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 - -disable-llvm-optzns | 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 // CK1: [[MEM_TY:%.+]] = type { [128 x i8] } 30 // CK1-DAG: [[SHARED_GLOBAL_RD:@.+]] = common addrspace(3) global [[MEM_TY]] zeroinitializer 31 // CK1-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null 32 // CK1-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i64 8 33 // CK1-DAG: [[KERNEL_SHARED:@.+]] = internal unnamed_addr constant i16 1 34 35 /// ========= In the worker function ========= /// 36 // CK1: {{.*}}define internal void @__omp_offloading{{.*}}test_ds{{.*}}_worker() 37 // CK1: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 38 // CK1-NOT: call void @__kmpc_data_sharing_init_stack 39 40 /// ========= In the kernel function ========= /// 41 42 // CK1: {{.*}}define weak void @__omp_offloading{{.*}}test_ds{{.*}}() 43 // CK1: [[SHAREDARGS1:%.+]] = alloca i8** 44 // CK1: [[SHAREDARGS2:%.+]] = alloca i8** 45 // CK1: call void @__kmpc_kernel_init 46 // CK1: call void @__kmpc_data_sharing_init_stack 47 // CK1: [[SHARED_MEM_FLAG:%.+]] = load i16, i16* [[KERNEL_SHARED]], 48 // CK1: [[SIZE:%.+]] = load i64, i64* [[KERNEL_SIZE]], 49 // CK1: call void @__kmpc_get_team_static_memory(i16 0, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i64 [[SIZE]], i16 [[SHARED_MEM_FLAG]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**)) 50 // CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]], 51 // CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i64 0 52 // CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty* 53 // CK1: [[A:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0 54 // CK1: [[B:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 1 55 // CK1: store i32 10, i32* [[A]] 56 // CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1) 57 // CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS1]], i64 1) 58 // CK1: [[SHARGSTMP1:%.+]] = load i8**, i8*** [[SHAREDARGS1]] 59 // CK1: [[SHARGSTMP2:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP1]], i64 0 60 // CK1: [[SHAREDVAR:%.+]] = bitcast i32* [[A]] to i8* 61 // CK1: store i8* [[SHAREDVAR]], i8** [[SHARGSTMP2]] 62 // CK1: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 63 // CK1: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 64 // CK1: call void @__kmpc_end_sharing_variables() 65 // CK1: store i32 100, i32* [[B]] 66 // CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1) 67 // CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS2]], i64 2) 68 // CK1: [[SHARGSTMP3:%.+]] = load i8**, i8*** [[SHAREDARGS2]] 69 // CK1: [[SHARGSTMP4:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]], i64 0 70 // CK1: [[SHAREDVAR1:%.+]] = bitcast i32* [[B]] to i8* 71 // CK1: store i8* [[SHAREDVAR1]], i8** [[SHARGSTMP4]] 72 // CK1: [[SHARGSTMP12:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]], i64 1 73 // CK1: [[SHAREDVAR2:%.+]] = bitcast i32* [[A]] to i8* 74 // CK1: store i8* [[SHAREDVAR2]], i8** [[SHARGSTMP12]] 75 // CK1: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 76 // CK1: call void @__kmpc_barrier_simple_spmd(%struct.ident_t* null, i32 0) 77 // CK1: call void @__kmpc_end_sharing_variables() 78 // CK1: [[SHARED_MEM_FLAG:%.+]] = load i16, i16* [[KERNEL_SHARED]], 79 // CK1: call void @__kmpc_restore_team_static_memory(i16 0, i16 [[SHARED_MEM_FLAG]]) 80 // CK1: call void @__kmpc_kernel_deinit(i16 1) 81 82 /// ========= In the data sharing wrapper function ========= /// 83 84 // CK1: {{.*}}define internal void @__omp_outlined{{.*}}wrapper({{.*}}) 85 // CK1: [[SHAREDARGS4:%.+]] = alloca i8** 86 // CK1: call void @__kmpc_get_shared_variables(i8*** [[SHAREDARGS4]]) 87 // CK1: [[SHARGSTMP13:%.+]] = load i8**, i8*** [[SHAREDARGS4]] 88 // CK1: [[SHARGSTMP14:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP13]], i64 0 89 // CK1: [[SHARGSTMP15:%.+]] = bitcast i8** [[SHARGSTMP14]] to i32** 90 // CK1: [[SHARGSTMP16:%.+]] = load i32*, i32** [[SHARGSTMP15]] 91 // CK1: call void @__omp_outlined__{{.*}}({{.*}}, i32* [[SHARGSTMP16]]) 92 93 /// outlined function for the second parallel region /// 94 95 // CK1: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.+}}, i32* dereferenceable{{.+}}) 96 // CK1-NOT: call i8* @__kmpc_data_sharing_push_stack( 97 // CK1: [[C_ADDR:%.+]] = alloca i32, 98 // CK1: store i32* [[C_ADDR]], i32** % 99 // CK1i-NOT: call void @__kmpc_data_sharing_pop_stack( 100 101 /// ========= In the data sharing wrapper function ========= /// 102 103 // CK1: {{.*}}define internal void @__omp_outlined{{.*}}wrapper({{.*}}) 104 // CK1: [[SHAREDARGS3:%.+]] = alloca i8** 105 // CK1: call void @__kmpc_get_shared_variables(i8*** [[SHAREDARGS3]]) 106 // CK1: [[SHARGSTMP5:%.+]] = load i8**, i8*** [[SHAREDARGS3]] 107 // CK1: [[SHARGSTMP6:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP5]], i64 0 108 // CK1: [[SHARGSTMP7:%.+]] = bitcast i8** [[SHARGSTMP6]] to i32** 109 // CK1: [[SHARGSTMP8:%.+]] = load i32*, i32** [[SHARGSTMP7]] 110 // CK1: [[SHARGSTMP9:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP5]], i64 1 111 // CK1: [[SHARGSTMP10:%.+]] = bitcast i8** [[SHARGSTMP9]] to i32** 112 // CK1: [[SHARGSTMP11:%.+]] = load i32*, i32** [[SHARGSTMP10]] 113 // CK1: call void @__omp_outlined__{{.*}}({{.*}}, i32* [[SHARGSTMP8]], i32* [[SHARGSTMP11]]) 114 115 #endif 116 117