1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
|
/*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software Foundation,
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*/
/** \file
* \ingroup bli
*
* Parallel tasks over all elements in a container.
*/
#include <stdlib.h>
#include "MEM_guardedalloc.h"
#include "DNA_listBase.h"
#include "BLI_listbase.h"
#include "BLI_math.h"
#include "BLI_mempool.h"
#include "BLI_mempool_private.h"
#include "BLI_task.h"
#include "BLI_threads.h"
#include "atomic_ops.h"
/* -------------------------------------------------------------------- */
/** \name Macros
* \{ */
/* Allows to avoid using malloc for userdata_chunk in tasks, when small enough. */
#define MALLOCA(_size) ((_size) <= 8192) ? alloca((_size)) : MEM_mallocN((_size), __func__)
#define MALLOCA_FREE(_mem, _size) \
if (((_mem) != NULL) && ((_size) > 8192)) { \
MEM_freeN((_mem)); \
} \
((void)0)
/** \} */
/* -------------------------------------------------------------------- */
/** \name Generic Iteration
* \{ */
BLI_INLINE void task_parallel_calc_chunk_size(const TaskParallelSettings *settings,
const int tot_items,
int num_tasks,
int *r_chunk_size)
{
int chunk_size = 0;
if (!settings->use_threading) {
/* Some users of this helper will still need a valid chunk size in case processing is not
* threaded. We can use a bigger one than in default threaded case then. */
chunk_size = 1024;
num_tasks = 1;
}
else if (settings->min_iter_per_thread > 0) {
/* Already set by user, no need to do anything here. */
chunk_size = settings->min_iter_per_thread;
}
else {
/* Multiplier used in heuristics below to define "optimal" chunk size.
* The idea here is to increase the chunk size to compensate for a rather measurable threading
* overhead caused by fetching tasks. With too many CPU threads we are starting
* to spend too much time in those overheads.
* First values are: 1 if num_tasks < 16;
* else 2 if num_tasks < 32;
* else 3 if num_tasks < 48;
* else 4 if num_tasks < 64;
* etc.
* Note: If we wanted to keep the 'power of two' multiplier, we'd need something like:
* 1 << max_ii(0, (int)(sizeof(int) * 8) - 1 - bitscan_reverse_i(num_tasks) - 3)
*/
const int num_tasks_factor = max_ii(1, num_tasks >> 3);
/* We could make that 'base' 32 number configurable in TaskParallelSettings too, or maybe just
* always use that heuristic using TaskParallelSettings.min_iter_per_thread as basis? */
chunk_size = 32 * num_tasks_factor;
/* Basic heuristic to avoid threading on low amount of items.
* We could make that limit configurable in settings too. */
if (tot_items > 0 && tot_items < max_ii(256, chunk_size * 2)) {
chunk_size = tot_items;
}
}
BLI_assert(chunk_size > 0);
*r_chunk_size = chunk_size;
}
typedef struct TaskParallelIteratorState {
void *userdata;
TaskParallelIteratorIterFunc iter_func;
TaskParallelIteratorFunc func;
/* *** Data used to 'acquire' chunks of items from the iterator. *** */
/* Common data also passed to the generator callback. */
TaskParallelIteratorStateShared iter_shared;
/* Total number of items. If unknown, set it to a negative number. */
int tot_items;
} TaskParallelIteratorState;
static void parallel_iterator_func_do(TaskParallelIteratorState *__restrict state,
void *userdata_chunk)
{
TaskParallelTLS tls = {
.userdata_chunk = userdata_chunk,
};
void **current_chunk_items;
int *current_chunk_indices;
int current_chunk_size;
const size_t items_size = sizeof(*current_chunk_items) * (size_t)state->iter_shared.chunk_size;
const size_t indices_size = sizeof(*current_chunk_indices) *
(size_t)state->iter_shared.chunk_size;
current_chunk_items = MALLOCA(items_size);
current_chunk_indices = MALLOCA(indices_size);
current_chunk_size = 0;
for (bool do_abort = false; !do_abort;) {
if (state->iter_shared.spin_lock != NULL) {
BLI_spin_lock(state->iter_shared.spin_lock);
}
/* Get current status. */
int index = state->iter_shared.next_index;
void *item = state->iter_shared.next_item;
int i;
/* 'Acquire' a chunk of items from the iterator function. */
for (i = 0; i < state->iter_shared.chunk_size && !state->iter_shared.is_finished; i++) {
current_chunk_indices[i] = index;
current_chunk_items[i] = item;
state->iter_func(state->userdata, &tls, &item, &index, &state->iter_shared.is_finished);
}
/* Update current status. */
state->iter_shared.next_index = index;
state->iter_shared.next_item = item;
current_chunk_size = i;
do_abort = state->iter_shared.is_finished;
if (state->iter_shared.spin_lock != NULL) {
BLI_spin_unlock(state->iter_shared.spin_lock);
}
for (i = 0; i < current_chunk_size; ++i) {
state->func(state->userdata, current_chunk_items[i], current_chunk_indices[i], &tls);
}
}
MALLOCA_FREE(current_chunk_items, items_size);
MALLOCA_FREE(current_chunk_indices, indices_size);
}
static void parallel_iterator_func(TaskPool *__restrict pool, void *userdata_chunk)
{
TaskParallelIteratorState *__restrict state = BLI_task_pool_user_data(pool);
parallel_iterator_func_do(state, userdata_chunk);
}
static void task_parallel_iterator_no_threads(const TaskParallelSettings *settings,
TaskParallelIteratorState *state)
{
/* Prepare user's TLS data. */
void *userdata_chunk = settings->userdata_chunk;
const size_t userdata_chunk_size = settings->userdata_chunk_size;
void *userdata_chunk_local = NULL;
const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);
if (use_userdata_chunk) {
userdata_chunk_local = MALLOCA(userdata_chunk_size);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
}
/* Also marking it as non-threaded for the iterator callback. */
state->iter_shared.spin_lock = NULL;
parallel_iterator_func_do(state, userdata_chunk);
if (use_userdata_chunk) {
if (settings->func_free != NULL) {
/* `func_free` should only free data that was created during execution of `func`. */
settings->func_free(state->userdata, userdata_chunk_local);
}
MALLOCA_FREE(userdata_chunk_local, userdata_chunk_size);
}
}
static void task_parallel_iterator_do(const TaskParallelSettings *settings,
TaskParallelIteratorState *state)
{
const int num_threads = BLI_task_scheduler_num_threads();
task_parallel_calc_chunk_size(
settings, state->tot_items, num_threads, &state->iter_shared.chunk_size);
if (!settings->use_threading) {
task_parallel_iterator_no_threads(settings, state);
return;
}
const int chunk_size = state->iter_shared.chunk_size;
const int tot_items = state->tot_items;
const size_t num_tasks = tot_items >= 0 ?
(size_t)min_ii(num_threads, state->tot_items / chunk_size) :
(size_t)num_threads;
BLI_assert(num_tasks > 0);
if (num_tasks == 1) {
task_parallel_iterator_no_threads(settings, state);
return;
}
SpinLock spin_lock;
BLI_spin_init(&spin_lock);
state->iter_shared.spin_lock = &spin_lock;
void *userdata_chunk = settings->userdata_chunk;
const size_t userdata_chunk_size = settings->userdata_chunk_size;
void *userdata_chunk_local = NULL;
void *userdata_chunk_array = NULL;
const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);
TaskPool *task_pool = BLI_task_pool_create(state, TASK_PRIORITY_HIGH, TASK_ISOLATION_ON);
if (use_userdata_chunk) {
userdata_chunk_array = MALLOCA(userdata_chunk_size * num_tasks);
}
for (size_t i = 0; i < num_tasks; i++) {
if (use_userdata_chunk) {
userdata_chunk_local = (char *)userdata_chunk_array + (userdata_chunk_size * i);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
}
/* Use this pool's pre-allocated tasks. */
BLI_task_pool_push(task_pool, parallel_iterator_func, userdata_chunk_local, false, NULL);
}
BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);
if (use_userdata_chunk) {
if (settings->func_reduce != NULL || settings->func_free != NULL) {
for (size_t i = 0; i < num_tasks; i++) {
userdata_chunk_local = (char *)userdata_chunk_array + (userdata_chunk_size * i);
if (settings->func_reduce != NULL) {
settings->func_reduce(state->userdata, userdata_chunk, userdata_chunk_local);
}
if (settings->func_free != NULL) {
settings->func_free(state->userdata, userdata_chunk_local);
}
}
}
MALLOCA_FREE(userdata_chunk_array, userdata_chunk_size * num_tasks);
}
BLI_spin_end(&spin_lock);
state->iter_shared.spin_lock = NULL;
}
/**
* This function allows to parallelize for loops using a generic iterator.
*
* \param userdata: Common userdata passed to all instances of \a func.
* \param iter_func: Callback function used to generate chunks of items.
* \param init_item: The initial item, if necessary (may be NULL if unused).
* \param init_index: The initial index.
* \param tot_items: The total amount of items to iterate over
* (if unknown, set it to a negative number).
* \param func: Callback function.
* \param settings: See public API doc of TaskParallelSettings for description of all settings.
*
* \note Static scheduling is only available when \a tot_items is >= 0.
*/
void BLI_task_parallel_iterator(void *userdata,
TaskParallelIteratorIterFunc iter_func,
void *init_item,
const int init_index,
const int tot_items,
TaskParallelIteratorFunc func,
const TaskParallelSettings *settings)
{
TaskParallelIteratorState state = {0};
state.tot_items = tot_items;
state.iter_shared.next_index = init_index;
state.iter_shared.next_item = init_item;
state.iter_shared.is_finished = false;
state.userdata = userdata;
state.iter_func = iter_func;
state.func = func;
task_parallel_iterator_do(settings, &state);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name ListBase Iteration
* \{ */
static void task_parallel_listbase_get(void *__restrict UNUSED(userdata),
const TaskParallelTLS *__restrict UNUSED(tls),
void **r_next_item,
int *r_next_index,
bool *r_do_abort)
{
/* Get current status. */
Link *link = *r_next_item;
if (link->next == NULL) {
*r_do_abort = true;
}
*r_next_item = link->next;
(*r_next_index)++;
}
/**
* This function allows to parallelize for loops over ListBase items.
*
* \param listbase: The double linked list to loop over.
* \param userdata: Common userdata passed to all instances of \a func.
* \param func: Callback function.
* \param settings: See public API doc of ParallelRangeSettings for description of all settings.
*
* \note There is no static scheduling here,
* since it would need another full loop over items to count them.
*/
void BLI_task_parallel_listbase(ListBase *listbase,
void *userdata,
TaskParallelIteratorFunc func,
const TaskParallelSettings *settings)
{
if (BLI_listbase_is_empty(listbase)) {
return;
}
TaskParallelIteratorState state = {0};
state.tot_items = BLI_listbase_count(listbase);
state.iter_shared.next_index = 0;
state.iter_shared.next_item = listbase->first;
state.iter_shared.is_finished = false;
state.userdata = userdata;
state.iter_func = task_parallel_listbase_get;
state.func = func;
task_parallel_iterator_do(settings, &state);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name MemPool Iteration
* \{ */
typedef struct ParallelMempoolState {
void *userdata;
TaskParallelMempoolFunc func;
} ParallelMempoolState;
static void parallel_mempool_func(TaskPool *__restrict pool, void *taskdata)
{
ParallelMempoolState *__restrict state = BLI_task_pool_user_data(pool);
BLI_mempool_threadsafe_iter *iter = &((ParallelMempoolTaskData *)taskdata)->ts_iter;
TaskParallelTLS *tls = &((ParallelMempoolTaskData *)taskdata)->tls;
MempoolIterData *item;
while ((item = mempool_iter_threadsafe_step(iter)) != NULL) {
state->func(state->userdata, item, tls);
}
}
/**
* This function allows to parallelize for loops over Mempool items.
*
* \param mempool: The iterable BLI_mempool to loop over.
* \param userdata: Common userdata passed to all instances of \a func.
* \param func: Callback function.
* \param settings: See public API doc of TaskParallelSettings for description of all settings.
*
* \note There is no static scheduling here.
*/
void BLI_task_parallel_mempool(BLI_mempool *mempool,
void *userdata,
TaskParallelMempoolFunc func,
const TaskParallelSettings *settings)
{
TaskPool *task_pool;
ParallelMempoolState state;
int i, num_threads, num_tasks;
if (BLI_mempool_len(mempool) == 0) {
return;
}
void *userdata_chunk = settings->userdata_chunk;
const size_t userdata_chunk_size = settings->userdata_chunk_size;
void *userdata_chunk_local = NULL;
void *userdata_chunk_array = NULL;
const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);
if (!settings->use_threading) {
TaskParallelTLS tls = {NULL};
if (use_userdata_chunk) {
userdata_chunk_local = MALLOCA(userdata_chunk_size);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
tls.userdata_chunk = userdata_chunk_local;
}
BLI_mempool_iter iter;
BLI_mempool_iternew(mempool, &iter);
void *item;
while ((item = BLI_mempool_iterstep(&iter))) {
func(userdata, item, &tls);
}
if (settings->func_free != NULL) {
/* `func_free` should only free data that was created during execution of `func`. */
settings->func_free(userdata, userdata_chunk_local);
}
MALLOCA_FREE(userdata_chunk_local, userdata_chunk_size);
return;
}
task_pool = BLI_task_pool_create(&state, TASK_PRIORITY_HIGH, TASK_ISOLATION_ON);
num_threads = BLI_task_scheduler_num_threads();
/* The idea here is to prevent creating task for each of the loop iterations
* and instead have tasks which are evenly distributed across CPU cores and
* pull next item to be crunched using the threaded-aware BLI_mempool_iter.
*/
num_tasks = num_threads + 2;
state.userdata = userdata;
state.func = func;
if (use_userdata_chunk) {
userdata_chunk_array = MALLOCA(userdata_chunk_size * num_tasks);
}
ParallelMempoolTaskData *mempool_iterator_data = mempool_iter_threadsafe_create(
mempool, (size_t)num_tasks);
for (i = 0; i < num_tasks; i++) {
if (use_userdata_chunk) {
userdata_chunk_local = (char *)userdata_chunk_array + (userdata_chunk_size * i);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
}
mempool_iterator_data[i].tls.userdata_chunk = userdata_chunk_local;
/* Use this pool's pre-allocated tasks. */
BLI_task_pool_push(task_pool, parallel_mempool_func, &mempool_iterator_data[i], false, NULL);
}
BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);
if (use_userdata_chunk) {
if ((settings->func_free != NULL) || (settings->func_reduce != NULL)) {
for (i = 0; i < num_tasks; i++) {
if (settings->func_reduce) {
settings->func_reduce(
userdata, userdata_chunk, mempool_iterator_data[i].tls.userdata_chunk);
}
if (settings->func_free) {
settings->func_free(userdata, mempool_iterator_data[i].tls.userdata_chunk);
}
}
}
MALLOCA_FREE(userdata_chunk_array, userdata_chunk_size * num_tasks);
}
mempool_iter_threadsafe_destroy(mempool_iterator_data);
}
#undef MALLOCA
#undef MALLOCA_FREE
/** \} */
|