[1] | 1 | /* Copyright (c) 2007-2009, Stanford University |
---|
| 2 | * All rights reserved. |
---|
| 3 | * |
---|
| 4 | * Redistribution and use in source and binary forms, with or without |
---|
| 5 | * modification, are permitted provided that the following conditions are met: |
---|
| 6 | * * Redistributions of source code must retain the above copyright |
---|
| 7 | * notice, this list of conditions and the following disclaimer. |
---|
| 8 | * * Redistributions in binary form must reproduce the above copyright |
---|
| 9 | * notice, this list of conditions and the following disclaimer in the |
---|
| 10 | * documentation and/or other materials provided with the distribution. |
---|
| 11 | * * Neither the name of Stanford University nor the names of its |
---|
| 12 | * contributors may be used to endorse or promote products derived from |
---|
| 13 | * this software without specific prior written permission. |
---|
| 14 | * |
---|
| 15 | * THIS SOFTWARE IS PROVIDED BY STANFORD UNIVERSITY ``AS IS'' AND ANY |
---|
| 16 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
---|
| 17 | * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
---|
| 18 | * DISCLAIMED. IN NO EVENT SHALL STANFORD UNIVERSITY BE LIABLE FOR ANY |
---|
| 19 | * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
---|
| 20 | * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
---|
| 21 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
---|
| 22 | * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
---|
| 23 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
---|
| 24 | * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
---|
| 25 | */ |
---|
| 26 | |
---|
| 27 | #include <assert.h> |
---|
| 28 | #include <stdlib.h> |
---|
| 29 | #include <stdio.h> |
---|
| 30 | #include <inttypes.h> |
---|
| 31 | #include <string.h> |
---|
| 32 | |
---|
| 33 | #include "memory.h" |
---|
| 34 | #include "taskQ.h" |
---|
| 35 | #include "queue.h" |
---|
| 36 | #include "synch.h" |
---|
| 37 | #include "locality.h" |
---|
| 38 | |
---|
| 39 | static int num_strands_per_chip = 0; |
---|
| 40 | |
---|
| 41 | typedef struct { |
---|
| 42 | task_t task; |
---|
| 43 | queue_elem_t queue_elem; |
---|
| 44 | } tq_entry_t; |
---|
| 45 | |
---|
| 46 | typedef struct { |
---|
| 47 | mr_lock_t parent; |
---|
| 48 | uintptr_t chksum; |
---|
| 49 | mr_lock_t *per_thread; |
---|
| 50 | } tq_lock_t; |
---|
| 51 | |
---|
| 52 | struct taskQ_t { |
---|
| 53 | int num_queues; |
---|
| 54 | int num_threads; |
---|
| 55 | queue_t **queues; |
---|
| 56 | queue_t **free_queues; |
---|
| 57 | tq_lock_t *locks; |
---|
| 58 | /* putting all seeds together may lead to extra coherence traffic among cpus |
---|
| 59 | * if it's a problem we can pad it by l1 line size */ |
---|
| 60 | /* per-thread random seed */ |
---|
| 61 | unsigned int *seeds; |
---|
| 62 | }; |
---|
| 63 | |
---|
| 64 | typedef int (*dequeue_fn)(taskQ_t *, int, int, queue_elem_t**); |
---|
| 65 | |
---|
| 66 | static inline taskQ_t* tq_init_normal(int numThreads); |
---|
| 67 | static inline void tq_finalize_normal(taskQ_t* tq); |
---|
| 68 | static inline int tq_dequeue_normal( |
---|
| 69 | taskQ_t* tq, task_t* task, int lgrp, int tid); |
---|
| 70 | static inline int tq_dequeue_normal_seq ( |
---|
| 71 | taskQ_t* tq, task_t* task, int lgrp, int tid); |
---|
| 72 | static inline int tq_dequeue_normal_internal ( |
---|
| 73 | taskQ_t* tq, task_t* task, int lgrp, int tid, dequeue_fn dequeue_fn); |
---|
| 74 | |
---|
| 75 | static queue_t* tq_alloc_queue(void); |
---|
| 76 | static void tq_free_queue(queue_t* q); |
---|
| 77 | static int tq_queue_init(taskQ_t* tq, unsigned int idx); |
---|
| 78 | static void tq_queue_destroy(taskQ_t* tq, unsigned int idx); |
---|
| 79 | static void tq_empty_queue(queue_t* q); |
---|
| 80 | |
---|
| 81 | taskQ_t* tq_init (int num_threads) |
---|
| 82 | { |
---|
| 83 | return tq_init_normal(num_threads); |
---|
| 84 | } |
---|
| 85 | |
---|
| 86 | void tq_reset (taskQ_t* tq, int num_threads) |
---|
| 87 | { |
---|
| 88 | } |
---|
| 89 | |
---|
| 90 | /** |
---|
| 91 | * Initialize task queue for a normal machine |
---|
| 92 | */ |
---|
| 93 | static inline taskQ_t* tq_init_normal(int numThreads) |
---|
| 94 | { |
---|
| 95 | int i; |
---|
| 96 | taskQ_t *tq = NULL; |
---|
| 97 | |
---|
| 98 | tq = mem_calloc(1, sizeof(taskQ_t)); |
---|
| 99 | if (tq == NULL) { |
---|
| 100 | return NULL; |
---|
| 101 | } |
---|
| 102 | |
---|
| 103 | /* XXX should this be local? */ |
---|
| 104 | num_strands_per_chip = loc_get_lgrp_size (); |
---|
| 105 | tq->num_threads = numThreads; |
---|
| 106 | tq->num_queues = tq->num_threads / num_strands_per_chip; |
---|
| 107 | |
---|
| 108 | if (tq->num_queues == 0) |
---|
| 109 | tq->num_queues = 1; |
---|
| 110 | |
---|
| 111 | tq->queues = (queue_t **)mem_calloc (tq->num_queues, sizeof (queue_t *)); |
---|
| 112 | if (tq->queues == NULL) goto fail_queues; |
---|
| 113 | |
---|
| 114 | tq->free_queues = (queue_t **)mem_calloc ( |
---|
| 115 | tq->num_queues, sizeof (queue_t *)); |
---|
| 116 | if (tq->free_queues == NULL) goto fail_free_queues; |
---|
| 117 | |
---|
| 118 | tq->locks = (tq_lock_t *)mem_calloc (tq->num_queues, sizeof (tq_lock_t)); |
---|
| 119 | if (tq->locks == NULL) goto fail_locks; |
---|
| 120 | |
---|
| 121 | tq->seeds = (unsigned int*)mem_calloc( |
---|
| 122 | tq->num_threads, sizeof(unsigned int)); |
---|
| 123 | if (tq->seeds == NULL) goto fail_seeds; |
---|
| 124 | mem_memset(tq->seeds, 0, sizeof(unsigned int) * tq->num_threads); |
---|
| 125 | |
---|
| 126 | for (i = 0; i < tq->num_queues; ++i) |
---|
| 127 | if (!tq_queue_init(tq, i)) |
---|
| 128 | goto fail_tq_init; |
---|
| 129 | |
---|
| 130 | return tq; |
---|
| 131 | |
---|
| 132 | fail_tq_init: |
---|
| 133 | /* destroy all queues that have been allocated */ |
---|
| 134 | i--; |
---|
| 135 | while (i >= 0) { |
---|
| 136 | tq_queue_destroy(tq, i); |
---|
| 137 | --i; |
---|
| 138 | } |
---|
| 139 | mem_free(tq->seeds); |
---|
| 140 | fail_seeds: |
---|
| 141 | mem_free(tq->locks); |
---|
| 142 | fail_locks: |
---|
| 143 | mem_free(tq->free_queues); |
---|
| 144 | fail_free_queues: |
---|
| 145 | mem_free(tq->queues); |
---|
| 146 | fail_queues: |
---|
| 147 | mem_free(tq); |
---|
| 148 | return NULL; |
---|
| 149 | } |
---|
| 150 | |
---|
| 151 | /** |
---|
| 152 | * Destroys an initialized queue (i.e. free queue and alloc queue) in task queue |
---|
| 153 | * @param tq tq to index |
---|
| 154 | * @param idx index of queue to destroy in tq |
---|
| 155 | */ |
---|
| 156 | static void tq_queue_destroy(taskQ_t* tq, unsigned int idx) |
---|
| 157 | { |
---|
| 158 | int j; |
---|
| 159 | uintptr_t chksum; |
---|
| 160 | |
---|
| 161 | // assert (idx < tq->num_queues); |
---|
| 162 | |
---|
| 163 | tq_empty_queue(tq->queues[idx]); |
---|
| 164 | tq_free_queue(tq->queues[idx]); |
---|
| 165 | |
---|
| 166 | tq_empty_queue(tq->free_queues[idx]); |
---|
| 167 | tq_free_queue(tq->free_queues[idx]); |
---|
| 168 | |
---|
| 169 | /* free all lock data associated with queue */ |
---|
| 170 | chksum = 0; |
---|
| 171 | for (j = 0; j < tq->num_threads; j++) { |
---|
| 172 | chksum += (uintptr_t)tq->locks[idx].per_thread[j]; |
---|
| 173 | lock_free_per_thread(tq->locks[idx].per_thread[j]); |
---|
| 174 | } |
---|
| 175 | |
---|
| 176 | lock_free (tq->locks[idx].parent); |
---|
| 177 | |
---|
| 178 | mem_free (tq->locks[idx].per_thread); |
---|
| 179 | tq->locks[idx].per_thread = NULL; |
---|
| 180 | } |
---|
| 181 | |
---|
| 182 | /** |
---|
| 183 | * Initialize a queue for a given index in the task queue |
---|
| 184 | * @return zero on failure, nonzero on success |
---|
| 185 | */ |
---|
| 186 | static int tq_queue_init(taskQ_t* tq, unsigned int idx) |
---|
| 187 | { |
---|
| 188 | int j; |
---|
| 189 | |
---|
| 190 | assert (idx < (unsigned)tq->num_queues); |
---|
| 191 | |
---|
| 192 | tq->queues[idx] = tq_alloc_queue(); |
---|
| 193 | if (tq->queues[idx] == NULL) return 0; |
---|
| 194 | |
---|
| 195 | tq->free_queues[idx] = tq_alloc_queue(); |
---|
| 196 | if (tq->free_queues[idx] == NULL) goto fail_free_queue; |
---|
| 197 | |
---|
| 198 | tq->locks[idx].parent = lock_alloc(); |
---|
| 199 | |
---|
| 200 | tq->locks[idx].per_thread = (mr_lock_t *)mem_calloc( |
---|
| 201 | tq->num_threads, sizeof(mr_lock_t)); |
---|
| 202 | if (tq->locks[idx].per_thread == NULL) goto fail_priv_alloc; |
---|
| 203 | |
---|
| 204 | tq->locks[idx].chksum = 0; |
---|
| 205 | for (j = 0; j < tq->num_threads; ++j) { |
---|
| 206 | mr_lock_t per_thread; |
---|
| 207 | per_thread = lock_alloc_per_thread(tq->locks[idx].parent); |
---|
| 208 | tq->locks[idx].per_thread[j] = per_thread; |
---|
| 209 | tq->locks[idx].chksum += (uintptr_t)per_thread; |
---|
| 210 | } |
---|
| 211 | |
---|
| 212 | return 1; |
---|
| 213 | |
---|
| 214 | fail_priv_alloc: |
---|
| 215 | lock_free(&tq->locks[idx].parent); |
---|
| 216 | tq_free_queue(tq->free_queues[idx]); |
---|
| 217 | tq->free_queues[idx] = NULL; |
---|
| 218 | fail_free_queue: |
---|
| 219 | tq_free_queue(tq->queues[idx]); |
---|
| 220 | tq->queues[idx] = NULL; |
---|
| 221 | |
---|
| 222 | return 0; |
---|
| 223 | } |
---|
| 224 | |
---|
| 225 | /** |
---|
| 226 | * Allocates an initialized queue |
---|
| 227 | * @return NULL on failure, initialized queue pointer on success |
---|
| 228 | */ |
---|
| 229 | static queue_t* tq_alloc_queue(void) |
---|
| 230 | { |
---|
| 231 | queue_t *q; |
---|
| 232 | |
---|
| 233 | q = (queue_t*) mem_malloc (sizeof(queue_t)); |
---|
| 234 | if (q == NULL) { |
---|
| 235 | return NULL; |
---|
| 236 | } |
---|
| 237 | |
---|
| 238 | queue_init(q); |
---|
| 239 | |
---|
| 240 | return q; |
---|
| 241 | } |
---|
| 242 | |
---|
| 243 | /** |
---|
| 244 | * Frees an initialized queue that was allocated on the heap. |
---|
| 245 | */ |
---|
| 246 | static void tq_free_queue(queue_t* q) |
---|
| 247 | { |
---|
| 248 | mem_free(q); |
---|
| 249 | } |
---|
| 250 | |
---|
| 251 | /** |
---|
| 252 | * Empties out a queue in the task queue by dequeuing and freeing |
---|
| 253 | * every task. |
---|
| 254 | */ |
---|
| 255 | static void tq_empty_queue(queue_t* q) |
---|
| 256 | { |
---|
| 257 | do { |
---|
| 258 | tq_entry_t *entry; |
---|
| 259 | queue_elem_t *queue_elem; |
---|
| 260 | |
---|
| 261 | if (queue_pop_front (q, &queue_elem) == 0) |
---|
| 262 | break; |
---|
| 263 | |
---|
| 264 | entry = queue_entry (queue_elem, tq_entry_t, queue_elem); |
---|
| 265 | assert (entry != NULL); |
---|
| 266 | mem_free (entry); |
---|
| 267 | } while (1); |
---|
| 268 | } |
---|
| 269 | |
---|
| 270 | static inline void tq_finalize_normal(taskQ_t* tq) |
---|
| 271 | { |
---|
| 272 | int i; |
---|
| 273 | |
---|
| 274 | assert (tq->queues != NULL); |
---|
| 275 | assert (tq->free_queues != NULL); |
---|
| 276 | assert (tq->locks != NULL); |
---|
| 277 | |
---|
| 278 | /* destroy all queues */ |
---|
| 279 | for (i = 0; i < tq->num_queues; ++i) { |
---|
| 280 | tq_queue_destroy(tq, i); |
---|
| 281 | } |
---|
| 282 | |
---|
| 283 | /* destroy all first level pointers in tq */ |
---|
| 284 | mem_free (tq->queues); |
---|
| 285 | mem_free (tq->free_queues); |
---|
| 286 | mem_free (tq->locks); |
---|
| 287 | mem_free (tq->seeds); |
---|
| 288 | |
---|
| 289 | /* finally kill tq */ |
---|
| 290 | mem_free (tq); |
---|
| 291 | } |
---|
| 292 | |
---|
| 293 | void tq_finalize (taskQ_t* tq) |
---|
| 294 | { |
---|
| 295 | tq_finalize_normal(tq); |
---|
| 296 | } |
---|
| 297 | |
---|
| 298 | /* Queue TASK at LGRP task queue with locking. |
---|
| 299 | LGRP is a locality hint denoting to which locality group this task |
---|
| 300 | should be queued at. If LGRP is less than 0, the locality group is |
---|
| 301 | randomly selected. TID is required for MCS locking. */ |
---|
| 302 | int tq_enqueue (taskQ_t* tq, task_t *task, int lgrp, int tid) |
---|
| 303 | { |
---|
| 304 | tq_entry_t *entry; |
---|
| 305 | int index; |
---|
| 306 | |
---|
| 307 | assert (tq != NULL); |
---|
| 308 | assert (task != NULL); |
---|
| 309 | |
---|
| 310 | entry = (tq_entry_t *)mem_malloc (sizeof (tq_entry_t)); |
---|
| 311 | if (entry == NULL) { |
---|
| 312 | return -1; |
---|
| 313 | } |
---|
| 314 | |
---|
| 315 | mem_memcpy (&entry->task, task, sizeof (task_t)); |
---|
| 316 | |
---|
| 317 | index = (lgrp < 0) ? rand_r(&tq->seeds[tid]) : lgrp; |
---|
| 318 | index %= tq->num_queues; |
---|
| 319 | |
---|
| 320 | lock_acquire (tq->locks[index].per_thread[tid]); |
---|
| 321 | queue_push_back (tq->queues[index], &entry->queue_elem); |
---|
| 322 | lock_release (tq->locks[index].per_thread[tid]); |
---|
| 323 | |
---|
| 324 | return 0; |
---|
| 325 | } |
---|
| 326 | |
---|
| 327 | /* Queue TASK at LGRP task queue without locking. |
---|
| 328 | LGRP is a locality hint denoting to which locality group this task |
---|
| 329 | should be queued at. If LGRP is less than 0, the locality group is |
---|
| 330 | randomly selected. */ |
---|
| 331 | int tq_enqueue_seq (taskQ_t* tq, task_t *task, int lgrp) |
---|
| 332 | { |
---|
| 333 | tq_entry_t *entry; |
---|
| 334 | int index; |
---|
| 335 | |
---|
| 336 | assert (task != NULL); |
---|
| 337 | |
---|
| 338 | entry = (tq_entry_t *)mem_malloc (sizeof (tq_entry_t)); |
---|
| 339 | if (entry == NULL) { |
---|
| 340 | return -1; |
---|
| 341 | } |
---|
| 342 | |
---|
| 343 | mem_memcpy (&entry->task, task, sizeof (task_t)); |
---|
| 344 | |
---|
| 345 | index = (lgrp < 0) ? rand() % tq->num_queues : lgrp % tq->num_queues; |
---|
| 346 | queue_push_back (tq->queues[index], &entry->queue_elem); |
---|
| 347 | |
---|
| 348 | return 0; |
---|
| 349 | } |
---|
| 350 | |
---|
| 351 | /** |
---|
| 352 | * Safely dequeues an element from the normal queue and queues onto free queue |
---|
| 353 | * @param tq taskQ to operate on |
---|
| 354 | * @param idx index of queue to use |
---|
| 355 | * @param tid task id |
---|
| 356 | * @param qe queue element of element we operated on |
---|
| 357 | * @return nonzero if normal queue was empty |
---|
| 358 | */ |
---|
| 359 | static inline int tq_elem_into_free_seq ( |
---|
| 360 | taskQ_t* tq, int idx, int tid, queue_elem_t** qe) |
---|
| 361 | { |
---|
| 362 | queue_elem_t *queue_elem = NULL; |
---|
| 363 | int ret; |
---|
| 364 | |
---|
| 365 | ret = queue_pop_front (tq->queues[idx], &queue_elem); |
---|
| 366 | if (ret != 0) |
---|
| 367 | queue_push_back (tq->free_queues[idx], queue_elem); |
---|
| 368 | |
---|
| 369 | *qe = queue_elem; |
---|
| 370 | |
---|
| 371 | return ret; |
---|
| 372 | } |
---|
| 373 | |
---|
| 374 | /** |
---|
| 375 | * Safely dequeues an element from the normal queue and queues onto free queue |
---|
| 376 | * @param tq taskQ to operate on |
---|
| 377 | * @param idx index of queue to use |
---|
| 378 | * @param tid task id |
---|
| 379 | * @param qe queue element of element we operated on |
---|
| 380 | * @return nonzero if normal queue was empty |
---|
| 381 | */ |
---|
| 382 | static inline int tq_elem_into_free ( |
---|
| 383 | taskQ_t* tq, int idx, int tid, queue_elem_t** qe) |
---|
| 384 | { |
---|
| 385 | int ret; |
---|
| 386 | |
---|
| 387 | lock_acquire (tq->locks[idx].per_thread[tid]); |
---|
| 388 | ret = tq_elem_into_free_seq (tq, idx, tid, qe); |
---|
| 389 | lock_release (tq->locks[idx].per_thread[tid]); |
---|
| 390 | |
---|
| 391 | return ret; |
---|
| 392 | } |
---|
| 393 | |
---|
| 394 | static inline int tq_dequeue_normal_seq ( |
---|
| 395 | taskQ_t* tq, task_t* task, int lgrp, int tid) |
---|
| 396 | { |
---|
| 397 | return tq_dequeue_normal_internal ( |
---|
| 398 | tq, task, lgrp, tid, tq_elem_into_free_seq); |
---|
| 399 | } |
---|
| 400 | |
---|
| 401 | static inline int tq_dequeue_normal( |
---|
| 402 | taskQ_t* tq, task_t* task, int lgrp, int tid) |
---|
| 403 | { |
---|
| 404 | return tq_dequeue_normal_internal ( |
---|
| 405 | tq, task, lgrp, tid, tq_elem_into_free); |
---|
| 406 | } |
---|
| 407 | |
---|
| 408 | static inline int tq_dequeue_normal_internal ( |
---|
| 409 | taskQ_t* tq, task_t* task, int lgrp, int tid, dequeue_fn dequeue_fn) |
---|
| 410 | { |
---|
| 411 | int i, ret, index; |
---|
| 412 | queue_elem_t *queue_elem; |
---|
| 413 | tq_entry_t *entry; |
---|
| 414 | |
---|
| 415 | assert (task != NULL); |
---|
| 416 | |
---|
| 417 | mem_memset (task, 0, sizeof (task_t)); |
---|
| 418 | |
---|
| 419 | index = (lgrp < 0) ? rand_r(&tq->seeds[tid]) : lgrp; |
---|
| 420 | index %= tq->num_queues; |
---|
| 421 | ret = (*dequeue_fn)(tq, index, tid, &queue_elem); |
---|
| 422 | |
---|
| 423 | /* Do task stealing if nothing on our queue. |
---|
| 424 | Cycle through all indexes until success or exhaustion */ |
---|
| 425 | for (i = (index + 1) % tq->num_queues; |
---|
| 426 | (ret == 0) && (i != index); |
---|
| 427 | i = (i + 1) % tq->num_queues) |
---|
| 428 | { |
---|
| 429 | ret = (*dequeue_fn)(tq, i, tid, &queue_elem); |
---|
| 430 | } |
---|
| 431 | |
---|
| 432 | if (ret == 0) { |
---|
| 433 | /* There really is no more work. */ |
---|
| 434 | return 0; |
---|
| 435 | } |
---|
| 436 | |
---|
| 437 | entry = queue_entry (queue_elem, tq_entry_t, queue_elem); |
---|
| 438 | assert (entry != NULL); |
---|
| 439 | |
---|
| 440 | mem_memcpy (task, &entry->task, sizeof (task_t)); |
---|
| 441 | |
---|
| 442 | return 1; |
---|
| 443 | } |
---|
| 444 | |
---|
| 445 | int tq_dequeue (taskQ_t* tq, task_t *task, int lgrp, int tid) |
---|
| 446 | { |
---|
| 447 | return tq_dequeue_normal(tq, task, lgrp, tid); |
---|
| 448 | } |
---|