1 // Test host codegen. 2 // RUN: %clang_cc1 -verify -Wno-vla -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 3 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 4 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 5 // RUN: %clang_cc1 -verify -Wno-vla -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 6 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 7 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 8 9 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s 10 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 11 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s 12 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s 13 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 14 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s 15 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}} 16 17 // Test target codegen - host bc file has to be created first. 18 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 19 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 20 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 21 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 22 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 23 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 24 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 25 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 26 27 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 28 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s 29 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 30 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s 31 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 32 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s 33 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 34 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s 35 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}} 36 37 // expected-no-diagnostics 38 #ifndef HEADER 39 #define HEADER 40 41 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } 42 // CHECK-DAG: [[ENTTY:%.+]] = type { i64, i16, i16, i32, ptr, ptr, i64, i64, ptr } 43 44 // TCHECK: [[ENTTY:%.+]] = type { i64, i16, i16, i32, ptr, ptr, i64, i64, ptr } 45 46 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i64] [i64 0, i64 4] 47 // CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] 48 // CHECK-DAG: @{{.*}} = weak constant i8 0 49 50 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 51 // TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] 52 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] 53 54 template<typename tx, typename ty> 55 struct TT{ 56 tx X; 57 ty Y; 58 }; 59 60 int global; 61 extern int global; 62 63 // CHECK: define {{.*}}[[FOO:@.+]]( 64 int foo(int n) { 65 int a = 0; 66 short aa = 0; 67 float b[10]; 68 float bn[n]; 69 double c[5][10]; 70 double cn[5][n]; 71 TT<long long, char> d; 72 static long *plocal; 73 74 // CHECK: [[ADD:%.+]] = add nsw i32 75 // CHECK: store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]], 76 // CHECK: [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 0 77 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]], 78 // CHECK: store i32 [[DEV]], ptr [[GEP]], 79 // CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr [[ID:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ:32|64]] {{20|40}}, i[[SZ]] 4, ptr [[TASK_ENTRY0:@.+]]) 80 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0 81 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1 82 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2 83 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 3 84 // CHECK: call void @__kmpc_omp_taskwait_deps_51(ptr [[ID]], i32 [[GTID]], i32 4, ptr %{{.+}}, i32 0, ptr null, i32 0) 85 // CHECK: call void @__kmpc_omp_task_begin_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]]) 86 // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], ptr [[TASK]]) 87 // CHECK: call void @__kmpc_omp_task_complete_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]]) 88 #pragma omp target teams distribute device(global + a) depend(in: global) depend(out: a, b, cn[4]) 89 for (int i = 0; i < 10; ++i) { 90 } 91 92 // CHECK: [[ADD:%.+]] = add nsw i32 93 // CHECK: store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]], 94 95 // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 96 // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] 97 // CHECK: [[THEN]]: 98 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP:%.+]], i32 0, i32 0 99 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P:%.+]], i32 0, i32 0 100 // CHECK-DAG: store ptr [[BP0:%[^,]+]], ptr [[BPADDR0]] 101 // CHECK-DAG: store ptr [[BP0]], ptr [[PADDR0]] 102 103 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1 104 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1 105 // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], ptr [[BPADDR1]] 106 // CHECK-DAG: store i[[SZ]] [[BP1]], ptr [[PADDR1]] 107 // CHECK-DAG: getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0 108 // CHECK-DAG: getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0 109 // CHECK: [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2 110 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]], 111 // CHECK: store i32 [[DEV]], ptr [[GEP]], 112 // CHECK: [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]], 113 // CHECK: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64 114 115 // CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{104|60}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1_:@.+]], i64 [[DEV2]]) 116 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0 117 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1 118 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2 119 // CHECK: call i32 @__kmpc_omp_task_with_deps(ptr [[ID]], i32 [[GTID]], ptr [[TASK]], i32 3, ptr %{{.+}}, i32 0, ptr null) 120 // CHECK: br label %[[EXIT:.+]] 121 122 // CHECK: [[ELSE]]: 123 // CHECK-NOT: getelementptr inbounds [2 x ptr], ptr 124 // CHECK: [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2 125 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]], 126 // CHECK: store i32 [[DEV]], ptr [[GEP]], 127 // CHECK: [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]], 128 // CHECK: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64 129 130 // CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1__:@.+]], i64 [[DEV2]]) 131 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0 132 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1 133 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2 134 // CHECK: call i32 @__kmpc_omp_task_with_deps(ptr [[ID]], i32 [[GTID]], ptr [[TASK]], i32 3, ptr %{{.+}}, i32 0, ptr null) 135 // CHECK: br label %[[EXIT:.+]] 136 // CHECK: [[EXIT]]: 137 138 #pragma omp target teams distribute device(global + a) nowait depend(inout: global, a, bn) if(a) 139 for (int i = 0; i < *plocal; ++i) { 140 static int local1; 141 *plocal = global; 142 local1 = global; 143 } 144 145 // CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, ptr [[TASK_ENTRY2:@.+]]) 146 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0 147 // CHECK: call void @__kmpc_omp_taskwait_deps_51(ptr [[ID]], i32 [[GTID]], i32 1, ptr %{{.+}}, i32 0, ptr null, i32 0) 148 // CHECK: call void @__kmpc_omp_task_begin_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]]) 149 // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], ptr [[TASK]]) 150 // CHECK: call void @__kmpc_omp_task_complete_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]]) 151 #pragma omp target teams distribute if(0) firstprivate(global) depend(out:global) 152 for (int i = 0; i < global; ++i) { 153 global += 1; 154 } 155 156 return a; 157 } 158 159 // Check that the offloading functions are emitted and that the arguments are 160 // correct and loaded correctly for the target regions in foo(). 161 162 // CHECK: define internal void [[HVT0:@.+]]() 163 164 // CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1) 165 // CHECK: store ptr null, ptr % 166 // CHECK: [[DEVICE_CAP:%.+]] = load ptr, 167 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]], 168 // CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 169 // CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, ptr @.{{.+}}.region_id, ptr %{{.+}}) 170 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 171 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 172 // CHECK: [[FAIL]] 173 // CHECK: call void [[HVT0]]() 174 // CHECK-NEXT: br label %[[END]] 175 // CHECK: [[END]] 176 // CHECK: ret i32 0 177 178 // CHECK: define internal void [[HVT1:@.+]](ptr noundef %{{.+}}, i[[SZ]] noundef %{{.+}}) 179 180 // CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, ptr noalias noundef %1) 181 // CHECK: call void {{%.*}}( 182 // CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2 183 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]], 184 // CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 185 // CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 0, i32 0, ptr @.{{.+}}.region_id, ptr %{{.+}}) 186 187 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 188 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 189 // CHECK: [[FAIL]] 190 // CHECK: [[BP0:%.+]] = load ptr, ptr % 191 // CHECK: [[BP1_I32:%.+]] = load i32, ptr @ 192 // CHECK-64: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]], 193 // CHECK-32: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]], 194 // CHECK: [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]], 195 // CHECK: call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]]) 196 // CHECK-NEXT: br label %[[END]] 197 // CHECK: [[END]] 198 // CHECK: ret i32 0 199 200 // CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, ptr noalias noundef %1) 201 // CHECK: call void {{%.*}}( 202 // CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2 203 // CHECK: [[BP0:%.+]] = load ptr, ptr % 204 // CHECK: [[BP1_I32:%.+]] = load i32, ptr @ 205 // CHECK-64: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]], 206 // CHECK-32: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]], 207 // CHECK: [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]], 208 // CHECK: call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]]) 209 // CHECK: ret i32 0 210 211 // CHECK: define internal void [[HVT2:@.+]](i[[SZ]] noundef %{{.+}}) 212 // Create stack storage and store argument in there. 213 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 214 // CHECK: store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align 215 // CHECK-64: load i32, ptr [[AA_ADDR]], align 216 // CHECK-32: load i32, ptr [[AA_ADDR]], align 217 218 // CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1) 219 // CHECK: call void {{%.*}}( 220 // CHECK: [[BP1_I32:%.+]] = load i32, ptr % 221 // CHECK-64: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]], 222 // CHECK-32: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]], 223 // CHECK: [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]], 224 // CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) 225 // CHECK: ret i32 0 226 227 228 #endif 229