1 // RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 2 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 3 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 4 // RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 5 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 6 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 7 8 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 9 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 10 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 11 12 // expected-no-diagnostics 13 #ifndef HEADER 14 #define HEADER 15 16 int x; 17 #pragma omp threadprivate(x) 18 19 template <typename T> 20 T tmain() { 21 int a[2]; 22 #pragma omp target 23 #pragma omp teams distribute parallel for copyin(x) 24 for (int i = 0; i < 2; ++i) { 25 a[i] = x; 26 } 27 return T(); 28 } 29 30 int main() { 31 int a[2]; 32 #ifdef LAMBDA 33 // LAMBDA-LABEL: @main 34 // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( 35 [&]() { 36 // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( 37 // LAMBDA: call i32 @__tgt_target_teams( 38 // LAMBDA: call void @[[LOFFL1:.+]]( 39 // LAMBDA: ret 40 #pragma omp target 41 #pragma omp teams distribute parallel for copyin(x) 42 for (int i = 0; i < 2; ++i) { 43 // LAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]]( 44 // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[LOUTL1:.+]] to {{.+}}) 45 // LAMBDA: ret void 46 47 // LAMBDA: define internal void @[[LOUTL1]]( 48 // Skip global, bound tid and loop vars 49 // LAMBDA: {{.+}} = alloca i32*, 50 // LAMBDA: {{.+}} = alloca i32*, 51 // LAMBDA: alloca i32, 52 // LAMBDA: alloca i32, 53 // LAMBDA: alloca i32, 54 // LAMBDA: alloca i32, 55 // LAMBDA: alloca i32, 56 a[i] = x; 57 58 // LAMBDA: call void @__kmpc_for_static_init_4( 59 // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to 60 // LAMBDA: call void @__kmpc_for_static_fini( 61 // LAMBDA: ret void 62 63 // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}) 64 // Skip global, bound tid and loop vars 65 // LAMBDA: {{.+}} = alloca i32*, 66 // LAMBDA: {{.+}} = alloca i32*, 67 // LAMBDA: alloca i{{[0-9]+}}, 68 // LAMBDA: alloca i{{[0-9]+}}, 69 // LAMBDA: {{%.+}} = alloca [2 x i{{[0-9]+}}]*, 70 // LAMBDA: [[X_ADDR:%.+]] = alloca i{{[0-9]+}}*, 71 // LAMBDA: alloca i32, 72 // LAMBDA: alloca i32, 73 // LAMBDA: alloca i32, 74 // LAMBDA: alloca i32, 75 // LAMBDA: alloca i32, 76 77 // LAMBDA: [[X_REF:%.+]] = load {{.+}}, {{.+}} [[X_ADDR]], 78 79 // LAMBDA: call void @__kmpc_for_static_init_4( 80 // LAMBDA: [[X_VAL:%.+]] = load {{.+}}, {{.+}} [[X_REF]], 81 // LAMBDA: store {{.+}} [[X_VAL]], 82 // LAMBDA: call void [[INNER_LAMBDA:@.+]]( 83 // LAMBDA: call void @__kmpc_for_static_fini( 84 // LAMBDA: ret void 85 [&]() { 86 // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) 87 // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], 88 a[i] = x; 89 }(); 90 } 91 }(); 92 return 0; 93 #else 94 #pragma omp target 95 #pragma omp teams distribute parallel for copyin(x) 96 for (int i = 0; i < 2; ++i) { 97 a[i] = x; 98 } 99 return tmain<int>(); 100 //return 0; 101 #endif 102 } 103 104 // CHECK: define {{.*}}i{{[0-9]+}} @main() 105 // CHECK: call i32 @__tgt_target_teams( 106 // CHECK: call void @[[OFFL1:.+]](i{{64|32}} %{{.+}}) 107 // CHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() 108 // CHECK: ret 109 110 // CHECK: define{{.*}} void @[[OFFL1]]({{.+}}) 111 // CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTL1:.+]] to {{.+}}) 112 // CHECK: ret void 113 114 // CHECK: define internal void @[[OUTL1]]({{.+}}) 115 // Skip global, bound tid and loop vars 116 // CHECK: {{.+}} = alloca i32*, 117 // CHECK: {{.+}} = alloca i32*, 118 // CHECK: {{.+}} = alloca i32, 119 // CHECK: {{.+}} = alloca i32, 120 // CHECK: {{.+}} = alloca i32, 121 // CHECK: {{.+}} = alloca i32, 122 // CHECK: {{.+}} = alloca i32, 123 124 // CHECK: call void @__kmpc_for_static_init_4( 125 // CHECK: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to 126 // CHECK: call void @__kmpc_for_static_fini( 127 // CHECK: ret void 128 129 // CHECK: define internal void @[[PAR_OUTL1]]({{.+}}) 130 // Skip global, bound tid and loop vars 131 // CHECK: {{.+}} = alloca i32*, 132 // CHECK: {{.+}} = alloca i32*, 133 // CHECK: {{.+}} = alloca i{{[0-9]+}}, 134 // CHECK: {{.+}} = alloca i{{[0-9]+}}, 135 // CHECK: {{%.+}} = alloca [2 x i{{[0-9]+}}]*, 136 // CHECK: [[X_ADDR:%.+]] = alloca i{{[0-9]+}}*, 137 // CHECK: {{.+}} = alloca i32, 138 // CHECK: {{.+}} = alloca i32, 139 // CHECK: {{.+}} = alloca i32, 140 // CHECK: {{.+}} = alloca i32, 141 // CHECK: {{.+}} = alloca i32, 142 143 // CHECK: [[X_REF:%.+]] = load {{.+}}, {{.+}} [[X_ADDR]], 144 // CHECK: call void @__kmpc_for_static_init_4( 145 // CHECK: [[X_VAL:%.+]] = load {{.+}}, {{.+}} [[X_REF]], 146 // CHECK: store {{.+}} [[X_VAL]], 147 // CHECK: call void @__kmpc_for_static_fini( 148 // CHECK: ret void 149 150 // CHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() 151 // CHECK: call i32 @__tgt_target_teams( 152 // CHECK: call void @[[TOFFL1:.+]]( 153 // CHECK: ret 154 155 // CHECK: define {{.*}}void @[[TOFFL1]]( 156 // CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[TOUTL1:.+]] to {{.+}}) 157 // CHECK: ret void 158 159 // CHECK: define internal void @[[TOUTL1]]({{.+}}) 160 // Skip global, bound tid and loop vars 161 // CHECK: {{.+}} = alloca i32*, 162 // CHECK: {{.+}} = alloca i32*, 163 // CHECK: alloca i{{[0-9]+}}, 164 // CHECK: alloca i{{[0-9]+}}, 165 // CHECK: alloca i{{[0-9]+}}, 166 // CHECK: alloca i{{[0-9]+}}, 167 // CHECK: alloca i{{[0-9]+}}, 168 169 // CHECK: call void @__kmpc_for_static_init_4( 170 // CHECK: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL1:.+]] to 171 // CHECK: call void @__kmpc_for_static_fini( 172 // CHECK: ret void 173 174 // CHECK: define internal void @[[TPAR_OUTL1]]({{.+}}) 175 // Skip global, bound tid and loop vars 176 // CHECK: {{.+}} = alloca i32*, 177 // CHECK: {{.+}} = alloca i32*, 178 // prev lb and ub 179 // CHECK: alloca i{{[0-9]+}}, 180 // CHECK: alloca i{{[0-9]+}}, 181 182 // CHECK: {{%.+}} = alloca [2 x i{{[0-9]+}}]*, 183 // CHECK: [[X_ADDR:%.+]] = alloca i{{[0-9]+}}*, 184 // iter variables 185 // CHECK: alloca i{{[0-9]+}}, 186 // CHECK: alloca i{{[0-9]+}}, 187 // CHECK: alloca i{{[0-9]+}}, 188 // CHECK: alloca i{{[0-9]+}}, 189 // CHECK: alloca i{{[0-9]+}}, 190 191 // CHECK: [[X_REF:%.+]] = load {{.+}}, {{.+}} [[X_ADDR]], 192 // CHECK: call void @__kmpc_for_static_init_4( 193 // CHECK: [[X_VAL:%.+]] = load {{.+}}, {{.+}} [[X_REF]], 194 // CHECK: store {{.+}} [[X_VAL]], 195 // CHECK: call void @__kmpc_for_static_fini( 196 // CHECK: ret void 197 198 199 #endif 200