forked from gcc-mirror/gcc
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtask.c
2339 lines (2173 loc) · 68 KB
/
task.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) 2007-2020 Free Software Foundation, Inc.
Contributed by Richard Henderson <[email protected]>.
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 tasks in response to task
creation and termination. */
#include "libgomp.h"
#include <stdlib.h>
#include <string.h>
#include "gomp-constants.h"
typedef struct gomp_task_depend_entry *hash_entry_type;
static inline void *
htab_alloc (size_t size)
{
return gomp_malloc (size);
}
static inline void
htab_free (void *ptr)
{
free (ptr);
}
#include "hashtab.h"
static inline hashval_t
htab_hash (hash_entry_type element)
{
return hash_pointer (element->addr);
}
static inline bool
htab_eq (hash_entry_type x, hash_entry_type y)
{
return x->addr == y->addr;
}
/* Create a new task data structure. */
void
gomp_init_task (struct gomp_task *task, struct gomp_task *parent_task,
struct gomp_task_icv *prev_icv)
{
/* It would seem that using memset here would be a win, but it turns
out that partially filling gomp_task allows us to keep the
overhead of task creation low. In the nqueens-1.c test, for a
sufficiently large N, we drop the overhead from 5-6% to 1%.
Note, the nqueens-1.c test in serial mode is a good test to
benchmark the overhead of creating tasks as there are millions of
tiny tasks created that all run undeferred. */
task->parent = parent_task;
task->icv = *prev_icv;
task->kind = GOMP_TASK_IMPLICIT;
task->taskwait = NULL;
task->in_tied_task = false;
task->final_task = false;
task->copy_ctors_done = false;
task->parent_depends_on = false;
priority_queue_init (&task->children_queue);
task->taskgroup = NULL;
task->dependers = NULL;
task->depend_hash = NULL;
task->depend_count = 0;
}
/* Clean up a task, after completing it. */
void
gomp_end_task (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task *task = thr->task;
gomp_finish_task (task);
thr->task = task->parent;
}
/* Clear the parent field of every task in LIST. */
static inline void
gomp_clear_parent_in_list (struct priority_list *list)
{
struct priority_node *p = list->tasks;
if (p)
do
{
priority_node_to_task (PQ_CHILDREN, p)->parent = NULL;
p = p->next;
}
while (p != list->tasks);
}
/* Splay tree version of gomp_clear_parent_in_list.
Clear the parent field of every task in NODE within SP, and free
the node when done. */
static void
gomp_clear_parent_in_tree (prio_splay_tree sp, prio_splay_tree_node node)
{
if (!node)
return;
prio_splay_tree_node left = node->left, right = node->right;
gomp_clear_parent_in_list (&node->key.l);
#if _LIBGOMP_CHECKING_
memset (node, 0xaf, sizeof (*node));
#endif
/* No need to remove the node from the tree. We're nuking
everything, so just free the nodes and our caller can clear the
entire splay tree. */
free (node);
gomp_clear_parent_in_tree (sp, left);
gomp_clear_parent_in_tree (sp, right);
}
/* Clear the parent field of every task in Q and remove every task
from Q. */
static inline void
gomp_clear_parent (struct priority_queue *q)
{
if (priority_queue_multi_p (q))
{
gomp_clear_parent_in_tree (&q->t, q->t.root);
/* All the nodes have been cleared in gomp_clear_parent_in_tree.
No need to remove anything. We can just nuke everything. */
q->t.root = NULL;
}
else
gomp_clear_parent_in_list (&q->l);
}
/* Helper function for GOMP_task and gomp_create_target_task.
For a TASK with in/out dependencies, fill in the various dependency
queues. PARENT is the parent of said task. DEPEND is as in
GOMP_task. */
static void
gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
void **depend)
{
size_t ndepend = (uintptr_t) depend[0];
size_t i;
hash_entry_type ent;
if (ndepend)
{
/* depend[0] is total # */
size_t nout = (uintptr_t) depend[1]; /* # of out: and inout: */
/* ndepend - nout is # of in: */
for (i = 0; i < ndepend; i++)
{
task->depend[i].addr = depend[2 + i];
task->depend[i].is_in = i >= nout;
}
}
else
{
ndepend = (uintptr_t) depend[1]; /* total # */
size_t nout = (uintptr_t) depend[2]; /* # of out: and inout: */
size_t nmutexinoutset = (uintptr_t) depend[3]; /* # of mutexinoutset: */
/* For now we treat mutexinoutset like out, which is compliant, but
inefficient. */
size_t nin = (uintptr_t) depend[4]; /* # of in: */
/* ndepend - nout - nmutexinoutset - nin is # of depobjs */
size_t normal = nout + nmutexinoutset + nin;
size_t n = 0;
for (i = normal; i < ndepend; i++)
{
void **d = (void **) (uintptr_t) depend[5 + i];
switch ((uintptr_t) d[1])
{
case GOMP_DEPEND_OUT:
case GOMP_DEPEND_INOUT:
case GOMP_DEPEND_MUTEXINOUTSET:
break;
case GOMP_DEPEND_IN:
continue;
default:
gomp_fatal ("unknown omp_depend_t dependence type %d",
(int) (uintptr_t) d[1]);
}
task->depend[n].addr = d[0];
task->depend[n++].is_in = 0;
}
for (i = 0; i < normal; i++)
{
task->depend[n].addr = depend[5 + i];
task->depend[n++].is_in = i >= nout + nmutexinoutset;
}
for (i = normal; i < ndepend; i++)
{
void **d = (void **) (uintptr_t) depend[5 + i];
if ((uintptr_t) d[1] != GOMP_DEPEND_IN)
continue;
task->depend[n].addr = d[0];
task->depend[n++].is_in = 1;
}
}
task->depend_count = ndepend;
task->num_dependees = 0;
if (parent->depend_hash == NULL)
parent->depend_hash = htab_create (2 * ndepend > 12 ? 2 * ndepend : 12);
for (i = 0; i < ndepend; i++)
{
task->depend[i].next = NULL;
task->depend[i].prev = NULL;
task->depend[i].task = task;
task->depend[i].redundant = false;
task->depend[i].redundant_out = false;
hash_entry_type *slot = htab_find_slot (&parent->depend_hash,
&task->depend[i], INSERT);
hash_entry_type out = NULL, last = NULL;
if (*slot)
{
/* If multiple depends on the same task are the same, all but the
first one are redundant. As inout/out come first, if any of them
is inout/out, it will win, which is the right semantics. */
if ((*slot)->task == task)
{
task->depend[i].redundant = true;
continue;
}
for (ent = *slot; ent; ent = ent->next)
{
if (ent->redundant_out)
break;
last = ent;
/* depend(in:...) doesn't depend on earlier depend(in:...). */
if (task->depend[i].is_in && ent->is_in)
continue;
if (!ent->is_in)
out = ent;
struct gomp_task *tsk = ent->task;
if (tsk->dependers == NULL)
{
tsk->dependers
= gomp_malloc (sizeof (struct gomp_dependers_vec)
+ 6 * sizeof (struct gomp_task *));
tsk->dependers->n_elem = 1;
tsk->dependers->allocated = 6;
tsk->dependers->elem[0] = task;
task->num_dependees++;
continue;
}
/* We already have some other dependency on tsk from earlier
depend clause. */
else if (tsk->dependers->n_elem
&& (tsk->dependers->elem[tsk->dependers->n_elem - 1]
== task))
continue;
else if (tsk->dependers->n_elem == tsk->dependers->allocated)
{
tsk->dependers->allocated
= tsk->dependers->allocated * 2 + 2;
tsk->dependers
= gomp_realloc (tsk->dependers,
sizeof (struct gomp_dependers_vec)
+ (tsk->dependers->allocated
* sizeof (struct gomp_task *)));
}
tsk->dependers->elem[tsk->dependers->n_elem++] = task;
task->num_dependees++;
}
task->depend[i].next = *slot;
(*slot)->prev = &task->depend[i];
}
*slot = &task->depend[i];
/* There is no need to store more than one depend({,in}out:) task per
address in the hash table chain for the purpose of creation of
deferred tasks, because each out depends on all earlier outs, thus it
is enough to record just the last depend({,in}out:). For depend(in:),
we need to keep all of the previous ones not terminated yet, because
a later depend({,in}out:) might need to depend on all of them. So, if
the new task's clause is depend({,in}out:), we know there is at most
one other depend({,in}out:) clause in the list (out). For
non-deferred tasks we want to see all outs, so they are moved to the
end of the chain, after first redundant_out entry all following
entries should be redundant_out. */
if (!task->depend[i].is_in && out)
{
if (out != last)
{
out->next->prev = out->prev;
out->prev->next = out->next;
out->next = last->next;
out->prev = last;
last->next = out;
if (out->next)
out->next->prev = out;
}
out->redundant_out = true;
}
}
}
/* Called when encountering an explicit task directive. If IF_CLAUSE is
false, then we must not delay in executing the task. If UNTIED is true,
then the task may be executed by any member of the team.
DEPEND is an array containing:
if depend[0] is non-zero, then:
depend[0]: number of depend elements.
depend[1]: number of depend elements of type "out/inout".
depend[2..N+1]: address of [1..N]th depend element.
otherwise, when depend[0] is zero, then:
depend[1]: number of depend elements.
depend[2]: number of depend elements of type "out/inout".
depend[3]: number of depend elements of type "mutexinoutset".
depend[4]: number of depend elements of type "in".
depend[5..4+depend[2]+depend[3]+depend[4]]: address of depend elements
depend[5+depend[2]+depend[3]+depend[4]..4+depend[1]]: address of
omp_depend_t objects. */
void
GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
long arg_size, long arg_align, bool if_clause, unsigned flags,
void **depend, int priority)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
#ifdef HAVE_BROKEN_POSIX_SEMAPHORES
/* If pthread_mutex_* is used for omp_*lock*, then each task must be
tied to one thread all the time. This means UNTIED tasks must be
tied and if CPYFN is non-NULL IF(0) must be forced, as CPYFN
might be running on different thread than FN. */
if (cpyfn)
if_clause = false;
flags &= ~GOMP_TASK_FLAG_UNTIED;
#endif
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (__builtin_expect (gomp_cancel_var, 0) && team)
{
if (gomp_team_barrier_cancelled (&team->barrier))
return;
if (thr->task->taskgroup)
{
if (thr->task->taskgroup->cancelled)
return;
if (thr->task->taskgroup->workshare
&& thr->task->taskgroup->prev
&& thr->task->taskgroup->prev->cancelled)
return;
}
}
if ((flags & GOMP_TASK_FLAG_PRIORITY) == 0)
priority = 0;
else if (priority > gomp_max_task_priority_var)
priority = gomp_max_task_priority_var;
if (!if_clause || team == NULL
|| (thr->task && thr->task->final_task)
|| team->task_count > 64 * team->nthreads)
{
struct gomp_task task;
/* If there are depend clauses and earlier deferred sibling tasks
with depend clauses, check if there isn't a dependency. If there
is, we need to wait for them. There is no need to handle
depend clauses for non-deferred tasks other than this, because
the parent task is suspended until the child task finishes and thus
it can't start further child tasks. */
if ((flags & GOMP_TASK_FLAG_DEPEND)
&& thr->task && thr->task->depend_hash)
gomp_task_maybe_wait_for_dependencies (depend);
gomp_init_task (&task, thr->task, gomp_icv (false));
task.kind = GOMP_TASK_UNDEFERRED;
task.final_task = (thr->task && thr->task->final_task)
|| (flags & GOMP_TASK_FLAG_FINAL);
task.priority = priority;
if (thr->task)
{
task.in_tied_task = thr->task->in_tied_task;
task.taskgroup = thr->task->taskgroup;
}
thr->task = &task;
if (__builtin_expect (cpyfn != NULL, 0))
{
char buf[arg_size + arg_align - 1];
char *arg = (char *) (((uintptr_t) buf + arg_align - 1)
& ~(uintptr_t) (arg_align - 1));
cpyfn (arg, data);
fn (arg);
}
else
fn (data);
/* Access to "children" is normally done inside a task_lock
mutex region, but the only way this particular task.children
can be set is if this thread's task work function (fn)
creates children. So since the setter is *this* thread, we
need no barriers here when testing for non-NULL. We can have
task.children set by the current thread then changed by a
child thread, but seeing a stale non-NULL value is not a
problem. Once past the task_lock acquisition, this thread
will see the real value of task.children. */
if (!priority_queue_empty_p (&task.children_queue, MEMMODEL_RELAXED))
{
gomp_mutex_lock (&team->task_lock);
gomp_clear_parent (&task.children_queue);
gomp_mutex_unlock (&team->task_lock);
}
gomp_end_task ();
}
else
{
struct gomp_task *task;
struct gomp_task *parent = thr->task;
struct gomp_taskgroup *taskgroup = parent->taskgroup;
char *arg;
bool do_wake;
size_t depend_size = 0;
if (flags & GOMP_TASK_FLAG_DEPEND)
depend_size = ((uintptr_t) (depend[0] ? depend[0] : depend[1])
* sizeof (struct gomp_task_depend_entry));
task = gomp_malloc (sizeof (*task) + depend_size
+ arg_size + arg_align - 1);
arg = (char *) (((uintptr_t) (task + 1) + depend_size + arg_align - 1)
& ~(uintptr_t) (arg_align - 1));
gomp_init_task (task, parent, gomp_icv (false));
task->priority = priority;
task->kind = GOMP_TASK_UNDEFERRED;
task->in_tied_task = parent->in_tied_task;
task->taskgroup = taskgroup;
thr->task = task;
if (cpyfn)
{
cpyfn (arg, data);
task->copy_ctors_done = true;
}
else
memcpy (arg, data, arg_size);
thr->task = parent;
task->kind = GOMP_TASK_WAITING;
task->fn = fn;
task->fn_data = arg;
task->final_task = (flags & GOMP_TASK_FLAG_FINAL) >> 1;
gomp_mutex_lock (&team->task_lock);
/* If parallel or taskgroup has been cancelled, don't start new
tasks. */
if (__builtin_expect (gomp_cancel_var, 0)
&& !task->copy_ctors_done)
{
if (gomp_team_barrier_cancelled (&team->barrier))
{
do_cancel:
gomp_mutex_unlock (&team->task_lock);
gomp_finish_task (task);
free (task);
return;
}
if (taskgroup)
{
if (taskgroup->cancelled)
goto do_cancel;
if (taskgroup->workshare
&& taskgroup->prev
&& taskgroup->prev->cancelled)
goto do_cancel;
}
}
if (taskgroup)
taskgroup->num_children++;
if (depend_size)
{
gomp_task_handle_depend (task, parent, depend);
if (task->num_dependees)
{
/* Tasks that depend on other tasks are not put into the
various waiting queues, so we are done for now. Said
tasks are instead put into the queues via
gomp_task_run_post_handle_dependers() after their
dependencies have been satisfied. After which, they
can be picked up by the various scheduling
points. */
gomp_mutex_unlock (&team->task_lock);
return;
}
}
priority_queue_insert (PQ_CHILDREN, &parent->children_queue,
task, priority,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup)
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task, priority,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
priority_queue_insert (PQ_TEAM, &team->task_queue,
task, priority,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
++team->task_count;
++team->task_queued_count;
gomp_team_barrier_set_task_pending (&team->barrier);
do_wake = team->task_running_count + !parent->in_tied_task
< team->nthreads;
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
gomp_team_barrier_wake (&team->barrier, 1);
}
}
ialias (GOMP_taskgroup_start)
ialias (GOMP_taskgroup_end)
ialias (GOMP_taskgroup_reduction_register)
#define TYPE long
#define UTYPE unsigned long
#define TYPE_is_long 1
#include "taskloop.c"
#undef TYPE
#undef UTYPE
#undef TYPE_is_long
#define TYPE unsigned long long
#define UTYPE TYPE
#define GOMP_taskloop GOMP_taskloop_ull
#include "taskloop.c"
#undef TYPE
#undef UTYPE
#undef GOMP_taskloop
static void inline
priority_queue_move_task_first (enum priority_queue_type type,
struct priority_queue *head,
struct gomp_task *task)
{
#if _LIBGOMP_CHECKING_
if (!priority_queue_task_in_queue_p (type, head, task))
gomp_fatal ("Attempt to move first missing task %p", task);
#endif
struct priority_list *list;
if (priority_queue_multi_p (head))
{
list = priority_queue_lookup_priority (head, task->priority);
#if _LIBGOMP_CHECKING_
if (!list)
gomp_fatal ("Unable to find priority %d", task->priority);
#endif
}
else
list = &head->l;
priority_list_remove (list, task_to_priority_node (type, task), 0);
priority_list_insert (type, list, task, task->priority,
PRIORITY_INSERT_BEGIN, type == PQ_CHILDREN,
task->parent_depends_on);
}
/* Actual body of GOMP_PLUGIN_target_task_completion that is executed
with team->task_lock held, or is executed in the thread that called
gomp_target_task_fn if GOMP_PLUGIN_target_task_completion has been
run before it acquires team->task_lock. */
static void
gomp_target_task_completion (struct gomp_team *team, struct gomp_task *task)
{
struct gomp_task *parent = task->parent;
if (parent)
priority_queue_move_task_first (PQ_CHILDREN, &parent->children_queue,
task);
struct gomp_taskgroup *taskgroup = task->taskgroup;
if (taskgroup)
priority_queue_move_task_first (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task);
priority_queue_insert (PQ_TEAM, &team->task_queue, task, task->priority,
PRIORITY_INSERT_BEGIN, false,
task->parent_depends_on);
task->kind = GOMP_TASK_WAITING;
if (parent && parent->taskwait)
{
if (parent->taskwait->in_taskwait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
parent->taskwait->in_taskwait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
else if (parent->taskwait->in_depend_wait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
parent->taskwait->in_depend_wait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
}
if (taskgroup && taskgroup->in_taskgroup_wait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
taskgroup->in_taskgroup_wait = false;
gomp_sem_post (&taskgroup->taskgroup_sem);
}
++team->task_queued_count;
gomp_team_barrier_set_task_pending (&team->barrier);
/* I'm afraid this can't be done after releasing team->task_lock,
as gomp_target_task_completion is run from unrelated thread and
therefore in between gomp_mutex_unlock and gomp_team_barrier_wake
the team could be gone already. */
if (team->nthreads > team->task_running_count)
gomp_team_barrier_wake (&team->barrier, 1);
}
/* Signal that a target task TTASK has completed the asynchronously
running phase and should be requeued as a task to handle the
variable unmapping. */
void
GOMP_PLUGIN_target_task_completion (void *data)
{
struct gomp_target_task *ttask = (struct gomp_target_task *) data;
struct gomp_task *task = ttask->task;
struct gomp_team *team = ttask->team;
gomp_mutex_lock (&team->task_lock);
if (ttask->state == GOMP_TARGET_TASK_READY_TO_RUN)
{
ttask->state = GOMP_TARGET_TASK_FINISHED;
gomp_mutex_unlock (&team->task_lock);
return;
}
ttask->state = GOMP_TARGET_TASK_FINISHED;
gomp_target_task_completion (team, task);
gomp_mutex_unlock (&team->task_lock);
}
static void gomp_task_run_post_handle_depend_hash (struct gomp_task *);
/* Called for nowait target tasks. */
bool
gomp_create_target_task (struct gomp_device_descr *devicep,
void (*fn) (void *), size_t mapnum, void **hostaddrs,
size_t *sizes, unsigned short *kinds,
unsigned int flags, void **depend, void **args,
enum gomp_target_task_state state)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (__builtin_expect (gomp_cancel_var, 0) && team)
{
if (gomp_team_barrier_cancelled (&team->barrier))
return true;
if (thr->task->taskgroup)
{
if (thr->task->taskgroup->cancelled)
return true;
if (thr->task->taskgroup->workshare
&& thr->task->taskgroup->prev
&& thr->task->taskgroup->prev->cancelled)
return true;
}
}
struct gomp_target_task *ttask;
struct gomp_task *task;
struct gomp_task *parent = thr->task;
struct gomp_taskgroup *taskgroup = parent->taskgroup;
bool do_wake;
size_t depend_size = 0;
uintptr_t depend_cnt = 0;
size_t tgt_align = 0, tgt_size = 0;
if (depend != NULL)
{
depend_cnt = (uintptr_t) (depend[0] ? depend[0] : depend[1]);
depend_size = depend_cnt * sizeof (struct gomp_task_depend_entry);
}
if (fn)
{
/* GOMP_MAP_FIRSTPRIVATE need to be copied first, as they are
firstprivate on the target task. */
size_t i;
for (i = 0; i < mapnum; i++)
if ((kinds[i] & 0xff) == GOMP_MAP_FIRSTPRIVATE)
{
size_t align = (size_t) 1 << (kinds[i] >> 8);
if (tgt_align < align)
tgt_align = align;
tgt_size = (tgt_size + align - 1) & ~(align - 1);
tgt_size += sizes[i];
}
if (tgt_align)
tgt_size += tgt_align - 1;
else
tgt_size = 0;
}
task = gomp_malloc (sizeof (*task) + depend_size
+ sizeof (*ttask)
+ mapnum * (sizeof (void *) + sizeof (size_t)
+ sizeof (unsigned short))
+ tgt_size);
gomp_init_task (task, parent, gomp_icv (false));
task->priority = 0;
task->kind = GOMP_TASK_WAITING;
task->in_tied_task = parent->in_tied_task;
task->taskgroup = taskgroup;
ttask = (struct gomp_target_task *) &task->depend[depend_cnt];
ttask->devicep = devicep;
ttask->fn = fn;
ttask->mapnum = mapnum;
ttask->args = args;
memcpy (ttask->hostaddrs, hostaddrs, mapnum * sizeof (void *));
ttask->sizes = (size_t *) &ttask->hostaddrs[mapnum];
memcpy (ttask->sizes, sizes, mapnum * sizeof (size_t));
ttask->kinds = (unsigned short *) &ttask->sizes[mapnum];
memcpy (ttask->kinds, kinds, mapnum * sizeof (unsigned short));
if (tgt_align)
{
char *tgt = (char *) &ttask->kinds[mapnum];
size_t i;
uintptr_t al = (uintptr_t) tgt & (tgt_align - 1);
if (al)
tgt += tgt_align - al;
tgt_size = 0;
for (i = 0; i < mapnum; i++)
if ((kinds[i] & 0xff) == GOMP_MAP_FIRSTPRIVATE)
{
size_t align = (size_t) 1 << (kinds[i] >> 8);
tgt_size = (tgt_size + align - 1) & ~(align - 1);
memcpy (tgt + tgt_size, hostaddrs[i], sizes[i]);
ttask->hostaddrs[i] = tgt + tgt_size;
tgt_size = tgt_size + sizes[i];
}
}
ttask->flags = flags;
ttask->state = state;
ttask->task = task;
ttask->team = team;
task->fn = NULL;
task->fn_data = ttask;
task->final_task = 0;
gomp_mutex_lock (&team->task_lock);
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (__builtin_expect (gomp_cancel_var, 0))
{
if (gomp_team_barrier_cancelled (&team->barrier))
{
do_cancel:
gomp_mutex_unlock (&team->task_lock);
gomp_finish_task (task);
free (task);
return true;
}
if (taskgroup)
{
if (taskgroup->cancelled)
goto do_cancel;
if (taskgroup->workshare
&& taskgroup->prev
&& taskgroup->prev->cancelled)
goto do_cancel;
}
}
if (depend_size)
{
gomp_task_handle_depend (task, parent, depend);
if (task->num_dependees)
{
if (taskgroup)
taskgroup->num_children++;
gomp_mutex_unlock (&team->task_lock);
return true;
}
}
if (state == GOMP_TARGET_TASK_DATA)
{
gomp_task_run_post_handle_depend_hash (task);
gomp_mutex_unlock (&team->task_lock);
gomp_finish_task (task);
free (task);
return false;
}
if (taskgroup)
taskgroup->num_children++;
/* For async offloading, if we don't need to wait for dependencies,
run the gomp_target_task_fn right away, essentially schedule the
mapping part of the task in the current thread. */
if (devicep != NULL
&& (devicep->capabilities & GOMP_OFFLOAD_CAP_OPENMP_400))
{
priority_queue_insert (PQ_CHILDREN, &parent->children_queue, task, 0,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup)
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task, 0, PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
task->pnode[PQ_TEAM].next = NULL;
task->pnode[PQ_TEAM].prev = NULL;
task->kind = GOMP_TASK_TIED;
++team->task_count;
gomp_mutex_unlock (&team->task_lock);
thr->task = task;
gomp_target_task_fn (task->fn_data);
thr->task = parent;
gomp_mutex_lock (&team->task_lock);
task->kind = GOMP_TASK_ASYNC_RUNNING;
/* If GOMP_PLUGIN_target_task_completion has run already
in between gomp_target_task_fn and the mutex lock,
perform the requeuing here. */
if (ttask->state == GOMP_TARGET_TASK_FINISHED)
gomp_target_task_completion (team, task);
else
ttask->state = GOMP_TARGET_TASK_RUNNING;
gomp_mutex_unlock (&team->task_lock);
return true;
}
priority_queue_insert (PQ_CHILDREN, &parent->children_queue, task, 0,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup)
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue, task, 0,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
priority_queue_insert (PQ_TEAM, &team->task_queue, task, 0,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
++team->task_count;
++team->task_queued_count;
gomp_team_barrier_set_task_pending (&team->barrier);
do_wake = team->task_running_count + !parent->in_tied_task
< team->nthreads;
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
gomp_team_barrier_wake (&team->barrier, 1);
return true;
}
/* Given a parent_depends_on task in LIST, move it to the front of its
priority so it is run as soon as possible.
Care is taken to update the list's LAST_PARENT_DEPENDS_ON field.
We rearrange the queue such that all parent_depends_on tasks are
first, and last_parent_depends_on points to the last such task we
rearranged. For example, given the following tasks in a queue
where PD[123] are the parent_depends_on tasks:
task->children
|
V
C1 -> C2 -> C3 -> PD1 -> PD2 -> PD3 -> C4
We rearrange such that:
task->children
| +--- last_parent_depends_on
| |
V V
PD1 -> PD2 -> PD3 -> C1 -> C2 -> C3 -> C4. */
static void inline
priority_list_upgrade_task (struct priority_list *list,
struct priority_node *node)
{
struct priority_node *last_parent_depends_on
= list->last_parent_depends_on;
if (last_parent_depends_on)
{
node->prev->next = node->next;
node->next->prev = node->prev;
node->prev = last_parent_depends_on;
node->next = last_parent_depends_on->next;
node->prev->next = node;
node->next->prev = node;
}
else if (node != list->tasks)
{
node->prev->next = node->next;
node->next->prev = node->prev;
node->prev = list->tasks->prev;
node->next = list->tasks;
list->tasks = node;
node->prev->next = node;
node->next->prev = node;
}
list->last_parent_depends_on = node;
}
/* Given a parent_depends_on TASK in its parent's children_queue, move
it to the front of its priority so it is run as soon as possible.
PARENT is passed as an optimization.
(This function could be defined in priority_queue.c, but we want it
inlined, and putting it in priority_queue.h is not an option, given
that gomp_task has not been properly defined at that point). */
static void inline
priority_queue_upgrade_task (struct gomp_task *task,
struct gomp_task *parent)
{
struct priority_queue *head = &parent->children_queue;
struct priority_node *node = &task->pnode[PQ_CHILDREN];
#if _LIBGOMP_CHECKING_
if (!task->parent_depends_on)
gomp_fatal ("priority_queue_upgrade_task: task must be a "
"parent_depends_on task");
if (!priority_queue_task_in_queue_p (PQ_CHILDREN, head, task))
gomp_fatal ("priority_queue_upgrade_task: cannot find task=%p", task);
#endif
if (priority_queue_multi_p (head))
{
struct priority_list *list
= priority_queue_lookup_priority (head, task->priority);
priority_list_upgrade_task (list, node);
}
else
priority_list_upgrade_task (&head->l, node);
}
/* Given a CHILD_TASK in LIST that is about to be executed, move it out of
the way in LIST so that other tasks can be considered for
execution. LIST contains tasks of type TYPE.
Care is taken to update the queue's LAST_PARENT_DEPENDS_ON field
if applicable. */
static void inline
priority_list_downgrade_task (enum priority_queue_type type,
struct priority_list *list,
struct gomp_task *child_task)
{
struct priority_node *node = task_to_priority_node (type, child_task);
if (list->tasks == node)
list->tasks = node->next;
else if (node->next != list->tasks)
{
/* The task in NODE is about to become TIED and TIED tasks
cannot come before WAITING tasks. If we're about to
leave the queue in such an indeterminate state, rewire
things appropriately. However, a TIED task at the end is
perfectly fine. */
struct gomp_task *next_task = priority_node_to_task (type, node->next);
if (next_task->kind == GOMP_TASK_WAITING)
{
/* Remove from list. */
node->prev->next = node->next;
node->next->prev = node->prev;
/* Rewire at the end. */
node->next = list->tasks;
node->prev = list->tasks->prev;
list->tasks->prev->next = node;
list->tasks->prev = node;