1 // Test device global memory data sharing codegen. 2 ///==========================================================================/// 3 4 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc 5 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CK1 6 7 // expected-no-diagnostics 8 9 #ifndef HEADER 10 #define HEADER 11 12 void test_ds(){ 13 #pragma omp target 14 { 15 int a = 10; 16 #pragma omp parallel 17 { 18 a = 1000; 19 } 20 int b = 100; 21 int c = 1000; 22 #pragma omp parallel private(c) 23 { 24 int *c1 = &c; 25 b = a + 10000; 26 } 27 } 28 } 29 30 /// ========= In the worker function ========= /// 31 // CK1: {{.*}}define internal void @__omp_offloading{{.*}}test_ds{{.*}}_worker() 32 // CK1: call void @llvm.nvvm.barrier0() 33 // CK1: call void @__kmpc_data_sharing_init_stack 34 35 /// ========= In the kernel function ========= /// 36 37 // CK1: {{.*}}define weak void @__omp_offloading{{.*}}test_ds{{.*}}() 38 // CK1: [[SHAREDARGS1:%.+]] = alloca i8** 39 // CK1: [[SHAREDARGS2:%.+]] = alloca i8** 40 // CK1: call void @__kmpc_kernel_init 41 // CK1: call void @__kmpc_data_sharing_init_stack 42 // CK1: [[GLOBALSTACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 0) 43 // CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty* 44 // CK1: [[A:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0 45 // CK1: [[B:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 1 46 // CK1: store i32 10, i32* [[A]] 47 // CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1) 48 // CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS1]], i64 1) 49 // CK1: [[SHARGSTMP1:%.+]] = load i8**, i8*** [[SHAREDARGS1]] 50 // CK1: [[SHARGSTMP2:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP1]], i64 0 51 // CK1: [[SHAREDVAR:%.+]] = bitcast i32* [[A]] to i8* 52 // CK1: store i8* [[SHAREDVAR]], i8** [[SHARGSTMP2]] 53 // CK1: call void @llvm.nvvm.barrier0() 54 // CK1: call void @llvm.nvvm.barrier0() 55 // CK1: call void @__kmpc_end_sharing_variables() 56 // CK1: store i32 100, i32* [[B]] 57 // CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1) 58 // CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS2]], i64 2) 59 // CK1: [[SHARGSTMP3:%.+]] = load i8**, i8*** [[SHAREDARGS2]] 60 // CK1: [[SHARGSTMP4:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]], i64 0 61 // CK1: [[SHAREDVAR1:%.+]] = bitcast i32* [[B]] to i8* 62 // CK1: store i8* [[SHAREDVAR1]], i8** [[SHARGSTMP4]] 63 // CK1: [[SHARGSTMP12:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]], i64 1 64 // CK1: [[SHAREDVAR2:%.+]] = bitcast i32* [[A]] to i8* 65 // CK1: store i8* [[SHAREDVAR2]], i8** [[SHARGSTMP12]] 66 // CK1: call void @llvm.nvvm.barrier0() 67 // CK1: call void @llvm.nvvm.barrier0() 68 // CK1: call void @__kmpc_end_sharing_variables() 69 // CK1: call void @__kmpc_data_sharing_pop_stack(i8* [[GLOBALSTACK]]) 70 // CK1: call void @__kmpc_kernel_deinit(i16 1) 71 72 /// ========= In the data sharing wrapper function ========= /// 73 74 // CK1: {{.*}}define internal void @__omp_outlined{{.*}}wrapper({{.*}}) 75 // CK1: [[SHAREDARGS4:%.+]] = alloca i8** 76 // CK1: call void @__kmpc_get_shared_variables(i8*** [[SHAREDARGS4]]) 77 // CK1: [[SHARGSTMP13:%.+]] = load i8**, i8*** [[SHAREDARGS4]] 78 // CK1: [[SHARGSTMP14:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP13]], i64 0 79 // CK1: [[SHARGSTMP15:%.+]] = bitcast i8** [[SHARGSTMP14]] to i32** 80 // CK1: [[SHARGSTMP16:%.+]] = load i32*, i32** [[SHARGSTMP15]] 81 // CK1: call void @__omp_outlined__{{.*}}({{.*}}, i32* [[SHARGSTMP16]]) 82 83 /// outlined function for the second parallel region /// 84 85 // CK1: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.+}}, i32* dereferenceable{{.+}}) 86 // CK1: [[RES:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 4, i16 0) 87 // CK1: [[GLOBALS:%.+]] = bitcast i8* [[RES]] to [[GLOBAL_TY:%.+]]* 88 // CK1: [[C_ADDR:%.+]] = getelementptr inbounds [[GLOBAL_TY]], [[GLOBAL_TY]]* [[GLOBALS]], i32 0, i32 0 89 // CK1: store i32* [[C_ADDR]], i32** % 90 // CK1: call void @__kmpc_data_sharing_pop_stack(i8* [[RES]]) 91 92 /// ========= In the data sharing wrapper function ========= /// 93 94 // CK1: {{.*}}define internal void @__omp_outlined{{.*}}wrapper({{.*}}) 95 // CK1: [[SHAREDARGS3:%.+]] = alloca i8** 96 // CK1: call void @__kmpc_get_shared_variables(i8*** [[SHAREDARGS3]]) 97 // CK1: [[SHARGSTMP5:%.+]] = load i8**, i8*** [[SHAREDARGS3]] 98 // CK1: [[SHARGSTMP6:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP5]], i64 0 99 // CK1: [[SHARGSTMP7:%.+]] = bitcast i8** [[SHARGSTMP6]] to i32** 100 // CK1: [[SHARGSTMP8:%.+]] = load i32*, i32** [[SHARGSTMP7]] 101 // CK1: [[SHARGSTMP9:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP5]], i64 1 102 // CK1: [[SHARGSTMP10:%.+]] = bitcast i8** [[SHARGSTMP9]] to i32** 103 // CK1: [[SHARGSTMP11:%.+]] = load i32*, i32** [[SHARGSTMP10]] 104 // CK1: call void @__omp_outlined__{{.*}}({{.*}}, i32* [[SHARGSTMP8]], i32* [[SHARGSTMP11]]) 105 106 #endif 107 108