| 1 | // RUN--: %libarcher-compile-and-run | FileCheck %s --check-prefix=NOENV |
| 2 | // RUN: %libarcher-compile && env ARCHER_OPTIONS="all_memory=1" \ |
| 3 | // RUN: %libarcher-run | FileCheck %s --check-prefix=ENV |
| 4 | // REQUIRES: tsan |
| 5 | |
| 6 | // The runtime currently does not get dependency information from GCC. |
| 7 | // UNSUPPORTED: gcc |
| 8 | |
| 9 | // Tests OMP 5.x task dependence "omp_all_memory", |
| 10 | // emulates compiler codegen versions for new dep kind |
| 11 | // |
| 12 | // Task tree created: |
| 13 | // task0 - task1 (in: i1, i2) |
| 14 | // \ |
| 15 | // task2 (inoutset: i2), (in: i1) |
| 16 | // / |
| 17 | // task3 (omp_all_memory) via flag=0x80 |
| 18 | // / |
| 19 | // task4 - task5 (in: i1, i2) |
| 20 | // / |
| 21 | // task6 (omp_all_memory) via addr=-1 |
| 22 | // / |
| 23 | // task7 (omp_all_memory) via flag=0x80 |
| 24 | // / |
| 25 | // task8 (in: i3) |
| 26 | // |
| 27 | |
| 28 | #include <omp.h> |
| 29 | #include <stdio.h> |
| 30 | |
| 31 | #ifdef _WIN32 |
| 32 | #include <windows.h> |
| 33 | #define mysleep(n) Sleep(n) |
| 34 | #else |
| 35 | #include <unistd.h> |
| 36 | #define mysleep(n) usleep((n)*1000) |
| 37 | #endif |
| 38 | |
| 39 | // to check the # of concurrent tasks (must be 1 for MTX, <3 for other kinds) |
| 40 | static int checker = 0; |
| 41 | static int err = 0; |
| 42 | #ifndef DELAY |
| 43 | #define DELAY 100 |
| 44 | #endif |
| 45 | |
| 46 | // --------------------------------------------------------------------------- |
| 47 | // internal data to emulate compiler codegen |
| 48 | typedef struct DEP { |
| 49 | size_t addr; |
| 50 | size_t len; |
| 51 | unsigned char flags; |
| 52 | } dep; |
| 53 | #define DEP_ALL_MEM 0x80 |
| 54 | typedef struct task { |
| 55 | void **shareds; |
| 56 | void *entry; |
| 57 | int part_id; |
| 58 | void *destr_thunk; |
| 59 | int priority; |
| 60 | long long device_id; |
| 61 | int f_priv; |
| 62 | } task_t; |
| 63 | #define TIED 1 |
| 64 | typedef int (*entry_t)(int, task_t *); |
| 65 | typedef struct ID { |
| 66 | int reserved_1; |
| 67 | int flags; |
| 68 | int reserved_2; |
| 69 | int reserved_3; |
| 70 | char *psource; |
| 71 | } id; |
| 72 | // thunk routine for tasks with ALL dependency |
| 73 | int thunk_m(int gtid, task_t *ptask) { |
| 74 | int lcheck, th; |
| 75 | #pragma omp atomic capture |
| 76 | lcheck = ++checker; |
| 77 | th = omp_get_thread_num(); |
| 78 | printf(format: "task m_%d, th %d, checker %d\n" , ptask->f_priv, th, lcheck); |
| 79 | if (lcheck != 1) { // no more than 1 task at a time |
| 80 | err++; |
| 81 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
| 82 | } |
| 83 | mysleep(DELAY); |
| 84 | #pragma omp atomic read |
| 85 | lcheck = checker; // must still be equal to 1 |
| 86 | if (lcheck != 1) { |
| 87 | err++; |
| 88 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
| 89 | } |
| 90 | #pragma omp atomic |
| 91 | --checker; |
| 92 | return 0; |
| 93 | } |
| 94 | // thunk routine for tasks with inoutset dependency |
| 95 | int thunk_s(int gtid, task_t *ptask) { |
| 96 | int lcheck, th; |
| 97 | #pragma omp atomic capture |
| 98 | lcheck = ++checker; // 1 |
| 99 | th = omp_get_thread_num(); |
| 100 | printf(format: "task 2_%d, th %d, checker %d\n" , ptask->f_priv, th, lcheck); |
| 101 | if (lcheck != 1) { // no more than 1 task at a time |
| 102 | err++; |
| 103 | printf(format: "Error s1, checker %d != 1\n" , lcheck); |
| 104 | } |
| 105 | mysleep(DELAY); |
| 106 | #pragma omp atomic read |
| 107 | lcheck = checker; // must still be equal to 1 |
| 108 | if (lcheck != 1) { |
| 109 | err++; |
| 110 | printf(format: "Error s2, checker %d != 1\n" , lcheck); |
| 111 | } |
| 112 | #pragma omp atomic |
| 113 | --checker; |
| 114 | return 0; |
| 115 | } |
| 116 | |
| 117 | #ifdef __cplusplus |
| 118 | extern "C" { |
| 119 | #endif |
| 120 | int __kmpc_global_thread_num(id *); |
| 121 | task_t *__kmpc_omp_task_alloc(id *loc, int gtid, int flags, size_t sz, |
| 122 | size_t shar, entry_t rtn); |
| 123 | int __kmpc_omp_task_with_deps(id *loc, int gtid, task_t *task, int ndeps, |
| 124 | dep *dep_lst, int nd_noalias, dep *noalias_lst); |
| 125 | static id loc = {.reserved_1: 0, .flags: 2, .reserved_2: 0, .reserved_3: 0, .psource: ";file;func;0;0;;" }; |
| 126 | #ifdef __cplusplus |
| 127 | } // extern "C" |
| 128 | #endif |
| 129 | // End of internal data |
| 130 | // --------------------------------------------------------------------------- |
| 131 | |
| 132 | int main() { |
| 133 | char *ompx_all_memory = (void *)0xffffffffffffffff; |
| 134 | int i1, i2, i3; |
| 135 | omp_set_num_threads(8); |
| 136 | omp_set_dynamic(0); |
| 137 | #pragma omp parallel |
| 138 | { |
| 139 | #pragma omp single nowait |
| 140 | { |
| 141 | dep sdep[2]; |
| 142 | task_t *ptr; |
| 143 | int gtid = __kmpc_global_thread_num(&loc); |
| 144 | int t = omp_get_thread_num(); |
| 145 | #pragma omp task depend(in : i1, i2) |
| 146 | { // task 0 |
| 147 | int lcheck, th; |
| 148 | #pragma omp atomic capture |
| 149 | lcheck = ++checker; // 1 or 2 |
| 150 | th = omp_get_thread_num(); |
| 151 | printf(format: "task 0_%d, th %d, checker %d\n" , t, th, lcheck); |
| 152 | if (lcheck > 2 || lcheck < 1) { |
| 153 | err++; // no more than 2 tasks concurrently |
| 154 | printf(format: "Error1, checker %d, not 1 or 2\n" , lcheck); |
| 155 | } |
| 156 | mysleep(DELAY); |
| 157 | #pragma omp atomic read |
| 158 | lcheck = checker; // 1 or 2 |
| 159 | if (lcheck > 2 || lcheck < 1) { |
| 160 | #pragma omp atomic |
| 161 | err++; |
| 162 | printf(format: "Error2, checker %d, not 1 or 2\n" , lcheck); |
| 163 | } |
| 164 | #pragma omp atomic |
| 165 | --checker; |
| 166 | } |
| 167 | #pragma omp task depend(in : i1, i2) |
| 168 | { // task 1 |
| 169 | int lcheck, th; |
| 170 | #pragma omp atomic capture |
| 171 | lcheck = ++checker; // 1 or 2 |
| 172 | th = omp_get_thread_num(); |
| 173 | printf(format: "task 1_%d, th %d, checker %d\n" , t, th, lcheck); |
| 174 | if (lcheck > 2 || lcheck < 1) { |
| 175 | err++; // no more than 2 tasks concurrently |
| 176 | printf(format: "Error3, checker %d, not 1 or 2\n" , lcheck); |
| 177 | } |
| 178 | mysleep(DELAY); |
| 179 | #pragma omp atomic read |
| 180 | lcheck = checker; // 1 or 2 |
| 181 | if (lcheck > 2 || lcheck < 1) { |
| 182 | err++; |
| 183 | printf(format: "Error4, checker %d, not 1 or 2\n" , lcheck); |
| 184 | } |
| 185 | #pragma omp atomic |
| 186 | --checker; |
| 187 | } |
| 188 | // compiler codegen start |
| 189 | // task2 |
| 190 | ptr = __kmpc_omp_task_alloc(loc: &loc, gtid, TIED, sz: sizeof(task_t), shar: 0, rtn: thunk_s); |
| 191 | sdep[0].addr = (size_t)&i1; |
| 192 | sdep[0].len = 0; // not used |
| 193 | sdep[0].flags = 1; // IN |
| 194 | sdep[1].addr = (size_t)&i2; |
| 195 | sdep[1].len = 0; // not used |
| 196 | sdep[1].flags = 8; // INOUTSET |
| 197 | ptr->f_priv = t + 10; // init single first-private variable |
| 198 | __kmpc_omp_task_with_deps(loc: &loc, gtid, task: ptr, ndeps: 2, dep_lst: sdep, nd_noalias: 0, noalias_lst: 0); |
| 199 | |
| 200 | // task3 |
| 201 | #pragma omp task depend(in : i1) depend(inout : ompx_all_memory[0]) |
| 202 | { |
| 203 | int lcheck, th; |
| 204 | #pragma omp atomic capture |
| 205 | lcheck = ++checker; |
| 206 | th = omp_get_thread_num(); |
| 207 | printf(format: "task 3_%d, th %d, checker %d\n" , t, th, lcheck); |
| 208 | if (lcheck != 1) { // no more than 1 task at a time |
| 209 | err++; |
| 210 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
| 211 | } |
| 212 | mysleep(DELAY); |
| 213 | #pragma omp atomic read |
| 214 | lcheck = checker; // must still be equal to 1 |
| 215 | if (lcheck != 1) { |
| 216 | err++; |
| 217 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
| 218 | } |
| 219 | #pragma omp atomic |
| 220 | --checker; |
| 221 | } |
| 222 | // compiler codegen end |
| 223 | #pragma omp task depend(in : i1, i2) |
| 224 | { // task 4 |
| 225 | int lcheck, th; |
| 226 | #pragma omp atomic capture |
| 227 | lcheck = ++checker; // 1 or 2 |
| 228 | th = omp_get_thread_num(); |
| 229 | printf(format: "task 4_%d, th %d, checker %d\n" , t, th, lcheck); |
| 230 | if (lcheck > 2 || lcheck < 1) { |
| 231 | err++; // no more than 2 tasks concurrently |
| 232 | printf(format: "Error5, checker %d, not 1 or 2\n" , lcheck); |
| 233 | } |
| 234 | mysleep(DELAY); |
| 235 | #pragma omp atomic read |
| 236 | lcheck = checker; // 1 or 2 |
| 237 | if (lcheck > 2 || lcheck < 1) { |
| 238 | err++; |
| 239 | printf(format: "Error6, checker %d, not 1 or 2\n" , lcheck); |
| 240 | } |
| 241 | #pragma omp atomic |
| 242 | --checker; |
| 243 | } |
| 244 | #pragma omp task depend(in : i1, i2) |
| 245 | { // task 5 |
| 246 | int lcheck, th; |
| 247 | #pragma omp atomic capture |
| 248 | lcheck = ++checker; // 1 or 2 |
| 249 | th = omp_get_thread_num(); |
| 250 | printf(format: "task 5_%d, th %d, checker %d\n" , t, th, lcheck); |
| 251 | if (lcheck > 2 || lcheck < 1) { |
| 252 | err++; // no more than 2 tasks concurrently |
| 253 | printf(format: "Error7, checker %d, not 1 or 2\n" , lcheck); |
| 254 | } |
| 255 | mysleep(DELAY); |
| 256 | #pragma omp atomic read |
| 257 | lcheck = checker; // 1 or 2 |
| 258 | if (lcheck > 2 || lcheck < 1) { |
| 259 | err++; |
| 260 | printf(format: "Error8, checker %d, not 1 or 2\n" , lcheck); |
| 261 | } |
| 262 | #pragma omp atomic |
| 263 | --checker; |
| 264 | } |
| 265 | // task6 |
| 266 | #pragma omp task depend(inout : ompx_all_memory[0]) |
| 267 | { |
| 268 | int lcheck, th; |
| 269 | #pragma omp atomic capture |
| 270 | lcheck = ++checker; |
| 271 | th = omp_get_thread_num(); |
| 272 | printf(format: "task 6_%d, th %d, checker %d\n" , t, th, lcheck); |
| 273 | if (lcheck != 1) { // no more than 1 task at a time |
| 274 | err++; |
| 275 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
| 276 | } |
| 277 | mysleep(DELAY); |
| 278 | #pragma omp atomic read |
| 279 | lcheck = checker; // must still be equal to 1 |
| 280 | if (lcheck != 1) { |
| 281 | err++; |
| 282 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
| 283 | } |
| 284 | #pragma omp atomic |
| 285 | --checker; |
| 286 | } |
| 287 | // task7 |
| 288 | #pragma omp task depend(inout : ompx_all_memory[0]) depend(mutexinoutset : i3) |
| 289 | { |
| 290 | int lcheck, th; |
| 291 | #pragma omp atomic capture |
| 292 | lcheck = ++checker; |
| 293 | th = omp_get_thread_num(); |
| 294 | printf(format: "task 7_%d, th %d, checker %d\n" , t, th, lcheck); |
| 295 | if (lcheck != 1) { // no more than 1 task at a time |
| 296 | err++; |
| 297 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
| 298 | } |
| 299 | mysleep(DELAY); |
| 300 | #pragma omp atomic read |
| 301 | lcheck = checker; // must still be equal to 1 |
| 302 | if (lcheck != 1) { |
| 303 | err++; |
| 304 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
| 305 | } |
| 306 | #pragma omp atomic |
| 307 | --checker; |
| 308 | } |
| 309 | #pragma omp task depend(in : i3) |
| 310 | { // task 8 |
| 311 | int lcheck, th; |
| 312 | #pragma omp atomic capture |
| 313 | lcheck = ++checker; // 1 |
| 314 | th = omp_get_thread_num(); |
| 315 | printf(format: "task 8_%d, th %d, checker %d\n" , t, th, lcheck); |
| 316 | if (lcheck != 1) { |
| 317 | err++; |
| 318 | printf(format: "Error9, checker %d, != 1\n" , lcheck); |
| 319 | } |
| 320 | mysleep(DELAY); |
| 321 | #pragma omp atomic read |
| 322 | lcheck = checker; |
| 323 | if (lcheck != 1) { |
| 324 | err++; |
| 325 | printf(format: "Error10, checker %d, != 1\n" , lcheck); |
| 326 | } |
| 327 | #pragma omp atomic |
| 328 | --checker; |
| 329 | } |
| 330 | } // single |
| 331 | } // parallel |
| 332 | if (err == 0 && checker == 0) { |
| 333 | printf(format: "passed\n" ); |
| 334 | return 0; |
| 335 | } else { |
| 336 | printf(format: "failed, err = %d, checker = %d\n" , err, checker); |
| 337 | return 1; |
| 338 | } |
| 339 | } |
| 340 | |
| 341 | // NOENV-NOT: ThreadSanitizer: data race |
| 342 | // NOENV-NOT: ThreadSanitizer: reported |
| 343 | // NOENV: omp_all_memory |
| 344 | // NOENV-NOT: ThreadSanitizer: data race |
| 345 | // NOENV-NOT: ThreadSanitizer: reported |
| 346 | // NOENV: passed |
| 347 | |
| 348 | // ENV-NOT: ThreadSanitizer: data race |
| 349 | // ENV-NOT: ThreadSanitizer: reported |
| 350 | // ENV: passed |
| 351 | |