1 /*
2 * Copyright © 2016 Advanced Micro Devices, Inc.
3 * All Rights Reserved.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining
6 * a copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
14 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
15 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
16 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
17 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
20 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
24 * of the Software.
25 */
26
27 #include "u_queue.h"
28
29 #include "c11/threads.h"
30 #include "util/u_cpu_detect.h"
31 #include "util/os_time.h"
32 #include "util/u_string.h"
33 #include "util/u_thread.h"
34 #include "u_process.h"
35
36 #if defined(__linux__)
37 #include <sys/time.h>
38 #include <sys/resource.h>
39 #include <sys/syscall.h>
40 #endif
41
42
43 /* Define 256MB */
44 #define S_256MB (256 * 1024 * 1024)
45
46 static void
47 util_queue_kill_threads(struct util_queue *queue, unsigned keep_num_threads,
48 bool locked);
49
50 /****************************************************************************
51 * Wait for all queues to assert idle when exit() is called.
52 *
53 * Otherwise, C++ static variable destructors can be called while threads
54 * are using the static variables.
55 */
56
57 static once_flag atexit_once_flag = ONCE_FLAG_INIT;
58 static struct list_head queue_list = {
59 .next = &queue_list,
60 .prev = &queue_list,
61 };
62 static mtx_t exit_mutex;
63
64 static void
atexit_handler(void)65 atexit_handler(void)
66 {
67 struct util_queue *iter;
68
69 mtx_lock(&exit_mutex);
70 /* Wait for all queues to assert idle. */
71 LIST_FOR_EACH_ENTRY(iter, &queue_list, head) {
72 util_queue_kill_threads(iter, 0, false);
73 }
74 mtx_unlock(&exit_mutex);
75 }
76
77 static void
global_init(void)78 global_init(void)
79 {
80 mtx_init(&exit_mutex, mtx_plain);
81 atexit(atexit_handler);
82 }
83
84 static void
add_to_atexit_list(struct util_queue * queue)85 add_to_atexit_list(struct util_queue *queue)
86 {
87 call_once(&atexit_once_flag, global_init);
88
89 mtx_lock(&exit_mutex);
90 list_add(&queue->head, &queue_list);
91 mtx_unlock(&exit_mutex);
92 }
93
94 static void
remove_from_atexit_list(struct util_queue * queue)95 remove_from_atexit_list(struct util_queue *queue)
96 {
97 struct util_queue *iter, *tmp;
98
99 mtx_lock(&exit_mutex);
100 LIST_FOR_EACH_ENTRY_SAFE(iter, tmp, &queue_list, head) {
101 if (iter == queue) {
102 list_del(&iter->head);
103 break;
104 }
105 }
106 mtx_unlock(&exit_mutex);
107 }
108
109 /****************************************************************************
110 * util_queue_fence
111 */
112
113 #ifdef UTIL_QUEUE_FENCE_FUTEX
114 static bool
do_futex_fence_wait(struct util_queue_fence * fence,bool timeout,int64_t abs_timeout)115 do_futex_fence_wait(struct util_queue_fence *fence,
116 bool timeout, int64_t abs_timeout)
117 {
118 uint32_t v = p_atomic_read_relaxed(&fence->val);
119 struct timespec ts;
120 ts.tv_sec = abs_timeout / (1000*1000*1000);
121 ts.tv_nsec = abs_timeout % (1000*1000*1000);
122
123 while (v != 0) {
124 if (v != 2) {
125 v = p_atomic_cmpxchg(&fence->val, 1, 2);
126 if (v == 0)
127 return true;
128 }
129
130 int r = futex_wait(&fence->val, 2, timeout ? &ts : NULL);
131 if (timeout && r < 0) {
132 if (errno == ETIMEDOUT)
133 return false;
134 }
135
136 v = p_atomic_read_relaxed(&fence->val);
137 }
138
139 return true;
140 }
141
142 void
_util_queue_fence_wait(struct util_queue_fence * fence)143 _util_queue_fence_wait(struct util_queue_fence *fence)
144 {
145 do_futex_fence_wait(fence, false, 0);
146 }
147
148 bool
_util_queue_fence_wait_timeout(struct util_queue_fence * fence,int64_t abs_timeout)149 _util_queue_fence_wait_timeout(struct util_queue_fence *fence,
150 int64_t abs_timeout)
151 {
152 return do_futex_fence_wait(fence, true, abs_timeout);
153 }
154
155 #endif
156
157 #ifdef UTIL_QUEUE_FENCE_STANDARD
158 void
util_queue_fence_signal(struct util_queue_fence * fence)159 util_queue_fence_signal(struct util_queue_fence *fence)
160 {
161 mtx_lock(&fence->mutex);
162 fence->signalled = true;
163 cnd_broadcast(&fence->cond);
164 mtx_unlock(&fence->mutex);
165 }
166
167 void
_util_queue_fence_wait(struct util_queue_fence * fence)168 _util_queue_fence_wait(struct util_queue_fence *fence)
169 {
170 mtx_lock(&fence->mutex);
171 while (!fence->signalled)
172 cnd_wait(&fence->cond, &fence->mutex);
173 mtx_unlock(&fence->mutex);
174 }
175
176 bool
_util_queue_fence_wait_timeout(struct util_queue_fence * fence,int64_t abs_timeout)177 _util_queue_fence_wait_timeout(struct util_queue_fence *fence,
178 int64_t abs_timeout)
179 {
180 /* This terrible hack is made necessary by the fact that we really want an
181 * internal interface consistent with os_time_*, but cnd_timedwait is spec'd
182 * to be relative to the TIME_UTC clock.
183 */
184 int64_t rel = abs_timeout - os_time_get_nano();
185
186 if (rel > 0) {
187 struct timespec ts;
188
189 #ifdef HAVE_TIMESPEC_GET
190 timespec_get(&ts, TIME_UTC);
191 #else
192 clock_gettime(CLOCK_REALTIME, &ts);
193 #endif
194
195 ts.tv_sec += abs_timeout / (1000*1000*1000);
196 ts.tv_nsec += abs_timeout % (1000*1000*1000);
197 if (ts.tv_nsec >= (1000*1000*1000)) {
198 ts.tv_sec++;
199 ts.tv_nsec -= (1000*1000*1000);
200 }
201
202 mtx_lock(&fence->mutex);
203 while (!fence->signalled) {
204 if (cnd_timedwait(&fence->cond, &fence->mutex, &ts) != thrd_success)
205 break;
206 }
207 mtx_unlock(&fence->mutex);
208 }
209
210 return fence->signalled;
211 }
212
213 void
util_queue_fence_init(struct util_queue_fence * fence)214 util_queue_fence_init(struct util_queue_fence *fence)
215 {
216 memset(fence, 0, sizeof(*fence));
217 (void) mtx_init(&fence->mutex, mtx_plain);
218 cnd_init(&fence->cond);
219 fence->signalled = true;
220 }
221
222 void
util_queue_fence_destroy(struct util_queue_fence * fence)223 util_queue_fence_destroy(struct util_queue_fence *fence)
224 {
225 assert(fence->signalled);
226
227 /* Ensure that another thread is not in the middle of
228 * util_queue_fence_signal (having set the fence to signalled but still
229 * holding the fence mutex).
230 *
231 * A common contract between threads is that as soon as a fence is signalled
232 * by thread A, thread B is allowed to destroy it. Since
233 * util_queue_fence_is_signalled does not lock the fence mutex (for
234 * performance reasons), we must do so here.
235 */
236 mtx_lock(&fence->mutex);
237 mtx_unlock(&fence->mutex);
238
239 cnd_destroy(&fence->cond);
240 mtx_destroy(&fence->mutex);
241 }
242 #endif
243
244 /****************************************************************************
245 * util_queue implementation
246 */
247
248 struct thread_input {
249 struct util_queue *queue;
250 int thread_index;
251 };
252
253 static int
util_queue_thread_func(void * input)254 util_queue_thread_func(void *input)
255 {
256 struct util_queue *queue = ((struct thread_input*)input)->queue;
257 int thread_index = ((struct thread_input*)input)->thread_index;
258
259 free(input);
260
261 if (queue->flags & UTIL_QUEUE_INIT_SET_FULL_THREAD_AFFINITY) {
262 /* Don't inherit the thread affinity from the parent thread.
263 * Set the full mask.
264 */
265 uint32_t mask[UTIL_MAX_CPUS / 32];
266
267 memset(mask, 0xff, sizeof(mask));
268
269 util_set_current_thread_affinity(mask, NULL,
270 util_get_cpu_caps()->num_cpu_mask_bits);
271 }
272
273 #if defined(__linux__)
274 if (queue->flags & UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY) {
275 /* The nice() function can only set a maximum of 19. */
276 setpriority(PRIO_PROCESS, syscall(SYS_gettid), 19);
277 }
278 #endif
279
280 if (strlen(queue->name) > 0) {
281 char name[16];
282 snprintf(name, sizeof(name), "%s%i", queue->name, thread_index);
283 u_thread_setname(name);
284 }
285
286 while (1) {
287 struct util_queue_job job;
288
289 mtx_lock(&queue->lock);
290 assert(queue->num_queued >= 0 && queue->num_queued <= queue->max_jobs);
291
292 /* wait if the queue is empty */
293 while (thread_index < queue->num_threads && queue->num_queued == 0)
294 cnd_wait(&queue->has_queued_cond, &queue->lock);
295
296 /* only kill threads that are above "num_threads" */
297 if (thread_index >= queue->num_threads) {
298 mtx_unlock(&queue->lock);
299 break;
300 }
301
302 job = queue->jobs[queue->read_idx];
303 memset(&queue->jobs[queue->read_idx], 0, sizeof(struct util_queue_job));
304 queue->read_idx = (queue->read_idx + 1) % queue->max_jobs;
305
306 queue->num_queued--;
307 cnd_signal(&queue->has_space_cond);
308 if (job.job)
309 queue->total_jobs_size -= job.job_size;
310 mtx_unlock(&queue->lock);
311
312 if (job.job) {
313 job.execute(job.job, job.global_data, thread_index);
314 if (job.fence)
315 util_queue_fence_signal(job.fence);
316 if (job.cleanup)
317 job.cleanup(job.job, job.global_data, thread_index);
318 }
319 }
320
321 /* signal remaining jobs if all threads are being terminated */
322 mtx_lock(&queue->lock);
323 if (queue->num_threads == 0) {
324 for (unsigned i = queue->read_idx; i != queue->write_idx;
325 i = (i + 1) % queue->max_jobs) {
326 if (queue->jobs[i].job) {
327 if (queue->jobs[i].fence)
328 util_queue_fence_signal(queue->jobs[i].fence);
329 queue->jobs[i].job = NULL;
330 }
331 }
332 queue->read_idx = queue->write_idx;
333 queue->num_queued = 0;
334 }
335 mtx_unlock(&queue->lock);
336 return 0;
337 }
338
339 static bool
util_queue_create_thread(struct util_queue * queue,unsigned index)340 util_queue_create_thread(struct util_queue *queue, unsigned index)
341 {
342 struct thread_input *input =
343 (struct thread_input *) malloc(sizeof(struct thread_input));
344 input->queue = queue;
345 input->thread_index = index;
346
347 if (thrd_success != u_thread_create(queue->threads + index, util_queue_thread_func, input)) {
348 free(input);
349 return false;
350 }
351
352 if (queue->flags & UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY) {
353 #if defined(__linux__) && defined(SCHED_BATCH)
354 struct sched_param sched_param = {0};
355
356 /* The nice() function can only set a maximum of 19.
357 * SCHED_BATCH gives the scheduler a hint that this is a latency
358 * insensitive thread.
359 *
360 * Note that Linux only allows decreasing the priority. The original
361 * priority can't be restored.
362 */
363 pthread_setschedparam(queue->threads[index], SCHED_BATCH, &sched_param);
364 #endif
365 }
366 return true;
367 }
368
369 void
util_queue_adjust_num_threads(struct util_queue * queue,unsigned num_threads,bool locked)370 util_queue_adjust_num_threads(struct util_queue *queue, unsigned num_threads,
371 bool locked)
372 {
373 num_threads = MIN2(num_threads, queue->max_threads);
374 num_threads = MAX2(num_threads, 1);
375
376 if (!locked)
377 mtx_lock(&queue->lock);
378
379 unsigned old_num_threads = queue->num_threads;
380
381 if (num_threads == old_num_threads) {
382 if (!locked)
383 mtx_unlock(&queue->lock);
384 return;
385 }
386
387 if (num_threads < old_num_threads) {
388 util_queue_kill_threads(queue, num_threads, true);
389 if (!locked)
390 mtx_unlock(&queue->lock);
391 return;
392 }
393
394 /* Create threads.
395 *
396 * We need to update num_threads first, because threads terminate
397 * when thread_index < num_threads.
398 */
399 queue->num_threads = num_threads;
400 for (unsigned i = old_num_threads; i < num_threads; i++) {
401 if (!util_queue_create_thread(queue, i)) {
402 queue->num_threads = i;
403 break;
404 }
405 }
406
407 if (!locked)
408 mtx_unlock(&queue->lock);
409 }
410
411 bool
util_queue_init(struct util_queue * queue,const char * name,unsigned max_jobs,unsigned num_threads,unsigned flags,void * global_data)412 util_queue_init(struct util_queue *queue,
413 const char *name,
414 unsigned max_jobs,
415 unsigned num_threads,
416 unsigned flags,
417 void *global_data)
418 {
419 unsigned i;
420
421 /* Form the thread name from process_name and name, limited to 13
422 * characters. Characters 14-15 are reserved for the thread number.
423 * Character 16 should be 0. Final form: "process:name12"
424 *
425 * If name is too long, it's truncated. If any space is left, the process
426 * name fills it.
427 */
428 const char *process_name = util_get_process_name();
429 int process_len = process_name ? strlen(process_name) : 0;
430 int name_len = strlen(name);
431 const int max_chars = sizeof(queue->name) - 1;
432
433 name_len = MIN2(name_len, max_chars);
434
435 /* See if there is any space left for the process name, reserve 1 for
436 * the colon. */
437 process_len = MIN2(process_len, max_chars - name_len - 1);
438 process_len = MAX2(process_len, 0);
439
440 memset(queue, 0, sizeof(*queue));
441
442 if (process_len) {
443 snprintf(queue->name, sizeof(queue->name), "%.*s:%s",
444 process_len, process_name, name);
445 } else {
446 snprintf(queue->name, sizeof(queue->name), "%s", name);
447 }
448
449 queue->create_threads_on_demand = true;
450 queue->flags = flags;
451 queue->max_threads = num_threads;
452 queue->num_threads = 1;
453 queue->max_jobs = max_jobs;
454 queue->global_data = global_data;
455
456 (void) mtx_init(&queue->lock, mtx_plain);
457
458 queue->num_queued = 0;
459 cnd_init(&queue->has_queued_cond);
460 cnd_init(&queue->has_space_cond);
461
462 queue->jobs = (struct util_queue_job*)
463 calloc(max_jobs, sizeof(struct util_queue_job));
464 if (!queue->jobs)
465 goto fail;
466
467 queue->threads = (thrd_t*) calloc(queue->max_threads, sizeof(thrd_t));
468 if (!queue->threads)
469 goto fail;
470
471 /* start threads */
472 for (i = 0; i < queue->num_threads; i++) {
473 if (!util_queue_create_thread(queue, i)) {
474 if (i == 0) {
475 /* no threads created, fail */
476 goto fail;
477 } else {
478 /* at least one thread created, so use it */
479 queue->num_threads = i;
480 break;
481 }
482 }
483 }
484
485 add_to_atexit_list(queue);
486 return true;
487
488 fail:
489 free(queue->threads);
490
491 if (queue->jobs) {
492 cnd_destroy(&queue->has_space_cond);
493 cnd_destroy(&queue->has_queued_cond);
494 mtx_destroy(&queue->lock);
495 free(queue->jobs);
496 }
497 /* also util_queue_is_initialized can be used to check for success */
498 memset(queue, 0, sizeof(*queue));
499 return false;
500 }
501
502 static void
util_queue_kill_threads(struct util_queue * queue,unsigned keep_num_threads,bool locked)503 util_queue_kill_threads(struct util_queue *queue, unsigned keep_num_threads,
504 bool locked)
505 {
506 /* Signal all threads to terminate. */
507 if (!locked)
508 mtx_lock(&queue->lock);
509
510 if (keep_num_threads >= queue->num_threads) {
511 if (!locked)
512 mtx_unlock(&queue->lock);
513 return;
514 }
515
516 unsigned old_num_threads = queue->num_threads;
517 /* Setting num_threads is what causes the threads to terminate.
518 * Then cnd_broadcast wakes them up and they will exit their function.
519 */
520 queue->num_threads = keep_num_threads;
521 cnd_broadcast(&queue->has_queued_cond);
522
523 /* Wait for threads to terminate. */
524 if (keep_num_threads < old_num_threads) {
525 /* We need to unlock the mutex to allow threads to terminate. */
526 mtx_unlock(&queue->lock);
527 for (unsigned i = keep_num_threads; i < old_num_threads; i++)
528 thrd_join(queue->threads[i], NULL);
529 if (locked)
530 mtx_lock(&queue->lock);
531 } else {
532 if (!locked)
533 mtx_unlock(&queue->lock);
534 }
535 }
536
537 static void
util_queue_finish_execute(void * data,void * gdata,int num_thread)538 util_queue_finish_execute(void *data, void *gdata, int num_thread)
539 {
540 util_barrier *barrier = data;
541 if (util_barrier_wait(barrier))
542 util_barrier_destroy(barrier);
543 }
544
545 void
util_queue_destroy(struct util_queue * queue)546 util_queue_destroy(struct util_queue *queue)
547 {
548 util_queue_kill_threads(queue, 0, false);
549
550 /* This makes it safe to call on a queue that failed util_queue_init. */
551 if (queue->head.next != NULL)
552 remove_from_atexit_list(queue);
553
554 cnd_destroy(&queue->has_space_cond);
555 cnd_destroy(&queue->has_queued_cond);
556 mtx_destroy(&queue->lock);
557 free(queue->jobs);
558 free(queue->threads);
559 }
560
561 static void
util_queue_add_job_locked(struct util_queue * queue,void * job,struct util_queue_fence * fence,util_queue_execute_func execute,util_queue_execute_func cleanup,const size_t job_size,bool locked)562 util_queue_add_job_locked(struct util_queue *queue,
563 void *job,
564 struct util_queue_fence *fence,
565 util_queue_execute_func execute,
566 util_queue_execute_func cleanup,
567 const size_t job_size,
568 bool locked)
569 {
570 struct util_queue_job *ptr;
571
572 if (!locked)
573 mtx_lock(&queue->lock);
574 if (queue->num_threads == 0) {
575 if (!locked)
576 mtx_unlock(&queue->lock);
577 /* well no good option here, but any leaks will be
578 * short-lived as things are shutting down..
579 */
580 return;
581 }
582
583 if (fence)
584 util_queue_fence_reset(fence);
585
586 assert(queue->num_queued >= 0 && queue->num_queued <= queue->max_jobs);
587
588 /* Scale the number of threads up if there's already one job waiting. */
589 if (queue->num_queued > 0 &&
590 queue->create_threads_on_demand &&
591 execute != util_queue_finish_execute &&
592 queue->num_threads < queue->max_threads) {
593 util_queue_adjust_num_threads(queue, queue->num_threads + 1, true);
594 }
595
596 if (queue->num_queued == queue->max_jobs) {
597 if (queue->flags & UTIL_QUEUE_INIT_RESIZE_IF_FULL &&
598 queue->total_jobs_size + job_size < S_256MB) {
599 /* If the queue is full, make it larger to avoid waiting for a free
600 * slot.
601 */
602 unsigned new_max_jobs = queue->max_jobs + 8;
603 struct util_queue_job *jobs =
604 (struct util_queue_job*)calloc(new_max_jobs,
605 sizeof(struct util_queue_job));
606 assert(jobs);
607
608 /* Copy all queued jobs into the new list. */
609 unsigned num_jobs = 0;
610 unsigned i = queue->read_idx;
611
612 do {
613 jobs[num_jobs++] = queue->jobs[i];
614 i = (i + 1) % queue->max_jobs;
615 } while (i != queue->write_idx);
616
617 assert(num_jobs == queue->num_queued);
618
619 free(queue->jobs);
620 queue->jobs = jobs;
621 queue->read_idx = 0;
622 queue->write_idx = num_jobs;
623 queue->max_jobs = new_max_jobs;
624 } else {
625 /* Wait until there is a free slot. */
626 while (queue->num_queued == queue->max_jobs)
627 cnd_wait(&queue->has_space_cond, &queue->lock);
628 }
629 }
630
631 ptr = &queue->jobs[queue->write_idx];
632 assert(ptr->job == NULL);
633 ptr->job = job;
634 ptr->global_data = queue->global_data;
635 ptr->fence = fence;
636 ptr->execute = execute;
637 ptr->cleanup = cleanup;
638 ptr->job_size = job_size;
639
640 queue->write_idx = (queue->write_idx + 1) % queue->max_jobs;
641 queue->total_jobs_size += ptr->job_size;
642
643 queue->num_queued++;
644 cnd_signal(&queue->has_queued_cond);
645 if (!locked)
646 mtx_unlock(&queue->lock);
647 }
648
649 void
util_queue_add_job(struct util_queue * queue,void * job,struct util_queue_fence * fence,util_queue_execute_func execute,util_queue_execute_func cleanup,const size_t job_size)650 util_queue_add_job(struct util_queue *queue,
651 void *job,
652 struct util_queue_fence *fence,
653 util_queue_execute_func execute,
654 util_queue_execute_func cleanup,
655 const size_t job_size)
656 {
657 util_queue_add_job_locked(queue, job, fence, execute, cleanup, job_size,
658 false);
659 }
660
661 /**
662 * Remove a queued job. If the job hasn't started execution, it's removed from
663 * the queue. If the job has started execution, the function waits for it to
664 * complete.
665 *
666 * In all cases, the fence is signalled when the function returns.
667 *
668 * The function can be used when destroying an object associated with the job
669 * when you don't care about the job completion state.
670 */
671 void
util_queue_drop_job(struct util_queue * queue,struct util_queue_fence * fence)672 util_queue_drop_job(struct util_queue *queue, struct util_queue_fence *fence)
673 {
674 bool removed = false;
675
676 if (util_queue_fence_is_signalled(fence))
677 return;
678
679 mtx_lock(&queue->lock);
680 for (unsigned i = queue->read_idx; i != queue->write_idx;
681 i = (i + 1) % queue->max_jobs) {
682 if (queue->jobs[i].fence == fence) {
683 if (queue->jobs[i].cleanup)
684 queue->jobs[i].cleanup(queue->jobs[i].job, queue->global_data, -1);
685
686 /* Just clear it. The threads will treat as a no-op job. */
687 memset(&queue->jobs[i], 0, sizeof(queue->jobs[i]));
688 removed = true;
689 break;
690 }
691 }
692 mtx_unlock(&queue->lock);
693
694 if (removed)
695 util_queue_fence_signal(fence);
696 else
697 util_queue_fence_wait(fence);
698 }
699
700 /**
701 * Wait until all previously added jobs have completed.
702 */
703 void
util_queue_finish(struct util_queue * queue)704 util_queue_finish(struct util_queue *queue)
705 {
706 util_barrier barrier;
707 struct util_queue_fence *fences;
708
709 /* If 2 threads were adding jobs for 2 different barries at the same time,
710 * a deadlock would happen, because 1 barrier requires that all threads
711 * wait for it exclusively.
712 */
713 mtx_lock(&queue->lock);
714
715 /* The number of threads can be changed to 0, e.g. by the atexit handler. */
716 if (!queue->num_threads) {
717 mtx_unlock(&queue->lock);
718 return;
719 }
720
721 /* We need to disable adding new threads in util_queue_add_job because
722 * the finish operation requires a fixed number of threads.
723 *
724 * Also note that util_queue_add_job can unlock the mutex if there is not
725 * enough space in the queue and wait for space.
726 */
727 queue->create_threads_on_demand = false;
728
729 fences = malloc(queue->num_threads * sizeof(*fences));
730 util_barrier_init(&barrier, queue->num_threads);
731
732 for (unsigned i = 0; i < queue->num_threads; ++i) {
733 util_queue_fence_init(&fences[i]);
734 util_queue_add_job_locked(queue, &barrier, &fences[i],
735 util_queue_finish_execute, NULL, 0, true);
736 }
737 queue->create_threads_on_demand = true;
738 mtx_unlock(&queue->lock);
739
740 for (unsigned i = 0; i < queue->num_threads; ++i) {
741 util_queue_fence_wait(&fences[i]);
742 util_queue_fence_destroy(&fences[i]);
743 }
744
745 free(fences);
746 }
747
748 int64_t
util_queue_get_thread_time_nano(struct util_queue * queue,unsigned thread_index)749 util_queue_get_thread_time_nano(struct util_queue *queue, unsigned thread_index)
750 {
751 /* Allow some flexibility by not raising an error. */
752 if (thread_index >= queue->num_threads)
753 return 0;
754
755 return util_thread_get_time_nano(queue->threads[thread_index]);
756 }
757