-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathteam.c
1125 lines (1024 loc) · 31.3 KB
/
team.c
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
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* Copyright (C) 2005-2024 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU Offloading and Multi Processing Library
(libgomp).
Libgomp 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 3, or (at your option)
any later version.
Libgomp 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.
Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.
You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
<http://www.gnu.org/licenses/>. */
/* This file handles the maintenance of threads in response to team
creation and termination. */
#include "libgomp.h"
#include "pool.h"
#include <stdlib.h>
#include <string.h>
#ifdef LIBGOMP_USE_PTHREADS
pthread_attr_t gomp_thread_attr;
/* This key is for the thread destructor. */
pthread_key_t gomp_thread_destructor;
/* This is the libgomp per-thread data structure. */
#if defined HAVE_TLS || defined USE_EMUTLS
__thread struct gomp_thread gomp_tls_data;
#else
pthread_key_t gomp_tls_key;
#endif
/* This structure is used to communicate across pthread_create. */
struct gomp_thread_start_data
{
void (*fn) (void *);
void *fn_data;
struct gomp_team_state ts;
struct gomp_task *task;
struct gomp_thread_pool *thread_pool;
unsigned int place;
unsigned int num_teams;
unsigned int team_num;
bool nested;
pthread_t handle;
};
/* This function is a pthread_create entry point. This contains the idle
loop in which a thread waits to be called up to become part of a team. */
static void *
gomp_thread_start (void *xdata)
{
struct gomp_thread_start_data *data = xdata;
struct gomp_thread *thr;
struct gomp_thread_pool *pool;
void (*local_fn) (void *);
void *local_data;
#if defined HAVE_TLS || defined USE_EMUTLS
thr = &gomp_tls_data;
#else
struct gomp_thread local_thr;
thr = &local_thr;
#endif
gomp_sem_init (&thr->release, 0);
/* Extract what we need from data. */
local_fn = data->fn;
local_data = data->fn_data;
thr->thread_pool = data->thread_pool;
thr->ts = data->ts;
thr->task = data->task;
thr->place = data->place;
thr->num_teams = data->num_teams;
thr->team_num = data->team_num;
#ifdef GOMP_NEEDS_THREAD_HANDLE
thr->handle = data->handle;
#endif
#if !(defined HAVE_TLS || defined USE_EMUTLS)
pthread_setspecific (gomp_tls_key, thr);
#endif
thr->ts.team->ordered_release[thr->ts.team_id] = &thr->release;
/* Make thread pool local. */
pool = thr->thread_pool;
if (data->nested)
{
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
gomp_barrier_wait (&team->barrier);
local_fn (local_data);
gomp_team_barrier_wait_final (&team->barrier);
gomp_finish_task (task);
gomp_barrier_wait_last (&team->barrier);
}
else
{
pool->threads[thr->ts.team_id] = thr;
gomp_simple_barrier_wait (&pool->threads_dock);
do
{
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
local_fn (local_data);
gomp_team_barrier_wait_final (&team->barrier);
gomp_finish_task (task);
gomp_simple_barrier_wait (&pool->threads_dock);
local_fn = thr->fn;
local_data = thr->data;
thr->fn = NULL;
}
while (local_fn);
}
gomp_sem_destroy (&thr->release);
pthread_detach (pthread_self ());
thr->thread_pool = NULL;
thr->task = NULL;
return NULL;
}
#endif
static inline struct gomp_team *
get_last_team (unsigned nthreads)
{
struct gomp_thread *thr = gomp_thread ();
if (thr->ts.team == NULL)
{
struct gomp_thread_pool *pool = gomp_get_thread_pool (thr, nthreads);
struct gomp_team *last_team = pool->last_team;
if (last_team != NULL && last_team->nthreads == nthreads)
{
pool->last_team = NULL;
return last_team;
}
}
return NULL;
}
/* Create a new team data structure. */
struct gomp_team *
gomp_new_team (unsigned nthreads)
{
struct gomp_team *team;
int i;
team = get_last_team (nthreads);
if (team == NULL)
{
size_t extra = sizeof (team->ordered_release[0])
+ sizeof (team->implicit_task[0]);
#ifdef GOMP_USE_ALIGNED_WORK_SHARES
team = gomp_aligned_alloc (__alignof (struct gomp_team),
sizeof (*team) + nthreads * extra);
#else
team = team_malloc (sizeof (*team) + nthreads * extra);
#endif
#ifndef HAVE_SYNC_BUILTINS
gomp_mutex_init (&team->work_share_list_free_lock);
#endif
gomp_barrier_init (&team->barrier, nthreads);
gomp_mutex_init (&team->task_lock);
team->nthreads = nthreads;
}
team->work_share_chunk = 8;
#ifdef HAVE_SYNC_BUILTINS
team->single_count = 0;
#endif
team->work_shares_to_free = &team->work_shares[0];
gomp_init_work_share (&team->work_shares[0], 0, nthreads);
team->work_shares[0].next_alloc = NULL;
team->work_share_list_free = NULL;
team->work_share_list_alloc = &team->work_shares[1];
for (i = 1; i < 7; i++)
team->work_shares[i].next_free = &team->work_shares[i + 1];
team->work_shares[i].next_free = NULL;
gomp_sem_init (&team->master_release, 0);
team->ordered_release = (void *) &team->implicit_task[nthreads];
team->ordered_release[0] = &team->master_release;
priority_queue_init (&team->task_queue);
team->task_count = 0;
team->task_queued_count = 0;
team->task_running_count = 0;
team->work_share_cancelled = 0;
team->team_cancelled = 0;
team->task_detach_count = 0;
return team;
}
/* Free a team data structure. */
static void
free_team (struct gomp_team *team)
{
#ifndef HAVE_SYNC_BUILTINS
gomp_mutex_destroy (&team->work_share_list_free_lock);
#endif
gomp_barrier_destroy (&team->barrier);
gomp_mutex_destroy (&team->task_lock);
priority_queue_free (&team->task_queue);
team_free (team);
}
static void
gomp_free_pool_helper (void *thread_pool)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_thread_pool *pool
= (struct gomp_thread_pool *) thread_pool;
gomp_simple_barrier_wait_last (&pool->threads_dock);
gomp_sem_destroy (&thr->release);
thr->thread_pool = NULL;
thr->task = NULL;
#ifdef LIBGOMP_USE_PTHREADS
pthread_detach (pthread_self ());
pthread_exit (NULL);
#elif defined(__nvptx__)
asm ("exit;");
#elif defined(__AMDGCN__)
asm ("s_endpgm");
#else
#error gomp_free_pool_helper must terminate the thread
#endif
}
/* Free a thread pool and release its threads. */
void
gomp_free_thread (void *arg __attribute__((unused)))
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_thread_pool *pool = thr->thread_pool;
if (pool)
{
if (pool->threads_used > 0)
{
int i;
for (i = 1; i < pool->threads_used; i++)
{
struct gomp_thread *nthr = pool->threads[i];
nthr->fn = gomp_free_pool_helper;
nthr->data = pool;
}
/* This barrier undocks threads docked on pool->threads_dock. */
gomp_simple_barrier_wait (&pool->threads_dock);
/* And this waits till all threads have called gomp_barrier_wait_last
in gomp_free_pool_helper. */
gomp_simple_barrier_wait (&pool->threads_dock);
/* Now it is safe to destroy the barrier and free the pool. */
gomp_simple_barrier_destroy (&pool->threads_dock);
#ifdef HAVE_SYNC_BUILTINS
__sync_fetch_and_add (&gomp_managed_threads,
1L - pool->threads_used);
#else
gomp_mutex_lock (&gomp_managed_threads_lock);
gomp_managed_threads -= pool->threads_used - 1L;
gomp_mutex_unlock (&gomp_managed_threads_lock);
#endif
}
if (pool->last_team)
free_team (pool->last_team);
#ifndef __nvptx__
team_free (pool->threads);
team_free (pool);
#endif
thr->thread_pool = NULL;
}
if (thr->ts.level == 0 && __builtin_expect (thr->ts.team != NULL, 0))
gomp_team_end ();
if (thr->task != NULL)
{
struct gomp_task *task = thr->task;
gomp_end_task ();
free (task);
}
}
/* Launch a team. */
#ifdef LIBGOMP_USE_PTHREADS
void
gomp_team_start (void (*fn) (void *), void *data, unsigned nthreads,
unsigned flags, struct gomp_team *team,
struct gomp_taskgroup *taskgroup)
{
struct gomp_thread_start_data *start_data = NULL;
struct gomp_thread *thr, *nthr;
struct gomp_task *task;
struct gomp_task_icv *icv;
bool nested;
struct gomp_thread_pool *pool;
unsigned i, n, old_threads_used = 0;
pthread_attr_t thread_attr, *attr;
unsigned long nthreads_var;
char bind, bind_var;
unsigned int s = 0, rest = 0, p = 0, k = 0;
unsigned int affinity_count = 0;
struct gomp_thread **affinity_thr = NULL;
bool force_display = false;
thr = gomp_thread ();
nested = thr->ts.level;
pool = thr->thread_pool;
task = thr->task;
icv = task ? &task->icv : &gomp_global_icv;
if (__builtin_expect (gomp_places_list != NULL, 0) && thr->place == 0)
{
gomp_init_affinity ();
if (__builtin_expect (gomp_display_affinity_var, 0) && nthreads == 1)
gomp_display_affinity_thread (gomp_thread_self (), &thr->ts,
thr->place);
}
/* Always save the previous state, even if this isn't a nested team.
In particular, we should save any work share state from an outer
orphaned work share construct. */
team->prev_ts = thr->ts;
thr->ts.team = team;
thr->ts.team_id = 0;
++thr->ts.level;
if (nthreads > 1)
++thr->ts.active_level;
thr->ts.work_share = &team->work_shares[0];
thr->ts.last_work_share = NULL;
#ifdef HAVE_SYNC_BUILTINS
thr->ts.single_count = 0;
#endif
thr->ts.static_trip = 0;
thr->task = &team->implicit_task[0];
#ifdef GOMP_NEEDS_THREAD_HANDLE
thr->handle = pthread_self ();
#endif
nthreads_var = icv->nthreads_var;
if (__builtin_expect (gomp_nthreads_var_list != NULL, 0)
&& thr->ts.level < gomp_nthreads_var_list_len)
nthreads_var = gomp_nthreads_var_list[thr->ts.level];
bind_var = icv->bind_var;
if (bind_var != omp_proc_bind_false && (flags & 7) != omp_proc_bind_false)
bind_var = flags & 7;
bind = bind_var;
if (__builtin_expect (gomp_bind_var_list != NULL, 0)
&& thr->ts.level < gomp_bind_var_list_len)
bind_var = gomp_bind_var_list[thr->ts.level];
gomp_init_task (thr->task, task, icv);
thr->task->taskgroup = taskgroup;
team->implicit_task[0].icv.nthreads_var = nthreads_var;
team->implicit_task[0].icv.bind_var = bind_var;
if (nthreads == 1)
return;
i = 1;
if (__builtin_expect (gomp_places_list != NULL, 0))
{
/* Depending on chosen proc_bind model, set subpartition
for the master thread and initialize helper variables
P and optionally S, K and/or REST used by later place
computation for each additional thread. */
p = thr->place - 1;
switch (bind)
{
case omp_proc_bind_true:
case omp_proc_bind_close:
if (nthreads > thr->ts.place_partition_len)
{
/* T > P. S threads will be placed in each place,
and the final REM threads placed one by one
into the already occupied places. */
s = nthreads / thr->ts.place_partition_len;
rest = nthreads % thr->ts.place_partition_len;
}
else
s = 1;
k = 1;
break;
case omp_proc_bind_master:
/* Each thread will be bound to master's place. */
break;
case omp_proc_bind_spread:
if (nthreads <= thr->ts.place_partition_len)
{
/* T <= P. Each subpartition will have in between s
and s+1 places (subpartitions starting at or
after rest will have s places, earlier s+1 places),
each thread will be bound to the first place in
its subpartition (except for the master thread
that can be bound to another place in its
subpartition). */
s = thr->ts.place_partition_len / nthreads;
rest = thr->ts.place_partition_len % nthreads;
rest = (s + 1) * rest + thr->ts.place_partition_off;
if (p < rest)
{
p -= (p - thr->ts.place_partition_off) % (s + 1);
thr->ts.place_partition_len = s + 1;
}
else
{
p -= (p - rest) % s;
thr->ts.place_partition_len = s;
}
thr->ts.place_partition_off = p;
}
else
{
/* T > P. Each subpartition will have just a single
place and we'll place between s and s+1
threads into each subpartition. */
s = nthreads / thr->ts.place_partition_len;
rest = nthreads % thr->ts.place_partition_len;
thr->ts.place_partition_off = p;
thr->ts.place_partition_len = 1;
k = 1;
}
break;
}
}
else
bind = omp_proc_bind_false;
/* We only allow the reuse of idle threads for non-nested PARALLEL
regions. This appears to be implied by the semantics of
threadprivate variables, but perhaps that's reading too much into
things. Certainly it does prevent any locking problems, since
only the initial program thread will modify gomp_threads. */
if (!nested)
{
old_threads_used = pool->threads_used;
if (nthreads <= old_threads_used)
n = nthreads;
else if (old_threads_used == 0)
{
n = 0;
gomp_simple_barrier_init (&pool->threads_dock, nthreads);
}
else
{
n = old_threads_used;
/* Increase the barrier threshold to make sure all new
threads arrive before the team is released. */
gomp_simple_barrier_reinit (&pool->threads_dock, nthreads);
}
/* Not true yet, but soon will be. We're going to release all
threads from the dock, and those that aren't part of the
team will exit. */
pool->threads_used = nthreads;
/* If necessary, expand the size of the gomp_threads array. It is
expected that changes in the number of threads are rare, thus we
make no effort to expand gomp_threads_size geometrically. */
if (nthreads >= pool->threads_size)
{
pool->threads_size = nthreads + 1;
pool->threads
= gomp_realloc (pool->threads,
pool->threads_size
* sizeof (struct gomp_thread *));
/* Add current (master) thread to threads[]. */
pool->threads[0] = thr;
}
/* Release existing idle threads. */
for (; i < n; ++i)
{
unsigned int place_partition_off = thr->ts.place_partition_off;
unsigned int place_partition_len = thr->ts.place_partition_len;
unsigned int place = 0;
if (__builtin_expect (gomp_places_list != NULL, 0))
{
switch (bind)
{
case omp_proc_bind_true:
case omp_proc_bind_close:
if (k == s)
{
++p;
if (p == (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len))
p = team->prev_ts.place_partition_off;
k = 1;
if (i == nthreads - rest)
s = 1;
}
else
++k;
break;
case omp_proc_bind_master:
break;
case omp_proc_bind_spread:
if (k == 0)
{
/* T <= P. */
if (p < rest)
p += s + 1;
else
p += s;
if (p == (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len))
p = team->prev_ts.place_partition_off;
place_partition_off = p;
if (p < rest)
place_partition_len = s + 1;
else
place_partition_len = s;
}
else
{
/* T > P. */
if (k == s)
{
++p;
if (p == (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len))
p = team->prev_ts.place_partition_off;
k = 1;
if (i == nthreads - rest)
s = 1;
}
else
++k;
place_partition_off = p;
place_partition_len = 1;
}
break;
}
if (affinity_thr != NULL
|| (bind != omp_proc_bind_true
&& pool->threads[i]->place != p + 1)
|| pool->threads[i]->place <= place_partition_off
|| pool->threads[i]->place > (place_partition_off
+ place_partition_len))
{
unsigned int l;
force_display = true;
if (affinity_thr == NULL)
{
unsigned int j;
if (team->prev_ts.place_partition_len > 64)
affinity_thr
= gomp_malloc (team->prev_ts.place_partition_len
* sizeof (struct gomp_thread *));
else
affinity_thr
= gomp_alloca (team->prev_ts.place_partition_len
* sizeof (struct gomp_thread *));
memset (affinity_thr, '\0',
team->prev_ts.place_partition_len
* sizeof (struct gomp_thread *));
for (j = i; j < old_threads_used; j++)
{
if (pool->threads[j]->place
> team->prev_ts.place_partition_off
&& (pool->threads[j]->place
<= (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len)))
{
l = pool->threads[j]->place - 1
- team->prev_ts.place_partition_off;
pool->threads[j]->data = affinity_thr[l];
affinity_thr[l] = pool->threads[j];
}
pool->threads[j] = NULL;
}
if (nthreads > old_threads_used)
memset (&pool->threads[old_threads_used],
'\0', ((nthreads - old_threads_used)
* sizeof (struct gomp_thread *)));
n = nthreads;
affinity_count = old_threads_used - i;
}
if (affinity_count == 0)
break;
l = p;
if (affinity_thr[l - team->prev_ts.place_partition_off]
== NULL)
{
if (bind != omp_proc_bind_true)
continue;
for (l = place_partition_off;
l < place_partition_off + place_partition_len;
l++)
if (affinity_thr[l - team->prev_ts.place_partition_off]
!= NULL)
break;
if (l == place_partition_off + place_partition_len)
continue;
}
nthr = affinity_thr[l - team->prev_ts.place_partition_off];
affinity_thr[l - team->prev_ts.place_partition_off]
= (struct gomp_thread *) nthr->data;
affinity_count--;
pool->threads[i] = nthr;
}
else
nthr = pool->threads[i];
place = p + 1;
}
else
nthr = pool->threads[i];
nthr->ts.team = team;
nthr->ts.work_share = &team->work_shares[0];
nthr->ts.last_work_share = NULL;
nthr->ts.team_id = i;
nthr->ts.level = team->prev_ts.level + 1;
nthr->ts.active_level = thr->ts.active_level;
nthr->ts.place_partition_off = place_partition_off;
nthr->ts.place_partition_len = place_partition_len;
nthr->ts.def_allocator = thr->ts.def_allocator;
#ifdef HAVE_SYNC_BUILTINS
nthr->ts.single_count = 0;
#endif
nthr->ts.static_trip = 0;
nthr->num_teams = thr->num_teams;
nthr->team_num = thr->team_num;
nthr->task = &team->implicit_task[i];
nthr->place = place;
gomp_init_task (nthr->task, task, icv);
team->implicit_task[i].icv.nthreads_var = nthreads_var;
team->implicit_task[i].icv.bind_var = bind_var;
nthr->task->taskgroup = taskgroup;
nthr->fn = fn;
nthr->data = data;
team->ordered_release[i] = &nthr->release;
}
if (__builtin_expect (affinity_thr != NULL, 0))
{
/* If AFFINITY_THR is non-NULL just because we had to
permute some threads in the pool, but we've managed
to find exactly as many old threads as we'd find
without affinity, we don't need to handle this
specially anymore. */
if (nthreads <= old_threads_used
? (affinity_count == old_threads_used - nthreads)
: (i == old_threads_used))
{
if (team->prev_ts.place_partition_len > 64)
free (affinity_thr);
affinity_thr = NULL;
affinity_count = 0;
}
else
{
i = 1;
/* We are going to compute the places/subpartitions
again from the beginning. So, we need to reinitialize
vars modified by the switch (bind) above inside
of the loop, to the state they had after the initial
switch (bind). */
switch (bind)
{
case omp_proc_bind_true:
case omp_proc_bind_close:
if (nthreads > thr->ts.place_partition_len)
/* T > P. S has been changed, so needs
to be recomputed. */
s = nthreads / thr->ts.place_partition_len;
k = 1;
p = thr->place - 1;
break;
case omp_proc_bind_master:
/* No vars have been changed. */
break;
case omp_proc_bind_spread:
p = thr->ts.place_partition_off;
if (k != 0)
{
/* T > P. */
s = nthreads / team->prev_ts.place_partition_len;
k = 1;
}
break;
}
/* Increase the barrier threshold to make sure all new
threads and all the threads we're going to let die
arrive before the team is released. */
if (affinity_count)
gomp_simple_barrier_reinit (&pool->threads_dock,
nthreads + affinity_count);
}
}
if (i == nthreads)
goto do_release;
}
if (__builtin_expect (nthreads + affinity_count > old_threads_used, 0))
{
long diff = (long) (nthreads + affinity_count) - (long) old_threads_used;
if (old_threads_used == 0)
--diff;
#ifdef HAVE_SYNC_BUILTINS
__sync_fetch_and_add (&gomp_managed_threads, diff);
#else
gomp_mutex_lock (&gomp_managed_threads_lock);
gomp_managed_threads += diff;
gomp_mutex_unlock (&gomp_managed_threads_lock);
#endif
}
attr = &gomp_thread_attr;
if (__builtin_expect (gomp_places_list != NULL, 0))
{
size_t stacksize;
pthread_attr_init (&thread_attr);
if (! pthread_attr_getstacksize (&gomp_thread_attr, &stacksize))
pthread_attr_setstacksize (&thread_attr, stacksize);
attr = &thread_attr;
}
if (i >= nthreads)
__builtin_unreachable ();
start_data = gomp_alloca (sizeof (struct gomp_thread_start_data)
* (nthreads - i));
/* Launch new threads. */
for (; i < nthreads; ++i)
{
int err;
start_data->ts.place_partition_off = thr->ts.place_partition_off;
start_data->ts.place_partition_len = thr->ts.place_partition_len;
start_data->place = 0;
if (__builtin_expect (gomp_places_list != NULL, 0))
{
switch (bind)
{
case omp_proc_bind_true:
case omp_proc_bind_close:
if (k == s)
{
++p;
if (p == (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len))
p = team->prev_ts.place_partition_off;
k = 1;
if (i == nthreads - rest)
s = 1;
}
else
++k;
break;
case omp_proc_bind_master:
break;
case omp_proc_bind_spread:
if (k == 0)
{
/* T <= P. */
if (p < rest)
p += s + 1;
else
p += s;
if (p == (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len))
p = team->prev_ts.place_partition_off;
start_data->ts.place_partition_off = p;
if (p < rest)
start_data->ts.place_partition_len = s + 1;
else
start_data->ts.place_partition_len = s;
}
else
{
/* T > P. */
if (k == s)
{
++p;
if (p == (team->prev_ts.place_partition_off
+ team->prev_ts.place_partition_len))
p = team->prev_ts.place_partition_off;
k = 1;
if (i == nthreads - rest)
s = 1;
}
else
++k;
start_data->ts.place_partition_off = p;
start_data->ts.place_partition_len = 1;
}
break;
}
start_data->place = p + 1;
if (affinity_thr != NULL && pool->threads[i] != NULL)
continue;
gomp_init_thread_affinity (attr, p);
}
start_data->fn = fn;
start_data->fn_data = data;
start_data->ts.team = team;
start_data->ts.work_share = &team->work_shares[0];
start_data->ts.last_work_share = NULL;
start_data->ts.team_id = i;
start_data->ts.level = team->prev_ts.level + 1;
start_data->ts.active_level = thr->ts.active_level;
start_data->ts.def_allocator = thr->ts.def_allocator;
#ifdef HAVE_SYNC_BUILTINS
start_data->ts.single_count = 0;
#endif
start_data->ts.static_trip = 0;
start_data->num_teams = thr->num_teams;
start_data->team_num = thr->team_num;
start_data->task = &team->implicit_task[i];
gomp_init_task (start_data->task, task, icv);
team->implicit_task[i].icv.nthreads_var = nthreads_var;
team->implicit_task[i].icv.bind_var = bind_var;
start_data->task->taskgroup = taskgroup;
start_data->thread_pool = pool;
start_data->nested = nested;
attr = gomp_adjust_thread_attr (attr, &thread_attr);
err = pthread_create (&start_data->handle, attr, gomp_thread_start,
start_data);
start_data++;
if (err != 0)
gomp_fatal ("Thread creation failed: %s", strerror (err));
}
if (__builtin_expect (attr == &thread_attr, 0))
pthread_attr_destroy (&thread_attr);
do_release:
if (nested)
gomp_barrier_wait (&team->barrier);
else
gomp_simple_barrier_wait (&pool->threads_dock);
/* Decrease the barrier threshold to match the number of threads
that should arrive back at the end of this team. The extra
threads should be exiting. Note that we arrange for this test
to never be true for nested teams. If AFFINITY_COUNT is non-zero,
the barrier as well as gomp_managed_threads was temporarily
set to NTHREADS + AFFINITY_COUNT. For NTHREADS < OLD_THREADS_COUNT,
AFFINITY_COUNT if non-zero will be always at least
OLD_THREADS_COUNT - NTHREADS. */
if (__builtin_expect (nthreads < old_threads_used, 0)
|| __builtin_expect (affinity_count, 0))
{
long diff = (long) nthreads - (long) old_threads_used;
if (affinity_count)
diff = -affinity_count;
gomp_simple_barrier_reinit (&pool->threads_dock, nthreads);
#ifdef HAVE_SYNC_BUILTINS
__sync_fetch_and_add (&gomp_managed_threads, diff);
#else
gomp_mutex_lock (&gomp_managed_threads_lock);
gomp_managed_threads += diff;
gomp_mutex_unlock (&gomp_managed_threads_lock);
#endif
}
if (__builtin_expect (gomp_display_affinity_var, 0))
{
if (nested
|| nthreads != old_threads_used
|| force_display)
{
gomp_display_affinity_thread (gomp_thread_self (), &thr->ts,
thr->place);
if (nested)
{
start_data -= nthreads - 1;
for (i = 1; i < nthreads; ++i)
{
gomp_display_affinity_thread (
#ifdef LIBGOMP_USE_PTHREADS
start_data->handle,
#else
gomp_thread_self (),
#endif
&start_data->ts,
start_data->place);
start_data++;
}
}
else
{
for (i = 1; i < nthreads; ++i)
{
gomp_thread_handle handle
= gomp_thread_to_pthread_t (pool->threads[i]);
gomp_display_affinity_thread (handle, &pool->threads[i]->ts,
pool->threads[i]->place);
}
}
}
}
if (__builtin_expect (affinity_thr != NULL, 0)
&& team->prev_ts.place_partition_len > 64)
free (affinity_thr);
}
#endif
/* Terminate the current team. This is only to be called by the master
thread. We assume that we must wait for the other threads. */
void
gomp_team_end (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
/* This barrier handles all pending explicit threads.
As #pragma omp cancel parallel might get awaited count in
team->barrier in a inconsistent state, we need to use a different
counter here. */
gomp_team_barrier_wait_final (&team->barrier);
if (__builtin_expect (team->team_cancelled, 0))
{
struct gomp_work_share *ws = team->work_shares_to_free;
do
{
struct gomp_work_share *next_ws = gomp_ptrlock_get (&ws->next_ws);
if (next_ws == NULL)
gomp_ptrlock_set (&ws->next_ws, ws);
gomp_fini_work_share (ws);
ws = next_ws;
}
while (ws != NULL);
}
else
gomp_fini_work_share (thr->ts.work_share);
gomp_end_task ();
thr->ts = team->prev_ts;
if (__builtin_expect (thr->ts.level != 0, 0))
{
#ifdef HAVE_SYNC_BUILTINS
__sync_fetch_and_add (&gomp_managed_threads, 1L - team->nthreads);
#else
gomp_mutex_lock (&gomp_managed_threads_lock);
gomp_managed_threads -= team->nthreads - 1L;
gomp_mutex_unlock (&gomp_managed_threads_lock);
#endif
/* This barrier has gomp_barrier_wait_last counterparts
and ensures the team can be safely destroyed. */
gomp_barrier_wait (&team->barrier);
}
if (__builtin_expect (team->work_shares[0].next_alloc != NULL, 0))
{
struct gomp_work_share *ws = team->work_shares[0].next_alloc;
do
{
struct gomp_work_share *next_ws = ws->next_alloc;
free (ws);
ws = next_ws;
}
while (ws != NULL);