1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s 2 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s 3 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s 4 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG 5 // RUN: %clang_cc1 -main-file-name for_codegen.cpp %s -o - -emit-llvm -fprofile-instrument=clang -fprofile-instrument-path=for_codegen-test.profraw | FileCheck %s --check-prefix=PROF-INSTR-PATH 6 7 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s 8 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s 9 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s 10 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s 11 // RUN: %clang_cc1 -main-file-name for_codegen.cpp %s -o - -emit-llvm -fprofile-instrument=clang -fprofile-instrument-path=for_codegen-test.profraw | FileCheck --check-prefix SIMD-ONLY0 %s 12 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}} 13 // 14 // expected-no-diagnostics 15 #ifndef HEADER 16 #define HEADER 17 // PROF-INSTR-PATH: constant [25 x i8] c"for_codegen-test.profraw\00" 18 19 // CHECK: [[IDENT_T_TY:%.+]] = type { i32, i32, i32, i32, i8* } 20 // CHECK-DAG: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8* 21 // CHECK-DAG: [[LOOP_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 514, i32 0, i32 0, i8* 22 // CHECK-DAG: [[I:@.+]] = global i8 1, 23 // CHECK-DAG: [[J:@.+]] = global i8 2, 24 // CHECK-DAG: [[K:@.+]] = global i8 3, 25 26 // CHECK-LABEL: loop_with_counter_collapse 27 void loop_with_counter_collapse() { 28 // CHECK: call void @__kmpc_for_static_init_8(%ident_t* @ 29 // CHECK: call void @__kmpc_for_static_fini(%ident_t* @ 30 #pragma omp for collapse(2) 31 for (int i = 0; i < 4; i++) { 32 for (int j = i; j < 4; j++) { 33 } 34 } 35 } 36 // CHECK-LABEL: define {{.*void}} @{{.*}}without_schedule_clause{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 37 void without_schedule_clause(float *a, float *b, float *c, float *d) { 38 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 39 #pragma omp for nowait 40 // CHECK: call void @__kmpc_for_static_init_4([[IDENT_T_TY]]* [[LOOP_LOC]], i32 [[GTID]], i32 34, i32* [[IS_LAST:%[^,]+]], i32* [[OMP_LB:%[^,]+]], i32* [[OMP_UB:%[^,]+]], i32* [[OMP_ST:%[^,]+]], i32 1, i32 1) 41 // UB = min(UB, GlobalUB) 42 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 43 // CHECK-NEXT: [[UBCMP:%.+]] = icmp sgt i32 [[UB]], 4571423 44 // CHECK-NEXT: br i1 [[UBCMP]], label [[UB_TRUE:%[^,]+]], label [[UB_FALSE:%[^,]+]] 45 // CHECK: [[UBRESULT:%.+]] = phi i32 [ 4571423, [[UB_TRUE]] ], [ [[UBVAL:%[^,]+]], [[UB_FALSE]] ] 46 // CHECK-NEXT: store i32 [[UBRESULT]], i32* [[OMP_UB]] 47 // CHECK-NEXT: [[LB:%.+]] = load i32, i32* [[OMP_LB]] 48 // CHECK-NEXT: store i32 [[LB]], i32* [[OMP_IV:[^,]+]] 49 // Loop header 50 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]] 51 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 52 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i32 [[IV]], [[UB]] 53 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 54 for (int i = 33; i < 32000000; i += 7) { 55 // CHECK: [[LOOP1_BODY]] 56 // Start of body: calculate i from IV: 57 // CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]] 58 // CHECK-NEXT: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 7 59 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 33, [[CALC_I_1]] 60 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]] 61 // ... loop body ... 62 // End of body: store into a[i]: 63 // CHECK: store float [[RESULT:%.+]], float* {{%.+}} 64 // CHECK-NOT: !llvm.mem.parallel_loop_access 65 a[i] = b[i] * c[i] * d[i]; 66 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}} 67 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1 68 // CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]] 69 // CHECK-NEXT: br label %{{.+}} 70 } 71 // CHECK: [[LOOP1_END]] 72 // CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[LOOP_LOC]], i32 [[GTID]]) 73 // CHECK-NOT: __kmpc_barrier 74 // CHECK: ret void 75 } 76 77 // CHECK-LABEL: define {{.*void}} @{{.*}}static_not_chunked{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 78 void static_not_chunked(float *a, float *b, float *c, float *d) { 79 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 80 #pragma omp for schedule(static) 81 // CHECK: call void @__kmpc_for_static_init_4([[IDENT_T_TY]]* [[LOOP_LOC]], i32 [[GTID]], i32 34, i32* [[IS_LAST:%[^,]+]], i32* [[OMP_LB:%[^,]+]], i32* [[OMP_UB:%[^,]+]], i32* [[OMP_ST:%[^,]+]], i32 1, i32 1) 82 // UB = min(UB, GlobalUB) 83 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 84 // CHECK-NEXT: [[UBCMP:%.+]] = icmp sgt i32 [[UB]], 4571423 85 // CHECK-NEXT: br i1 [[UBCMP]], label [[UB_TRUE:%[^,]+]], label [[UB_FALSE:%[^,]+]] 86 // CHECK: [[UBRESULT:%.+]] = phi i32 [ 4571423, [[UB_TRUE]] ], [ [[UBVAL:%[^,]+]], [[UB_FALSE]] ] 87 // CHECK-NEXT: store i32 [[UBRESULT]], i32* [[OMP_UB]] 88 // CHECK-NEXT: [[LB:%.+]] = load i32, i32* [[OMP_LB]] 89 // CHECK-NEXT: store i32 [[LB]], i32* [[OMP_IV:[^,]+]] 90 // Loop header 91 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]] 92 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 93 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i32 [[IV]], [[UB]] 94 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 95 for (int i = 32000000; i > 33; i += -7) { 96 // CHECK: [[LOOP1_BODY]] 97 // Start of body: calculate i from IV: 98 // CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]] 99 // CHECK-NEXT: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 7 100 // CHECK-NEXT: [[CALC_I_2:%.+]] = sub nsw i32 32000000, [[CALC_I_1]] 101 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]] 102 // ... loop body ... 103 // End of body: store into a[i]: 104 // CHECK: store float [[RESULT:%.+]], float* {{%.+}} 105 // CHECK-NOT: !llvm.mem.parallel_loop_access 106 a[i] = b[i] * c[i] * d[i]; 107 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}} 108 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1 109 // CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]] 110 // CHECK-NEXT: br label %{{.+}} 111 } 112 // CHECK: [[LOOP1_END]] 113 // CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[LOOP_LOC]], i32 [[GTID]]) 114 // CHECK: call {{.+}} @__kmpc_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]]) 115 // CHECK: ret void 116 } 117 118 // CHECK-LABEL: define {{.*void}} @{{.*}}static_chunked{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 119 void static_chunked(float *a, float *b, float *c, float *d) { 120 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 121 #pragma omp for schedule(monotonic: static, 5) 122 // CHECK: call void @__kmpc_for_static_init_4u([[IDENT_T_TY]]* [[LOOP_LOC]], i32 [[GTID]], i32 536870945, i32* [[IS_LAST:%[^,]+]], i32* [[OMP_LB:%[^,]+]], i32* [[OMP_UB:%[^,]+]], i32* [[OMP_ST:%[^,]+]], i32 1, i32 5) 123 // UB = min(UB, GlobalUB) 124 // CHECK: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 125 // CHECK-NEXT: [[UBCMP:%.+]] = icmp ugt i32 [[UB]], 16908288 126 // CHECK-NEXT: br i1 [[UBCMP]], label [[UB_TRUE:%[^,]+]], label [[UB_FALSE:%[^,]+]] 127 // CHECK: [[UBRESULT:%.+]] = phi i32 [ 16908288, [[UB_TRUE]] ], [ [[UBVAL:%[^,]+]], [[UB_FALSE]] ] 128 // CHECK-NEXT: store i32 [[UBRESULT]], i32* [[OMP_UB]] 129 // CHECK-NEXT: [[LB:%.+]] = load i32, i32* [[OMP_LB]] 130 // CHECK-NEXT: store i32 [[LB]], i32* [[OMP_IV:[^,]+]] 131 132 // Outer loop header 133 // CHECK: [[O_IV:%.+]] = load i32, i32* [[OMP_IV]] 134 // CHECK-NEXT: [[O_UB:%.+]] = load i32, i32* [[OMP_UB]] 135 // CHECK-NEXT: [[O_CMP:%.+]] = icmp ule i32 [[O_IV]], [[O_UB]] 136 // CHECK-NEXT: br i1 [[O_CMP]], label %[[O_LOOP1_BODY:[^,]+]], label %[[O_LOOP1_END:[^,]+]] 137 138 // Loop header 139 // CHECK: [[O_LOOP1_BODY]] 140 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]] 141 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 142 // CHECK-NEXT: [[CMP:%.+]] = icmp ule i32 [[IV]], [[UB]] 143 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 144 for (unsigned i = 131071; i <= 2147483647; i += 127) { 145 // CHECK: [[LOOP1_BODY]] 146 // Start of body: calculate i from IV: 147 // CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]] 148 // CHECK-NEXT: [[CALC_I_1:%.+]] = mul i32 [[IV1_1]], 127 149 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 131071, [[CALC_I_1]] 150 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]] 151 // ... loop body ... 152 // End of body: store into a[i]: 153 // CHECK: store float [[RESULT:%.+]], float* {{%.+}} 154 // CHECK-NOT: !llvm.mem.parallel_loop_access 155 a[i] = b[i] * c[i] * d[i]; 156 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}} 157 // CHECK-NEXT: [[ADD1_2:%.+]] = add i32 [[IV1_2]], 1 158 // CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]] 159 // CHECK-NEXT: br label %{{.+}} 160 } 161 // CHECK: [[LOOP1_END]] 162 // Update the counters, adding stride 163 // CHECK: [[LB:%.+]] = load i32, i32* [[OMP_LB]] 164 // CHECK-NEXT: [[ST:%.+]] = load i32, i32* [[OMP_ST]] 165 // CHECK-NEXT: [[ADD_LB:%.+]] = add i32 [[LB]], [[ST]] 166 // CHECK-NEXT: store i32 [[ADD_LB]], i32* [[OMP_LB]] 167 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 168 // CHECK-NEXT: [[ST:%.+]] = load i32, i32* [[OMP_ST]] 169 // CHECK-NEXT: [[ADD_UB:%.+]] = add i32 [[UB]], [[ST]] 170 // CHECK-NEXT: store i32 [[ADD_UB]], i32* [[OMP_UB]] 171 172 // CHECK: [[O_LOOP1_END]] 173 // CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[LOOP_LOC]], i32 [[GTID]]) 174 // CHECK: call {{.+}} @__kmpc_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]]) 175 // CHECK: ret void 176 } 177 178 // CHECK-LABEL: define {{.*void}} @{{.*}}dynamic1{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 179 void dynamic1(float *a, float *b, float *c, float *d) { 180 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 181 #pragma omp for schedule(nonmonotonic: dynamic) 182 // CHECK: call void @__kmpc_dispatch_init_8u([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32 1073741859, i64 0, i64 16908287, i64 1, i64 1) 183 // 184 // CHECK: [[HASWORK:%.+]] = call i32 @__kmpc_dispatch_next_8u([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32* [[OMP_ISLAST:%[^,]+]], i64* [[OMP_LB:%[^,]+]], i64* [[OMP_UB:%[^,]+]], i64* [[OMP_ST:%[^,]+]]) 185 // CHECK-NEXT: [[O_CMP:%.+]] = icmp ne i32 [[HASWORK]], 0 186 // CHECK-NEXT: br i1 [[O_CMP]], label %[[O_LOOP1_BODY:[^,]+]], label %[[O_LOOP1_END:[^,]+]] 187 188 // Loop header 189 // CHECK: [[O_LOOP1_BODY]] 190 // CHECK: [[LB:%.+]] = load i64, i64* [[OMP_LB]] 191 // CHECK-NEXT: store i64 [[LB]], i64* [[OMP_IV:[^,]+]] 192 // CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]] 193 194 // CHECK-NEXT: [[UB:%.+]] = load i64, i64* [[OMP_UB]] 195 // CHECK-NEXT: [[CMP:%.+]] = icmp ule i64 [[IV]], [[UB]] 196 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 197 for (unsigned long long i = 131071; i < 2147483647; i += 127) { 198 // CHECK: [[LOOP1_BODY]] 199 // Start of body: calculate i from IV: 200 // CHECK: [[IV1_1:%.+]] = load i64, i64* [[OMP_IV]] 201 // CHECK-NEXT: [[CALC_I_1:%.+]] = mul i64 [[IV1_1]], 127 202 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i64 131071, [[CALC_I_1]] 203 // CHECK-NEXT: store i64 [[CALC_I_2]], i64* [[LC_I:.+]] 204 // ... loop body ... 205 // End of body: store into a[i]: 206 // CHECK: store float [[RESULT:%.+]], float* {{%.+}}!llvm.mem.parallel_loop_access 207 a[i] = b[i] * c[i] * d[i]; 208 // CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}} 209 // CHECK-NEXT: [[ADD1_2:%.+]] = add i64 [[IV1_2]], 1 210 // CHECK-NEXT: store i64 [[ADD1_2]], i64* [[OMP_IV]] 211 // CHECK-NEXT: br label %{{.+}} 212 } 213 // CHECK: [[LOOP1_END]] 214 // CHECK: [[O_LOOP1_END]] 215 // CHECK: call {{.+}} @__kmpc_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]]) 216 // CHECK: ret void 217 } 218 219 // CHECK-LABEL: define {{.*void}} @{{.*}}guided7{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 220 void guided7(float *a, float *b, float *c, float *d) { 221 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 222 #pragma omp for schedule(guided, 7) 223 // CHECK: call void @__kmpc_dispatch_init_8u([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32 36, i64 0, i64 16908287, i64 1, i64 7) 224 // 225 // CHECK: [[HASWORK:%.+]] = call i32 @__kmpc_dispatch_next_8u([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32* [[OMP_ISLAST:%[^,]+]], i64* [[OMP_LB:%[^,]+]], i64* [[OMP_UB:%[^,]+]], i64* [[OMP_ST:%[^,]+]]) 226 // CHECK-NEXT: [[O_CMP:%.+]] = icmp ne i32 [[HASWORK]], 0 227 // CHECK-NEXT: br i1 [[O_CMP]], label %[[O_LOOP1_BODY:[^,]+]], label %[[O_LOOP1_END:[^,]+]] 228 229 // Loop header 230 // CHECK: [[O_LOOP1_BODY]] 231 // CHECK: [[LB:%.+]] = load i64, i64* [[OMP_LB]] 232 // CHECK-NEXT: store i64 [[LB]], i64* [[OMP_IV:[^,]+]] 233 // CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]] 234 235 // CHECK-NEXT: [[UB:%.+]] = load i64, i64* [[OMP_UB]] 236 // CHECK-NEXT: [[CMP:%.+]] = icmp ule i64 [[IV]], [[UB]] 237 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 238 for (unsigned long long i = 131071; i < 2147483647; i += 127) { 239 // CHECK: [[LOOP1_BODY]] 240 // Start of body: calculate i from IV: 241 // CHECK: [[IV1_1:%.+]] = load i64, i64* [[OMP_IV]] 242 // CHECK-NEXT: [[CALC_I_1:%.+]] = mul i64 [[IV1_1]], 127 243 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i64 131071, [[CALC_I_1]] 244 // CHECK-NEXT: store i64 [[CALC_I_2]], i64* [[LC_I:.+]] 245 // ... loop body ... 246 // End of body: store into a[i]: 247 // CHECK: store float [[RESULT:%.+]], float* {{%.+}}!llvm.mem.parallel_loop_access 248 a[i] = b[i] * c[i] * d[i]; 249 // CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}} 250 // CHECK-NEXT: [[ADD1_2:%.+]] = add i64 [[IV1_2]], 1 251 // CHECK-NEXT: store i64 [[ADD1_2]], i64* [[OMP_IV]] 252 // CHECK-NEXT: br label %{{.+}} 253 } 254 // CHECK: [[LOOP1_END]] 255 // CHECK: [[O_LOOP1_END]] 256 // CHECK: call {{.+}} @__kmpc_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]]) 257 // CHECK: ret void 258 } 259 260 // CHECK-LABEL: define {{.*void}} @{{.*}}test_auto{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 261 void test_auto(float *a, float *b, float *c, float *d) { 262 unsigned int x = 0; 263 unsigned int y = 0; 264 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 265 #pragma omp for schedule(auto) collapse(2) 266 // CHECK: call void @__kmpc_dispatch_init_8([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32 38, i64 0, i64 [[LAST_ITER:%[^,]+]], i64 1, i64 1) 267 // 268 // CHECK: [[HASWORK:%.+]] = call i32 @__kmpc_dispatch_next_8([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32* [[OMP_ISLAST:%[^,]+]], i64* [[OMP_LB:%[^,]+]], i64* [[OMP_UB:%[^,]+]], i64* [[OMP_ST:%[^,]+]]) 269 // CHECK-NEXT: [[O_CMP:%.+]] = icmp ne i32 [[HASWORK]], 0 270 // CHECK-NEXT: br i1 [[O_CMP]], label %[[O_LOOP1_BODY:[^,]+]], label %[[O_LOOP1_END:[^,]+]] 271 272 // Loop header 273 // CHECK: [[O_LOOP1_BODY]] 274 // CHECK: [[LB:%.+]] = load i64, i64* [[OMP_LB]] 275 // CHECK-NEXT: store i64 [[LB]], i64* [[OMP_IV:[^,]+]] 276 // CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]] 277 278 // CHECK-NEXT: [[UB:%.+]] = load i64, i64* [[OMP_UB]] 279 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i64 [[IV]], [[UB]] 280 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 281 // FIXME: When the iteration count of some nested loop is not a known constant, 282 // we should pre-calculate it, like we do for the total number of iterations! 283 for (char i = static_cast<char>(y); i <= '9'; ++i) 284 for (x = 11; x > 0; --x) { 285 // CHECK: [[LOOP1_BODY]] 286 // Start of body: indices are calculated from IV: 287 // CHECK: store i8 {{%[^,]+}}, i8* {{%[^,]+}} 288 // CHECK: store i32 {{%[^,]+}}, i32* {{%[^,]+}} 289 // ... loop body ... 290 // End of body: store into a[i]: 291 // CHECK: store float [[RESULT:%.+]], float* {{%.+}} 292 // CHECK-NOT: !llvm.mem.parallel_loop_access 293 a[i] = b[i] * c[i] * d[i]; 294 // CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}} 295 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i64 [[IV1_2]], 1 296 // CHECK-NEXT: store i64 [[ADD1_2]], i64* [[OMP_IV]] 297 // CHECK-NEXT: br label %{{.+}} 298 } 299 // CHECK: [[LOOP1_END]] 300 // CHECK: [[O_LOOP1_END]] 301 // CHECK: call {{.+}} @__kmpc_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]]) 302 // CHECK: ret void 303 } 304 305 // CHECK-LABEL: define {{.*void}} @{{.*}}runtime{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 306 void runtime(float *a, float *b, float *c, float *d) { 307 int x = 0; 308 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]]) 309 #pragma omp for collapse(2) schedule(runtime) 310 // CHECK: call void @__kmpc_dispatch_init_4([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32 37, i32 0, i32 199, i32 1, i32 1) 311 // 312 // CHECK: [[HASWORK:%.+]] = call i32 @__kmpc_dispatch_next_4([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32* [[OMP_ISLAST:%[^,]+]], i32* [[OMP_LB:%[^,]+]], i32* [[OMP_UB:%[^,]+]], i32* [[OMP_ST:%[^,]+]]) 313 // CHECK-NEXT: [[O_CMP:%.+]] = icmp ne i32 [[HASWORK]], 0 314 // CHECK-NEXT: br i1 [[O_CMP]], label %[[O_LOOP1_BODY:[^,]+]], label %[[O_LOOP1_END:[^,]+]] 315 316 // Loop header 317 // CHECK: [[O_LOOP1_BODY]] 318 // CHECK: [[LB:%.+]] = load i32, i32* [[OMP_LB]] 319 // CHECK-NEXT: store i32 [[LB]], i32* [[OMP_IV:[^,]+]] 320 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]] 321 322 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]] 323 // CHECK-NEXT: [[CMP:%.+]] = icmp sle i32 [[IV]], [[UB]] 324 // CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1_BODY:[^,]+]], label %[[LOOP1_END:[^,]+]] 325 for (unsigned char i = '0' ; i <= '9'; ++i) 326 for (x = -10; x < 10; ++x) { 327 // CHECK: [[LOOP1_BODY]] 328 // Start of body: indices are calculated from IV: 329 // CHECK: store i8 {{%[^,]+}}, i8* {{%[^,]+}} 330 // CHECK: store i32 {{%[^,]+}}, i32* {{%[^,]+}} 331 // ... loop body ... 332 // End of body: store into a[i]: 333 // CHECK: store float [[RESULT:%.+]], float* {{%.+}} 334 // CHECK-NOT: !llvm.mem.parallel_loop_access 335 a[i] = b[i] * c[i] * d[i]; 336 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}} 337 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1 338 // CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]] 339 // CHECK-NEXT: br label %{{.+}} 340 } 341 // CHECK: [[LOOP1_END]] 342 // CHECK: [[O_LOOP1_END]] 343 // CHECK: call {{.+}} @__kmpc_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]]) 344 // CHECK: ret void 345 } 346 347 // CHECK-LABEL: test_precond 348 void test_precond() { 349 // CHECK: [[A_ADDR:%.+]] = alloca i8, 350 // CHECK: [[I_ADDR:%.+]] = alloca i8, 351 // CHECK: [[CAP:%.+]] = alloca i8, 352 char a = 0; 353 // CHECK: store i8 0, 354 // CHECK: store i32 355 // CHECK: store i8 356 // CHECK: [[A:%.+]] = load i8, i8* [[CAP]], 357 // CHECK: [[CONV:%.+]] = sext i8 [[A]] to i32 358 // CHECK: [[CMP:%.+]] = icmp slt i32 [[CONV]], 10 359 // CHECK: br i1 [[CMP]], label %[[PRECOND_THEN:[^,]+]], label %[[PRECOND_END:[^,]+]] 360 // CHECK: [[PRECOND_THEN]] 361 // CHECK: call void @__kmpc_for_static_init_4 362 #pragma omp for 363 for(char i = a; i < 10; ++i); 364 // CHECK: call void @__kmpc_for_static_fini 365 // CHECK: [[PRECOND_END]] 366 } 367 368 // TERM_DEBUG-LABEL: foo 369 int foo() {return 0;}; 370 371 // TERM_DEBUG-LABEL: parallel_for 372 void parallel_for(float *a) { 373 #pragma omp parallel 374 #pragma omp for schedule(static, 5) 375 // TERM_DEBUG-NOT: __kmpc_global_thread_num 376 // TERM_DEBUG: call void @__kmpc_for_static_init_4u({{.+}}), !dbg [[DBG_LOC_START:![0-9]+]] 377 // TERM_DEBUG: invoke i32 {{.*}}foo{{.*}}() 378 // TERM_DEBUG: unwind label %[[TERM_LPAD:.+]], 379 // TERM_DEBUG-NOT: __kmpc_global_thread_num 380 // TERM_DEBUG: call void @__kmpc_for_static_fini({{.+}}), !dbg [[DBG_LOC_END:![0-9]+]] 381 // TERM_DEBUG: call {{.+}} @__kmpc_barrier({{.+}}), !dbg [[DBG_LOC_CANCEL:![0-9]+]] 382 // TERM_DEBUG: [[TERM_LPAD]] 383 // TERM_DEBUG: call void @__clang_call_terminate 384 // TERM_DEBUG: unreachable 385 for (unsigned i = 131071; i <= 2147483647; i += 127) 386 a[i] += foo(); 387 } 388 // Check source line corresponds to "#pragma omp for schedule(static, 5)" above: 389 // TERM_DEBUG-DAG: [[DBG_LOC_START]] = !DILocation(line: [[@LINE-15]], 390 // TERM_DEBUG-DAG: [[DBG_LOC_END]] = !DILocation(line: [[@LINE-16]], 391 // TERM_DEBUG-DAG: [[DBG_LOC_CANCEL]] = !DILocation(line: [[@LINE-17]], 392 393 char i = 1, j = 2, k = 3; 394 // CHECK-LABEL: for_with_global_lcv 395 void for_with_global_lcv() { 396 // CHECK: alloca i8, 397 // CHECK: [[I_ADDR:%.+]] = alloca i8, 398 // CHECK: alloca i8, 399 // CHECK: [[J_ADDR:%.+]] = alloca i8, 400 401 // CHECK: call void @__kmpc_for_static_init_4( 402 // CHECK-NOT: [[I]] 403 // CHECK: store i8 %{{.+}}, i8* [[I_ADDR]] 404 // CHECK-NOT: [[I]] 405 // CHECK: [[I_VAL:%.+]] = load i8, i8* [[I_ADDR]], 406 // CHECK-NOT: [[I]] 407 // CHECK: store i8 [[I_VAL]], i8* [[K]] 408 // CHECK-NOT: [[I]] 409 // CHECK: call void @__kmpc_for_static_fini( 410 // CHECK: call void @__kmpc_barrier( 411 #pragma omp for 412 for (i = 0; i < 2; ++i) { 413 k = i; 414 } 415 // CHECK: call void @__kmpc_for_static_init_4( 416 // CHECK-NOT: [[J]] 417 // CHECK: store i8 %{{.+}}, i8* [[J_ADDR]] 418 // CHECK-NOT: [[J]] 419 // CHECK: [[J_VAL:%.+]] = load i8, i8* [[J_ADDR]], 420 // CHECK-NOT: [[J]] 421 // CHECK: store i8 [[J_VAL]], i8* [[K]] 422 // CHECK-NOT: [[J]] 423 // CHECK: call void @__kmpc_for_static_fini( 424 #pragma omp for collapse(2) 425 for (int i = 0; i < 2; ++i) 426 for (j = 0; j < 2; ++j) { 427 k = i; 428 k = j; 429 } 430 char &cnt = i; 431 #pragma omp for 432 for (cnt = 0; cnt < 2; ++cnt) 433 k = cnt; 434 } 435 436 // CHECK-LABEL: for_with_references 437 void for_with_references() { 438 // CHECK: [[I:%.+]] = alloca i8, 439 // CHECK: [[CNT:%.+]] = alloca i8*, 440 // CHECK: [[CNT_PRIV:%.+]] = alloca i8, 441 // CHECK: call void @__kmpc_for_static_init_4( 442 // CHECK-NOT: load i8, i8* [[CNT]], 443 // CHECK: call void @__kmpc_for_static_fini( 444 char i = 0; 445 char &cnt = i; 446 #pragma omp for 447 for (cnt = 0; cnt < 2; ++cnt) 448 k = cnt; 449 } 450 451 struct Bool { 452 Bool(bool b) : b(b) {} 453 operator bool() const { return b; } 454 const bool b; 455 }; 456 457 template <typename T> 458 struct It { 459 It() : p(0) {} 460 It(const It &, int = 0) ; 461 template <typename U> 462 It(U &, int = 0) ; 463 It &operator=(const It &); 464 It &operator=(It &); 465 ~It() {} 466 467 It(T *p) : p(p) {} 468 469 operator T *&() { return p; } 470 operator T *() const { return p; } 471 T *operator->() const { return p; } 472 473 It &operator++() { ++p; return *this; } 474 It &operator--() { --p; return *this; } 475 It &operator+=(unsigned n) { p += n; return *this; } 476 It &operator-=(unsigned n) { p -= n; return *this; } 477 478 T *p; 479 }; 480 481 template <typename T> 482 It<T> operator+(It<T> a, typename It<T>::difference_type n) { return a.p + n; } 483 484 template <typename T> 485 It<T> operator+(typename It<T>::difference_type n, It<T> a) { return a.p + n; } 486 487 template <typename T> 488 It<T> operator-(It<T> a, typename It<T>::difference_type n) { return a.p - n; } 489 490 typedef Bool BoolType; 491 492 template <typename T> 493 BoolType operator<(It<T> a, It<T> b) { return a.p < b.p; } 494 495 void loop_with_It(It<char> begin, It<char> end) { 496 #pragma omp for 497 for (It<char> it = begin; it < end; ++it) { 498 *it = 0; 499 } 500 } 501 502 // CHECK-LABEL: loop_with_It 503 // CHECK: call i32 @__kmpc_global_thread_num( 504 // CHECK: call void @__kmpc_for_static_init_8( 505 // CHECK: call void @__kmpc_for_static_fini( 506 507 void loop_with_It_plus(It<char> begin, It<char> end) { 508 #pragma omp for 509 for (It<char> it = begin; it < end; it+=1u) { 510 *it = 0; 511 } 512 } 513 514 // CHECK-LABEL: loop_with_It_plus 515 // CHECK: call i32 @__kmpc_global_thread_num( 516 // CHECK: call void @__kmpc_for_static_init_8( 517 // CHECK: call void @__kmpc_for_static_fini( 518 519 void loop_with_stmt_expr() { 520 #pragma omp for 521 for (int i = __extension__({float b = 0;b; }); i < __extension__({double c = 1;c; }); i += __extension__({char d = 1; d; })) 522 ; 523 } 524 // CHECK-LABEL: loop_with_stmt_expr 525 // CHECK: call i32 @__kmpc_global_thread_num( 526 // CHECK: call void @__kmpc_for_static_init_4( 527 // CHECK: call void @__kmpc_for_static_fini( 528 529 530 // CHECK-LABEL: fint 531 // CHECK: call {{.*}}i32 {{.*}}ftemplate 532 // CHECK: ret i32 533 534 // CHECK: load i16, i16* 535 // CHECK: store i16 % 536 // CHECK: call void {{.+}}@__kmpc_fork_call( 537 // CHECK: call void @__kmpc_for_static_init_4( 538 template <typename T> 539 T ftemplate() { 540 short aa = 0; 541 542 #pragma omp parallel for schedule(static, aa) 543 for (int i = 0; i < 100; i++) { 544 } 545 return T(); 546 } 547 548 int fint(void) { return ftemplate<int>(); } 549 550 #endif // HEADER 551