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