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