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