} 163; CHECK: [[META9]] = !{!"omnipotent char", [[META10:![0-9]+]], i64 0} 164; CHECK: [[META10]] = !{!"Simple C/C++ TBAA"} 165;. 166