| 1 | /* -*- Mode: C; c-basic-offset:4 ; -*- */ |
|---|
| 2 | /* |
|---|
| 3 | * (C) 2008 by Argonne National Laboratory. |
|---|
| 4 | * See COPYRIGHT in top-level directory. |
|---|
| 5 | */ |
|---|
| 6 | |
|---|
| 7 | #include "hydra.h" |
|---|
| 8 | #include "pmi_handle.h" |
|---|
| 9 | #include "pmi_handle_v1.h" |
|---|
| 10 | #include "pmi_handle_v2.h" |
|---|
| 11 | |
|---|
| 12 | HYD_PMCD_pmi_pg_t *HYD_pg_list = NULL; |
|---|
| 13 | struct HYD_PMCD_pmi_handle *HYD_PMCD_pmi_handle = { 0 }; |
|---|
| 14 | |
|---|
| 15 | struct segment { |
|---|
| 16 | int start_pid; |
|---|
| 17 | int proc_count; |
|---|
| 18 | int node_id; |
|---|
| 19 | struct segment *next; |
|---|
| 20 | }; |
|---|
| 21 | |
|---|
| 22 | struct block { |
|---|
| 23 | int start_node_id; |
|---|
| 24 | int num_blocks; |
|---|
| 25 | int block_size; |
|---|
| 26 | struct block *next; |
|---|
| 27 | }; |
|---|
| 28 | |
|---|
| 29 | static HYD_Status allocate_kvs(HYD_PMCD_pmi_kvs_t ** kvs, int pgid) |
|---|
| 30 | { |
|---|
| 31 | HYD_Status status = HYD_SUCCESS; |
|---|
| 32 | |
|---|
| 33 | HYDU_FUNC_ENTER(); |
|---|
| 34 | |
|---|
| 35 | HYDU_MALLOC(*kvs, HYD_PMCD_pmi_kvs_t *, sizeof(HYD_PMCD_pmi_kvs_t), status); |
|---|
| 36 | HYDU_snprintf((*kvs)->kvs_name, MAXNAMELEN, "kvs_%d_%d", (int) getpid(), pgid); |
|---|
| 37 | (*kvs)->key_pair = NULL; |
|---|
| 38 | |
|---|
| 39 | fn_exit: |
|---|
| 40 | HYDU_FUNC_EXIT(); |
|---|
| 41 | return status; |
|---|
| 42 | |
|---|
| 43 | fn_fail: |
|---|
| 44 | goto fn_exit; |
|---|
| 45 | } |
|---|
| 46 | |
|---|
| 47 | |
|---|
| 48 | static HYD_Status create_pg(HYD_PMCD_pmi_pg_t ** pg, int pgid) |
|---|
| 49 | { |
|---|
| 50 | HYD_Status status = HYD_SUCCESS; |
|---|
| 51 | |
|---|
| 52 | HYDU_FUNC_ENTER(); |
|---|
| 53 | |
|---|
| 54 | HYDU_MALLOC(*pg, HYD_PMCD_pmi_pg_t *, sizeof(HYD_PMCD_pmi_pg_t), status); |
|---|
| 55 | (*pg)->id = pgid; |
|---|
| 56 | (*pg)->num_procs = 0; |
|---|
| 57 | (*pg)->num_subgroups = 0; |
|---|
| 58 | (*pg)->conn_procs = NULL; |
|---|
| 59 | (*pg)->barrier_count = 0; |
|---|
| 60 | (*pg)->node_list = NULL; |
|---|
| 61 | |
|---|
| 62 | status = allocate_kvs(&(*pg)->kvs, pgid); |
|---|
| 63 | HYDU_ERR_POP(status, "unable to allocate kvs space\n"); |
|---|
| 64 | |
|---|
| 65 | (*pg)->next = NULL; |
|---|
| 66 | |
|---|
| 67 | fn_exit: |
|---|
| 68 | HYDU_FUNC_EXIT(); |
|---|
| 69 | return status; |
|---|
| 70 | |
|---|
| 71 | fn_fail: |
|---|
| 72 | goto fn_exit; |
|---|
| 73 | } |
|---|
| 74 | |
|---|
| 75 | |
|---|
| 76 | static HYD_Status free_pmi_process_list(HYD_PMCD_pmi_process_t * process_list) |
|---|
| 77 | { |
|---|
| 78 | HYD_PMCD_pmi_process_t *process, *tmp; |
|---|
| 79 | HYD_Status status = HYD_SUCCESS; |
|---|
| 80 | |
|---|
| 81 | HYDU_FUNC_ENTER(); |
|---|
| 82 | |
|---|
| 83 | process = process_list; |
|---|
| 84 | while (process) { |
|---|
| 85 | tmp = process->next; |
|---|
| 86 | HYDU_FREE(process); |
|---|
| 87 | process = tmp; |
|---|
| 88 | } |
|---|
| 89 | |
|---|
| 90 | HYDU_FUNC_EXIT(); |
|---|
| 91 | return status; |
|---|
| 92 | } |
|---|
| 93 | |
|---|
| 94 | |
|---|
| 95 | static HYD_Status free_pmi_kvs_list(HYD_PMCD_pmi_kvs_t * kvs_list) |
|---|
| 96 | { |
|---|
| 97 | HYD_PMCD_pmi_kvs_pair_t *key_pair, *tmp; |
|---|
| 98 | HYD_Status status = HYD_SUCCESS; |
|---|
| 99 | |
|---|
| 100 | HYDU_FUNC_ENTER(); |
|---|
| 101 | |
|---|
| 102 | key_pair = kvs_list->key_pair; |
|---|
| 103 | while (key_pair) { |
|---|
| 104 | tmp = key_pair->next; |
|---|
| 105 | HYDU_FREE(key_pair); |
|---|
| 106 | key_pair = tmp; |
|---|
| 107 | } |
|---|
| 108 | HYDU_FREE(kvs_list); |
|---|
| 109 | |
|---|
| 110 | HYDU_FUNC_EXIT(); |
|---|
| 111 | return status; |
|---|
| 112 | } |
|---|
| 113 | |
|---|
| 114 | |
|---|
| 115 | static HYD_Status free_pmi_node_list(HYD_PMCD_pmi_node_t * node_list) |
|---|
| 116 | { |
|---|
| 117 | HYD_PMCD_pmi_node_t *node, *tmp; |
|---|
| 118 | HYD_Status status = HYD_SUCCESS; |
|---|
| 119 | |
|---|
| 120 | HYDU_FUNC_ENTER(); |
|---|
| 121 | |
|---|
| 122 | node = node_list; |
|---|
| 123 | while (node) { |
|---|
| 124 | tmp = node->next; |
|---|
| 125 | free_pmi_process_list(node->process_list); |
|---|
| 126 | free_pmi_kvs_list(node->kvs); |
|---|
| 127 | HYDU_FREE(node); |
|---|
| 128 | node = tmp; |
|---|
| 129 | } |
|---|
| 130 | |
|---|
| 131 | HYDU_FUNC_EXIT(); |
|---|
| 132 | return status; |
|---|
| 133 | } |
|---|
| 134 | |
|---|
| 135 | |
|---|
| 136 | static struct HYD_PMCD_pmi_node *allocate_node(HYD_PMCD_pmi_pg_t * pg, int node_id) |
|---|
| 137 | { |
|---|
| 138 | struct HYD_PMCD_pmi_node *node; |
|---|
| 139 | HYD_Status status = HYD_SUCCESS; |
|---|
| 140 | |
|---|
| 141 | HYDU_FUNC_ENTER(); |
|---|
| 142 | |
|---|
| 143 | HYDU_MALLOC(node, HYD_PMCD_pmi_node_t *, sizeof(HYD_PMCD_pmi_node_t), status); |
|---|
| 144 | node->node_id = node_id; |
|---|
| 145 | node->pg = pg; |
|---|
| 146 | node->process_list = NULL; |
|---|
| 147 | node->kvs = NULL; |
|---|
| 148 | node->next = NULL; |
|---|
| 149 | |
|---|
| 150 | fn_exit: |
|---|
| 151 | HYDU_FUNC_EXIT(); |
|---|
| 152 | return node; |
|---|
| 153 | |
|---|
| 154 | fn_fail: |
|---|
| 155 | node = NULL; |
|---|
| 156 | goto fn_exit; |
|---|
| 157 | } |
|---|
| 158 | |
|---|
| 159 | |
|---|
| 160 | HYD_Status HYD_PMCD_pmi_add_kvs(const char *key, char *val, HYD_PMCD_pmi_kvs_t * kvs, |
|---|
| 161 | char **key_pair_str, int *ret) |
|---|
| 162 | { |
|---|
| 163 | HYD_PMCD_pmi_kvs_pair_t *key_pair, *run; |
|---|
| 164 | HYD_Status status = HYD_SUCCESS; |
|---|
| 165 | |
|---|
| 166 | HYDU_FUNC_ENTER(); |
|---|
| 167 | |
|---|
| 168 | HYDU_MALLOC(key_pair, HYD_PMCD_pmi_kvs_pair_t *, sizeof(HYD_PMCD_pmi_kvs_pair_t), status); |
|---|
| 169 | HYDU_snprintf(key_pair->key, MAXKEYLEN, "%s", key); |
|---|
| 170 | HYDU_snprintf(key_pair->val, MAXVALLEN, "%s", val); |
|---|
| 171 | key_pair->next = NULL; |
|---|
| 172 | |
|---|
| 173 | key_pair_str = NULL; |
|---|
| 174 | *ret = 0; |
|---|
| 175 | |
|---|
| 176 | if (kvs->key_pair == NULL) { |
|---|
| 177 | kvs->key_pair = key_pair; |
|---|
| 178 | } |
|---|
| 179 | else { |
|---|
| 180 | run = kvs->key_pair; |
|---|
| 181 | while (run->next) { |
|---|
| 182 | if (!strcmp(run->key, key_pair->key)) { |
|---|
| 183 | /* duplicate key found */ |
|---|
| 184 | *key_pair_str = HYDU_strdup(key_pair->key); |
|---|
| 185 | *ret = -1; |
|---|
| 186 | break; |
|---|
| 187 | } |
|---|
| 188 | run = run->next; |
|---|
| 189 | } |
|---|
| 190 | run->next = key_pair; |
|---|
| 191 | } |
|---|
| 192 | |
|---|
| 193 | fn_exit: |
|---|
| 194 | HYDU_FUNC_EXIT(); |
|---|
| 195 | return status; |
|---|
| 196 | |
|---|
| 197 | fn_fail: |
|---|
| 198 | goto fn_exit; |
|---|
| 199 | } |
|---|
| 200 | |
|---|
| 201 | |
|---|
| 202 | HYD_Status HYD_PMCD_pmi_id_to_rank(int id, int *rank) |
|---|
| 203 | { |
|---|
| 204 | HYD_Status status = HYD_SUCCESS; |
|---|
| 205 | |
|---|
| 206 | HYDU_FUNC_ENTER(); |
|---|
| 207 | |
|---|
| 208 | if (HYD_handle.ranks_per_proc == -1) { |
|---|
| 209 | /* If multiple procs per rank is not defined, use ID as the rank */ |
|---|
| 210 | *rank = id; |
|---|
| 211 | } |
|---|
| 212 | else { |
|---|
| 213 | *rank = (id * HYD_handle.ranks_per_proc) + HYD_pg_list->conn_procs[id]; |
|---|
| 214 | HYD_pg_list->conn_procs[id]++; |
|---|
| 215 | } |
|---|
| 216 | |
|---|
| 217 | HYDU_FUNC_EXIT(); |
|---|
| 218 | return status; |
|---|
| 219 | } |
|---|
| 220 | |
|---|
| 221 | |
|---|
| 222 | HYD_Status HYD_PMCD_pmi_process_mapping(HYD_PMCD_pmi_process_t * process, |
|---|
| 223 | enum HYD_PMCD_pmi_process_mapping_type type, |
|---|
| 224 | char **process_mapping_str) |
|---|
| 225 | { |
|---|
| 226 | int i, j, k, node_id, *process_mapping; |
|---|
| 227 | char *tmp[HYD_NUM_TMP_STRINGS]; |
|---|
| 228 | struct HYD_Partition *partition; |
|---|
| 229 | struct HYD_Partition_segment *segment; |
|---|
| 230 | struct segment *seglist_head, *seglist_tail = NULL, *seg; |
|---|
| 231 | struct block *blocklist_head, *blocklist_tail = NULL, *block; |
|---|
| 232 | int done; |
|---|
| 233 | HYD_Status status = HYD_SUCCESS; |
|---|
| 234 | |
|---|
| 235 | HYDU_FUNC_ENTER(); |
|---|
| 236 | |
|---|
| 237 | seglist_head = NULL; |
|---|
| 238 | node_id = -1; |
|---|
| 239 | FORALL_PARTITIONS(partition, HYD_handle.partition_list) { |
|---|
| 240 | node_id++; |
|---|
| 241 | for (segment = partition->segment_list; segment; segment = segment->next) { |
|---|
| 242 | HYDU_MALLOC(seg, struct segment *, sizeof(struct segment), status); |
|---|
| 243 | seg->start_pid = segment->start_pid; |
|---|
| 244 | seg->proc_count = segment->proc_count; |
|---|
| 245 | seg->node_id = node_id; |
|---|
| 246 | seg->next = NULL; |
|---|
| 247 | |
|---|
| 248 | if (seglist_head == NULL) { |
|---|
| 249 | seglist_head = seg; |
|---|
| 250 | seglist_tail = seg; |
|---|
| 251 | } |
|---|
| 252 | else { |
|---|
| 253 | seglist_tail->next = seg; |
|---|
| 254 | seglist_tail = seg; |
|---|
| 255 | } |
|---|
| 256 | } |
|---|
| 257 | } |
|---|
| 258 | |
|---|
| 259 | /* Sort the segment list */ |
|---|
| 260 | done = 1; |
|---|
| 261 | while (1) { |
|---|
| 262 | for (seg = seglist_head; seg; seg = seg->next) { |
|---|
| 263 | if (seg->next && (seg->start_pid > seg->next->start_pid)) { |
|---|
| 264 | seg->start_pid = seg->start_pid + seg->next->start_pid; |
|---|
| 265 | seg->next->start_pid = seg->start_pid - seg->next->start_pid; |
|---|
| 266 | seg->start_pid = seg->start_pid - seg->next->start_pid; |
|---|
| 267 | |
|---|
| 268 | seg->proc_count = seg->proc_count + seg->next->proc_count; |
|---|
| 269 | seg->next->proc_count = seg->proc_count - seg->next->proc_count; |
|---|
| 270 | seg->proc_count = seg->proc_count - seg->next->proc_count; |
|---|
| 271 | |
|---|
| 272 | seg->node_id = seg->node_id + seg->next->node_id; |
|---|
| 273 | seg->next->node_id = seg->node_id - seg->next->node_id; |
|---|
| 274 | seg->node_id = seg->node_id - seg->next->node_id; |
|---|
| 275 | |
|---|
| 276 | done = 0; |
|---|
| 277 | } |
|---|
| 278 | } |
|---|
| 279 | if (done) |
|---|
| 280 | break; |
|---|
| 281 | } |
|---|
| 282 | |
|---|
| 283 | /* Create a block list off the segment list */ |
|---|
| 284 | blocklist_head = NULL; |
|---|
| 285 | for (seg = seglist_head; seg; seg = seg->next) { |
|---|
| 286 | if (blocklist_head == NULL) { |
|---|
| 287 | HYDU_MALLOC(blocklist_head, struct block *, sizeof(struct block), status); |
|---|
| 288 | blocklist_head->start_node_id = seg->node_id; |
|---|
| 289 | blocklist_head->block_size = seg->proc_count; |
|---|
| 290 | blocklist_head->num_blocks = 1; |
|---|
| 291 | blocklist_head->next = NULL; |
|---|
| 292 | blocklist_tail = blocklist_head; |
|---|
| 293 | } |
|---|
| 294 | else if ((blocklist_tail->start_node_id + blocklist_tail->num_blocks == seg->node_id) |
|---|
| 295 | && (blocklist_tail->block_size == seg->proc_count)) { |
|---|
| 296 | blocklist_tail->num_blocks++; |
|---|
| 297 | } |
|---|
| 298 | else { |
|---|
| 299 | HYDU_MALLOC(blocklist_tail->next, struct block *, sizeof(struct block), status); |
|---|
| 300 | blocklist_tail = blocklist_tail->next; |
|---|
| 301 | blocklist_tail->start_node_id = seg->node_id; |
|---|
| 302 | blocklist_tail->block_size = seg->proc_count; |
|---|
| 303 | blocklist_tail->num_blocks = 1; |
|---|
| 304 | blocklist_tail->next = NULL; |
|---|
| 305 | } |
|---|
| 306 | } |
|---|
| 307 | |
|---|
| 308 | if (type == HYD_PMCD_pmi_explicit) { |
|---|
| 309 | /* Explicit process mapping */ |
|---|
| 310 | HYDU_MALLOC(process_mapping, int *, process->node->pg->num_procs * sizeof(int), |
|---|
| 311 | status); |
|---|
| 312 | |
|---|
| 313 | k = 0; |
|---|
| 314 | for (block = blocklist_head; block; block = block->next) |
|---|
| 315 | for (i = 0; i < block->num_blocks; i++) |
|---|
| 316 | for (j = 0; j < block->block_size; j++) |
|---|
| 317 | process_mapping[k++] = block->start_node_id + i; |
|---|
| 318 | |
|---|
| 319 | i = 0; |
|---|
| 320 | tmp[i++] = HYDU_strdup("explicit,"); |
|---|
| 321 | for (j = 0; j < k; j++) { |
|---|
| 322 | tmp[i++] = HYDU_int_to_str(process_mapping[j]); |
|---|
| 323 | if (j < k - 1) |
|---|
| 324 | tmp[i++] = HYDU_strdup(","); |
|---|
| 325 | HYDU_STRLIST_CONSOLIDATE(tmp, i, status); |
|---|
| 326 | } |
|---|
| 327 | tmp[i++] = NULL; |
|---|
| 328 | |
|---|
| 329 | status = HYDU_str_alloc_and_join(tmp, process_mapping_str); |
|---|
| 330 | HYDU_ERR_POP(status, "error while joining strings\n"); |
|---|
| 331 | |
|---|
| 332 | HYDU_free_strlist(tmp); |
|---|
| 333 | } |
|---|
| 334 | else if (type == HYD_PMCD_pmi_vector) { |
|---|
| 335 | i = 0; |
|---|
| 336 | tmp[i++] = HYDU_strdup("("); |
|---|
| 337 | tmp[i++] = HYDU_strdup("vector,"); |
|---|
| 338 | for (block = blocklist_head; block; block = block->next) { |
|---|
| 339 | tmp[i++] = HYDU_strdup("("); |
|---|
| 340 | tmp[i++] = HYDU_int_to_str(block->start_node_id); |
|---|
| 341 | tmp[i++] = HYDU_strdup(","); |
|---|
| 342 | tmp[i++] = HYDU_int_to_str(block->num_blocks); |
|---|
| 343 | tmp[i++] = HYDU_strdup(","); |
|---|
| 344 | tmp[i++] = HYDU_int_to_str(block->block_size); |
|---|
| 345 | if (block->next) |
|---|
| 346 | tmp[i++] = HYDU_strdup(","); |
|---|
| 347 | tmp[i++] = HYDU_strdup(")"); |
|---|
| 348 | HYDU_STRLIST_CONSOLIDATE(tmp, i, status); |
|---|
| 349 | } |
|---|
| 350 | tmp[i++] = HYDU_strdup(")"); |
|---|
| 351 | tmp[i++] = NULL; |
|---|
| 352 | |
|---|
| 353 | status = HYDU_str_alloc_and_join(tmp, process_mapping_str); |
|---|
| 354 | HYDU_ERR_POP(status, "error while joining strings\n"); |
|---|
| 355 | |
|---|
| 356 | HYDU_free_strlist(tmp); |
|---|
| 357 | } |
|---|
| 358 | else { |
|---|
| 359 | HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "unrecognized process mapping\n"); |
|---|
| 360 | } |
|---|
| 361 | |
|---|
| 362 | fn_exit: |
|---|
| 363 | HYDU_FUNC_EXIT(); |
|---|
| 364 | return status; |
|---|
| 365 | |
|---|
| 366 | fn_fail: |
|---|
| 367 | goto fn_exit; |
|---|
| 368 | } |
|---|
| 369 | |
|---|
| 370 | |
|---|
| 371 | static struct HYD_PMCD_pmi_node *find_node(HYD_PMCD_pmi_pg_t * pg, int rank) |
|---|
| 372 | { |
|---|
| 373 | int found = 0, node_id, srank; |
|---|
| 374 | struct HYD_Partition *partition; |
|---|
| 375 | struct HYD_Partition_segment *segment; |
|---|
| 376 | struct HYD_PMCD_pmi_node *node, *tmp; |
|---|
| 377 | HYD_Status status = HYD_SUCCESS; |
|---|
| 378 | |
|---|
| 379 | srank = rank % HYD_handle.global_core_count; |
|---|
| 380 | |
|---|
| 381 | node_id = 0; |
|---|
| 382 | FORALL_PARTITIONS(partition, HYD_handle.partition_list) { |
|---|
| 383 | for (segment = partition->segment_list; segment; segment = segment->next) { |
|---|
| 384 | if ((srank >= segment->start_pid) && |
|---|
| 385 | (srank < (segment->start_pid + segment->proc_count))) { |
|---|
| 386 | /* We found our rank */ |
|---|
| 387 | found = 1; |
|---|
| 388 | break; |
|---|
| 389 | } |
|---|
| 390 | } |
|---|
| 391 | if (found) |
|---|
| 392 | break; |
|---|
| 393 | node_id++; |
|---|
| 394 | } |
|---|
| 395 | |
|---|
| 396 | /* See if the node already exists */ |
|---|
| 397 | for (node = pg->node_list; node; node = node->next) |
|---|
| 398 | if (node->node_id == node_id) |
|---|
| 399 | return node; |
|---|
| 400 | |
|---|
| 401 | /* allocate node-level KVS space */ |
|---|
| 402 | node = allocate_node(pg, node_id); |
|---|
| 403 | HYDU_ERR_CHKANDJUMP(status, !node, HYD_INTERNAL_ERROR, "unable to allocate PMI node\n"); |
|---|
| 404 | |
|---|
| 405 | status = allocate_kvs(&node->kvs, 0); |
|---|
| 406 | HYDU_ERR_POP(status, "unable to allocate kvs space\n"); |
|---|
| 407 | |
|---|
| 408 | if (pg->node_list == NULL) |
|---|
| 409 | pg->node_list = node; |
|---|
| 410 | else { |
|---|
| 411 | tmp = pg->node_list; |
|---|
| 412 | while (tmp->next) |
|---|
| 413 | tmp = tmp->next; |
|---|
| 414 | tmp->next = node; |
|---|
| 415 | } |
|---|
| 416 | |
|---|
| 417 | fn_exit: |
|---|
| 418 | return node; |
|---|
| 419 | |
|---|
| 420 | fn_fail: |
|---|
| 421 | goto fn_exit; |
|---|
| 422 | } |
|---|
| 423 | |
|---|
| 424 | HYD_Status HYD_PMCD_pmi_add_process_to_pg(HYD_PMCD_pmi_pg_t * pg, int fd, int rank) |
|---|
| 425 | { |
|---|
| 426 | HYD_PMCD_pmi_process_t *process, *tmp; |
|---|
| 427 | struct HYD_PMCD_pmi_node *node; |
|---|
| 428 | HYD_Status status = HYD_SUCCESS; |
|---|
| 429 | |
|---|
| 430 | HYDU_FUNC_ENTER(); |
|---|
| 431 | |
|---|
| 432 | /* Find the node corresponding to the rank */ |
|---|
| 433 | node = find_node(pg, rank); |
|---|
| 434 | |
|---|
| 435 | /* Add process to the node */ |
|---|
| 436 | HYDU_MALLOC(process, HYD_PMCD_pmi_process_t *, sizeof(HYD_PMCD_pmi_process_t), status); |
|---|
| 437 | process->fd = fd; |
|---|
| 438 | process->rank = rank; |
|---|
| 439 | process->epoch = 0; |
|---|
| 440 | process->node = node; |
|---|
| 441 | process->next = NULL; |
|---|
| 442 | if (node->process_list == NULL) |
|---|
| 443 | node->process_list = process; |
|---|
| 444 | else { |
|---|
| 445 | tmp = node->process_list; |
|---|
| 446 | while (tmp->next) |
|---|
| 447 | tmp = tmp->next; |
|---|
| 448 | tmp->next = process; |
|---|
| 449 | } |
|---|
| 450 | |
|---|
| 451 | fn_exit: |
|---|
| 452 | HYDU_FUNC_EXIT(); |
|---|
| 453 | return status; |
|---|
| 454 | |
|---|
| 455 | fn_fail: |
|---|
| 456 | goto fn_exit; |
|---|
| 457 | } |
|---|
| 458 | |
|---|
| 459 | |
|---|
| 460 | HYD_PMCD_pmi_process_t *HYD_PMCD_pmi_find_process(int fd) |
|---|
| 461 | { |
|---|
| 462 | HYD_PMCD_pmi_pg_t *pg; |
|---|
| 463 | HYD_PMCD_pmi_node_t *node; |
|---|
| 464 | HYD_PMCD_pmi_process_t *process = NULL; |
|---|
| 465 | |
|---|
| 466 | for (pg = HYD_pg_list; pg; pg = pg->next) { |
|---|
| 467 | for (node = pg->node_list; node; node = node->next) { |
|---|
| 468 | for (process = node->process_list; process; process = process->next) { |
|---|
| 469 | if (process->fd == fd) |
|---|
| 470 | return process; |
|---|
| 471 | } |
|---|
| 472 | } |
|---|
| 473 | } |
|---|
| 474 | |
|---|
| 475 | return NULL; |
|---|
| 476 | } |
|---|
| 477 | |
|---|
| 478 | |
|---|
| 479 | HYD_Status HYD_PMCD_pmi_init(void) |
|---|
| 480 | { |
|---|
| 481 | struct HYD_Partition *partition; |
|---|
| 482 | struct HYD_Partition_exec *exec; |
|---|
| 483 | int i; |
|---|
| 484 | HYD_Status status = HYD_SUCCESS; |
|---|
| 485 | |
|---|
| 486 | HYDU_FUNC_ENTER(); |
|---|
| 487 | |
|---|
| 488 | status = create_pg(&HYD_pg_list, 0); |
|---|
| 489 | HYDU_ERR_POP(status, "unable to create pg\n"); |
|---|
| 490 | |
|---|
| 491 | /* Find the number of processes in the PG */ |
|---|
| 492 | HYD_pg_list->num_subgroups = 0; |
|---|
| 493 | FORALL_ACTIVE_PARTITIONS(partition, HYD_handle.partition_list) { |
|---|
| 494 | for (exec = partition->exec_list; exec; exec = exec->next) |
|---|
| 495 | HYD_pg_list->num_subgroups += exec->proc_count; |
|---|
| 496 | } |
|---|
| 497 | |
|---|
| 498 | if (HYD_handle.ranks_per_proc != -1) |
|---|
| 499 | HYD_pg_list->num_procs = HYD_pg_list->num_subgroups * HYD_handle.ranks_per_proc; |
|---|
| 500 | else |
|---|
| 501 | HYD_pg_list->num_procs = HYD_pg_list->num_subgroups; |
|---|
| 502 | |
|---|
| 503 | /* Allocate and initialize the connected ranks */ |
|---|
| 504 | HYDU_MALLOC(HYD_pg_list->conn_procs, int *, HYD_pg_list->num_subgroups * sizeof(int), status); |
|---|
| 505 | for (i = 0; i < HYD_pg_list->num_subgroups; i++) |
|---|
| 506 | HYD_pg_list->conn_procs[i] = 0; |
|---|
| 507 | |
|---|
| 508 | fn_exit: |
|---|
| 509 | HYDU_FUNC_EXIT(); |
|---|
| 510 | return status; |
|---|
| 511 | |
|---|
| 512 | fn_fail: |
|---|
| 513 | goto fn_exit; |
|---|
| 514 | } |
|---|
| 515 | |
|---|
| 516 | |
|---|
| 517 | HYD_Status HYD_PMCD_pmi_finalize(void) |
|---|
| 518 | { |
|---|
| 519 | HYD_PMCD_pmi_pg_t *pg, *tmp; |
|---|
| 520 | HYD_Status status = HYD_SUCCESS; |
|---|
| 521 | |
|---|
| 522 | HYDU_FUNC_ENTER(); |
|---|
| 523 | |
|---|
| 524 | pg = HYD_pg_list; |
|---|
| 525 | while (pg) { |
|---|
| 526 | tmp = pg->next; |
|---|
| 527 | |
|---|
| 528 | if (pg->conn_procs) |
|---|
| 529 | HYDU_FREE(pg->conn_procs); |
|---|
| 530 | |
|---|
| 531 | status = free_pmi_node_list(pg->node_list); |
|---|
| 532 | HYDU_ERR_POP(status, "unable to free process list\n"); |
|---|
| 533 | |
|---|
| 534 | status = free_pmi_kvs_list(pg->kvs); |
|---|
| 535 | HYDU_ERR_POP(status, "unable to free kvs list\n"); |
|---|
| 536 | |
|---|
| 537 | HYDU_FREE(pg); |
|---|
| 538 | pg = tmp; |
|---|
| 539 | } |
|---|
| 540 | |
|---|
| 541 | fn_exit: |
|---|
| 542 | HYDU_FUNC_EXIT(); |
|---|
| 543 | return status; |
|---|
| 544 | |
|---|
| 545 | fn_fail: |
|---|
| 546 | goto fn_exit; |
|---|
| 547 | } |
|---|