-
Notifications
You must be signed in to change notification settings - Fork 0
/
libgomp.h
1679 lines (1411 loc) · 53.1 KB
/
libgomp.h
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) 2005-2017 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 contains data types and function declarations that are not
part of the official OpenACC or OpenMP user interfaces. There are
declarations in here that are part of the GNU Offloading and Multi
Processing ABI, in that the compiler is required to know about them
and use them.
The convention is that the all caps prefix "GOMP" is used group items
that are part of the external ABI, and the lower case prefix "gomp"
is used group items that are completely private to the library. */
#ifndef LIBGOMP_H
#define LIBGOMP_H 1
#ifndef _LIBGOMP_CHECKING_
/* Define to 1 to perform internal sanity checks. */
#define _LIBGOMP_CHECKING_ 0
#endif
#ifndef _LIBGOMP_TEAM_TIMING_
/* Define to 1 to measure time to complete parallel region. */
#define _LIBGOMP_TEAM_TIMING_ 1
#endif
#ifndef _LIBGOMP_TASK_TIMING_
/* Define to 1 to collect tasks workaround time. */
#define _LIBGOMP_TASK_TIMING_ 1
#endif
#ifndef _LIBGOMP_TASK_GRANULARITY_
/* Define to 1 to measure tasks granularity at run-time. */
#define _LIBGOMP_TASK_GRANULARITY_ 0
#endif
#ifndef _LIBGOMP_TASK_SWITCH_AUDITING_
/* Define to 1 to monitor ULT-based task-switch occurrences. */
#define _LIBGOMP_TASK_SWITCH_AUDITING_ 1
/* Define to 1 to monitor ULT-based task-switches that occur
in place of the BASELINE function calls. */
#define _LIBGOMP_IN_FLOW_TASK_SWITCH_AUDITING_ 0
#endif
#ifndef _LIBGOMP_LIBGOMP_TIMING_
/* Define to 1 to monitor time spent executing libGOMP functions. */
#define _LIBGOMP_LIBGOMP_TIMING_ 1
#endif
#ifndef _LIBGOMP_TEAM_LOCK_TIMING_
/* Define to 1 to monitor time spent executing with team's lock acquired. */
#define _LIBGOMP_TEAM_LOCK_TIMING_ 1
#endif
#define DEFAULT_NUM_PRIORITIES 16
#include "config.h"
#include "gstdint.h"
#include "libgomp-plugin.h"
#ifdef HAVE_PTHREAD_H
#include <pthread.h>
#endif
#include <stdbool.h>
#include <stdlib.h>
#include <stdarg.h>
#include <string.h>
#if _LIBGOMP_TEAM_TIMING_ || _LIBGOMP_TASK_TIMING_ || _LIBGOMP_TASK_GRANULARITY_ || \
_LIBGOMP_TASK_SWITCH_AUDITING_ || _LIBGOMP_LIBGOMP_TIMING_ || _LIBGOMP_TEAM_LOCK_TIMING_
#include <stdio.h>
#include <time.h>
#endif
/* Needed for memset in priority_queue.c. */
#if _LIBGOMP_CHECKING_
# ifdef STRING_WITH_STRINGS
# include <string.h>
# include <strings.h>
# else
# ifdef HAVE_STRING_H
# include <string.h>
# else
# ifdef HAVE_STRINGS_H
# include <strings.h>
# endif
# endif
# endif
#endif
#ifdef HAVE_ATTRIBUTE_VISIBILITY
# pragma GCC visibility push(hidden)
#endif
/* If we were a C++ library, we'd get this from <std/atomic>. */
enum memmodel
{
MEMMODEL_RELAXED = 0,
MEMMODEL_CONSUME = 1,
MEMMODEL_ACQUIRE = 2,
MEMMODEL_RELEASE = 3,
MEMMODEL_ACQ_REL = 4,
MEMMODEL_SEQ_CST = 5
};
/* alloc.c */
extern void *gomp_malloc (size_t) __attribute__((malloc));
extern void *gomp_malloc_cleared (size_t) __attribute__((malloc));
extern void *gomp_aligned_alloc (size_t, size_t) __attribute__((malloc));
extern void *gomp_realloc (void *, size_t);
/* Avoid conflicting prototypes of alloca() in system headers by using
GCC's builtin alloca(). */
#define gomp_alloca(x) __builtin_alloca(x)
/* error.c */
extern void gomp_vdebug (int, const char *, va_list);
extern void gomp_debug (int, const char *, ...)
__attribute__ ((format (printf, 2, 3)));
#define gomp_vdebug(KIND, FMT, VALIST) \
do { \
if (__builtin_expect (gomp_debug_var, 0)) \
(gomp_vdebug) ((KIND), (FMT), (VALIST)); \
} while (0)
#define gomp_debug(KIND, ...) \
do { \
if (__builtin_expect (gomp_debug_var, 0)) \
(gomp_debug) ((KIND), __VA_ARGS__); \
} while (0)
extern void gomp_verror (const char *, va_list);
extern void gomp_error (const char *, ...)
__attribute__ ((format (printf, 1, 2)));
extern void gomp_vfatal (const char *, va_list)
__attribute__ ((noreturn));
extern void gomp_fatal (const char *, ...)
__attribute__ ((noreturn, format (printf, 1, 2)));
enum gomp_task_type
{
/* Tied Task, as it has been labeled by the programmer. */
GOMP_TASK_TYPE_TIED,
/* Untied Task, as it has been labeled by the programmer. */
GOMP_TASK_TYPE_UNTIED,
/* The number of task types that are usable at run-time. */
GOMP_TASK_TYPE_ENUM_SIZE
};
enum gomp_task_kind
{
/* Implicit task. */
GOMP_TASK_IMPLICIT,
/* Undeferred task. */
GOMP_TASK_UNDEFERRED,
/* Task created by GOMP_task and waiting to be run. */
GOMP_TASK_WAITING,
/* Tied Task currently executing or scheduled and about to execute. */
GOMP_TASK_TIED,
/* Tied Task currently suspended on at most one thread's suspending-queue. */
GOMP_TASK_TIED_SUSPENDED,
/* Used for target tasks that have vars mapped and async run started,
but not yet completed. Once that completes, they will be readded
into the queues as GOMP_TASK_WAITING in order to perform the var
unmapping. */
GOMP_TASK_ASYNC_RUNNING,
/* The number of task kinds that are usable at run-time. */
GOMP_TASK_KIND_ENUM_SIZE
#if defined HAVE_TLS || defined USE_EMUTLS
, GOMP_TASK_KIND_ENUM_PAD = 0x7fffffff
#endif
};
struct gomp_thread;
struct gomp_task_icv;
struct gomp_task;
struct gomp_taskgroup;
struct htab;
#include "priority_queue.h"
#include "sem.h"
#include "mutex.h"
#include "bar.h"
#include "simple-bar.h"
#include "ptrlock.h"
#include "spinlock.h"
#include "context.h"
#include "state.h"
/* This structure contains the data to control one work-sharing construct,
either a LOOP (FOR/DO) or a SECTIONS. */
enum gomp_schedule_type
{
GFS_RUNTIME,
GFS_STATIC,
GFS_DYNAMIC,
GFS_GUIDED,
GFS_AUTO
};
struct gomp_doacross_work_share
{
union {
/* chunk_size copy, as ws->chunk_size is multiplied by incr for
GFS_DYNAMIC. */
long chunk_size;
/* Likewise, but for ull implementation. */
unsigned long long chunk_size_ull;
/* For schedule(static,0) this is the number
of iterations assigned to the last thread, i.e. number of
iterations / number of threads. */
long q;
/* Likewise, but for ull implementation. */
unsigned long long q_ull;
};
/* Size of each array entry (padded to cache line size). */
unsigned long elt_sz;
/* Number of dimensions in sink vectors. */
unsigned int ncounts;
/* True if the iterations can be flattened. */
bool flattened;
/* Actual array (of elt_sz sized units), aligned to cache line size.
This is indexed by team_id for GFS_STATIC and outermost iteration
/ chunk_size for other schedules. */
unsigned char *array;
/* These two are only used for schedule(static,0). */
/* This one is number of iterations % number of threads. */
long t;
union {
/* And this one is cached t * (q + 1). */
long boundary;
/* Likewise, but for the ull implementation. */
unsigned long long boundary_ull;
};
/* Array of shift counts for each dimension if they can be flattened. */
unsigned int shift_counts[];
};
struct gomp_work_share
{
/* This member records the SCHEDULE clause to be used for this construct.
The user specification of "runtime" will already have been resolved.
If this is a SECTIONS construct, this value will always be DYNAMIC. */
enum gomp_schedule_type sched;
int mode;
union {
struct {
/* This is the chunk_size argument to the SCHEDULE clause. */
long chunk_size;
/* This is the iteration end point. If this is a SECTIONS construct,
this is the number of contained sections. */
long end;
/* This is the iteration step. If this is a SECTIONS construct, this
is always 1. */
long incr;
};
struct {
/* The same as above, but for the unsigned long long loop variants. */
unsigned long long chunk_size_ull;
unsigned long long end_ull;
unsigned long long incr_ull;
};
};
union {
/* This is a circular queue that details which threads will be allowed
into the ordered region and in which order. When a thread allocates
iterations on which it is going to work, it also registers itself at
the end of the array. When a thread reaches the ordered region, it
checks to see if it is the one at the head of the queue. If not, it
blocks on its RELEASE semaphore. */
unsigned *ordered_team_ids;
/* This is a pointer to DOACROSS work share data. */
struct gomp_doacross_work_share *doacross;
};
/* This is the number of threads that have registered themselves in
the circular queue ordered_team_ids. */
unsigned ordered_num_used;
/* This is the team_id of the currently acknowledged owner of the ordered
section, or -1u if the ordered section has not been acknowledged by
any thread. This is distinguished from the thread that is *allowed*
to take the section next. */
unsigned ordered_owner;
/* This is the index into the circular queue ordered_team_ids of the
current thread that's allowed into the ordered reason. */
unsigned ordered_cur;
/* This is a chain of allocated gomp_work_share blocks, valid only
in the first gomp_work_share struct in the block. */
struct gomp_work_share *next_alloc;
/* The above fields are written once during workshare initialization,
or related to ordered worksharing. Make sure the following fields
are in a different cache line. */
/* This lock protects the update of the following members. */
gomp_mutex_t lock __attribute__((aligned (64)));
/* This is the count of the number of threads that have exited the work
share construct. If the construct was marked nowait, they have moved on
to other work; otherwise they're blocked on a barrier. The last member
of the team to exit the work share construct must deallocate it. */
unsigned threads_completed;
union {
/* This is the next iteration value to be allocated. In the case of
GFS_STATIC loops, this the iteration start point and never changes. */
long next;
/* The same, but with unsigned long long type. */
unsigned long long next_ull;
/* This is the returned data structure for SINGLE COPYPRIVATE. */
void *copyprivate;
};
union {
/* Link to gomp_work_share struct for next work sharing construct
encountered after this one. */
gomp_ptrlock_t next_ws;
/* gomp_work_share structs are chained in the free work share cache
through this. */
struct gomp_work_share *next_free;
};
/* If only few threads are in the team, ordered_team_ids can point
to this array which fills the padding at the end of this struct. */
unsigned inline_ordered_team_ids[0];
};
/* This structure contains all of the thread-local data associated with
a thread team. This is the data that must be saved when a thread
encounters a nested PARALLEL construct. */
struct gomp_team_state
{
/* This is the team of which the thread is currently a member.
DO NOT MOVE IT FROM THE FIRST POSITION! */
struct gomp_team *team;
/* This is the work share construct which this thread is currently
processing. Recall that with NOWAIT, not all threads may be
processing the same construct. */
struct gomp_work_share *work_share;
/* This is the previous work share construct or NULL if there wasn't any.
When all threads are done with the current work sharing construct,
the previous one can be freed. The current one can't, as its
next_ws field is used. */
struct gomp_work_share *last_work_share;
/* This is the ID of this thread within the team. This value is
guaranteed to be between 0 and N-1, where N is the number of
threads in the team. */
unsigned team_id;
#if defined HAVE_TLS || defined USE_EMUTLS
/* This is the ID of the CPU-core on which this thread has been
pinned to, and it's needed to retrieve the ID of the IBS device
where this thread is also registered. */
int core_id;
/* This variable points to the initial address of an alternate-stack
memory area which is required to safely perform control-flow-
variation upon the occurrence of a programmed interrupt. It represents
a memory mapped landing area that can never lead to page-fault. */
void * alt_stack;
/* It representes the amount of bytes that the alternate-stack memory
area must be composed of. */
unsigned long alt_stack_size;
/* This is the file descriptor associated to the IBS device once
it has been opened by the current thread. */
int ibs_fd;
/* This is the file descriptor associated to the IPI device once
it has been opened by the current thread. */
int ipi_fd;
#endif
/* Nesting level. */
unsigned level;
/* Active nesting level. Only active parallel regions are counted. */
unsigned active_level;
/* Place-partition-var, offset and length into gomp_places_list array. */
unsigned place_partition_off;
unsigned place_partition_len;
#ifdef HAVE_SYNC_BUILTINS
/* Number of single stmts encountered. */
unsigned long single_count;
#endif
/* For GFS_RUNTIME loops that resolved to GFS_STATIC, this is the
trip number through the loop. So first time a particular loop
is encountered this number is 0, the second time through the loop
is 1, etc. This is unused when the compiler knows in advance that
the loop is statically scheduled. */
unsigned long static_trip;
};
struct target_mem_desc;
/* These are the OpenMP 4.0 Internal Control Variables described in
section 2.3.1. Those described as having one copy per task are
stored within the structure; those described as having one copy
for the whole program are (naturally) global variables. */
struct gomp_task_icv
{
unsigned long nthreads_var;
enum gomp_schedule_type run_sched_var;
int run_sched_chunk_size;
int default_device_var;
unsigned int thread_limit_var;
bool dyn_var;
bool nest_var;
bool wf_sched_var;
bool untied_block_var;
bool ult_var;
unsigned long ult_stack_size;
#if defined HAVE_TLS || defined USE_EMUTLS
void *text_section_start;
void *text_section_end;
#endif
char bind_var;
/* Internal ICV. */
struct target_mem_desc *target_data;
};
#if defined HAVE_TLS || defined USE_EMUTLS
/* This structure helps to displace within relative structs by
maintaining the offset associated to the specific field. This
packed structure has N fields, each one aligned to 8 bytes. */
struct gomp_struct_fields_offset
{
/* "0" */
size_t in_thread_team_offset;
/* "1" */
size_t in_thread_preemptable_offset;
/* "2" */
size_t in_thread_task_offset;
/* "3" */
size_t in_task_kind_offset;
/* "4" */
size_t in_task_state_offset;
/* "5" */
size_t in_state_context_offset;
} __attribute__((packed,aligned(8)));
extern struct gomp_struct_fields_offset gomp_global_sfo;
#endif
extern struct gomp_task_icv gomp_global_icv;
#ifndef HAVE_SYNC_BUILTINS
extern gomp_mutex_t gomp_managed_threads_lock;
#endif
extern unsigned long gomp_max_active_levels_var;
extern bool gomp_cancel_var;
extern bool gomp_auto_cutoff_var;
extern bool gomp_signal_unblock;
extern bool gomp_ipi_var;
extern double gomp_ipi_decision_model;
extern unsigned long gomp_ipi_priority_gap;
extern unsigned long gomp_ipi_sending_cap;
extern unsigned long gomp_ibs_rate_var;
extern unsigned long gomp_queue_policy_var;
extern int gomp_max_task_priority_var;
extern unsigned long long gomp_spin_count_var, gomp_throttled_spin_count_var;
extern unsigned long gomp_available_cpus, gomp_managed_threads;
extern unsigned long *gomp_nthreads_var_list, gomp_nthreads_var_list_len;
extern char *gomp_bind_var_list;
extern unsigned long gomp_bind_var_list_len;
extern void **gomp_places_list;
extern unsigned long gomp_places_list_len;
extern unsigned int gomp_num_teams_var;
extern int gomp_debug_var;
extern int goacc_device_num;
extern char *goacc_device_type;
#if _LIBGOMP_TEAM_TIMING_
#include "team-timing.h"
#endif
#include "task-timing.h"
#if _LIBGOMP_TASK_GRANULARITY_
#include "task-granularity.h"
#endif
#if _LIBGOMP_LIBGOMP_TIMING_
#include "libgomp-timing.h"
#endif
#if _LIBGOMP_TEAM_LOCK_TIMING_
#include "team-lock-timing.h"
#endif
struct gomp_task_depend_entry
{
/* Address of dependency. */
void *addr;
struct gomp_task_depend_entry *next;
struct gomp_task_depend_entry *prev;
/* Task that provides the dependency in ADDR. */
struct gomp_task *task;
/* Depend entry is of type "IN". */
bool is_in;
bool redundant;
bool redundant_out;
};
struct gomp_dependers_vec
{
size_t n_elem;
size_t allocated;
struct gomp_task *elem[];
};
/* Used when in GOMP_taskwait or in gomp_task_maybe_wait_for_dependencies. */
struct gomp_taskwait
{
bool in_taskwait;
bool in_depend_wait;
/* Number of tasks we are waiting for. */
size_t n_depend;
gomp_sem_t taskwait_sem;
};
/* This structure describes a "task" to be run by a thread. */
struct gomp_task
{
/* Parent of this task. */
struct gomp_task *parent;
/* The task that was tied to the current thread and executed
after this tied task. Otherwise is NULL. */
struct gomp_task *next_tied_task;
/* The task that was tied to the current thread before start
executing this tied task. Otherwise is NULL. */
struct gomp_task *previous_tied_task;
/* The first tied task encountered in its ascending creation order. */
struct gomp_task *ascending_tied_task;
/* Points to the first task into an UNDEFERRED tasks group. */
struct gomp_task *undeferred_ancestor;
/* TIED children of this task. */
struct priority_queue tied_children_queue;
/* UNTIED children of this task. */
struct priority_queue untied_children_queue;
/* Taskgroup this task belongs in. */
struct gomp_taskgroup *taskgroup;
/* Tasks that depend on this task. */
struct gomp_dependers_vec *dependers;
struct htab *depend_hash;
struct gomp_taskwait *taskwait;
/* Number of items in DEPEND. */
size_t depend_count;
/* Number of tasks this task depends on. Once this counter reaches
0, we have no unsatisfied dependencies, and this task can be put
into the various queues to be scheduled. */
size_t num_dependees;
/* Priority of this task. */
int priority;
uint64_t creation_time;
uint64_t completion_time;
#if _LIBGOMP_TASK_GRANULARITY_
uint64_t stage_start;
uint64_t stage_end;
uint64_t sum_stages;
#endif
/* If it's not NULL, this variable points to the gomp_thread data
structure of the thread that has suspended this task and this
task is currently suspended into either the BLOCKED suspended
tasks queue or NON-BLOCKED suspended tasks queue. */
struct gomp_thread *suspending_thread;
/* The priority node for this task in each of the different queues.
We put this here to avoid allocating space for each priority
node. Then we play offsetof() games to convert between pnode[]
entries and the gomp_task in which they reside. */
struct priority_node pnode[9];
/* Points to a specific "gomp_task_state" data structure, in order to
allow this task executing within its own private execution-context
in a private alternate stack memory space. */
struct gomp_task_state *state;
struct gomp_task_icv icv;
void (*fn) (void *);
void *fn_data;
enum gomp_task_type type;
enum gomp_task_kind kind;
bool in_tied_task;
/* Indicates whether this task belongs to or must be inserted in a
blocked list due to Taskwait blocking conditions. */
bool is_blocked;
bool final_task;
bool copy_ctors_done;
/* Set for undeferred tasks with unsatisfied dependencies which
block further execution of their parent until the dependencies
are satisfied. */
bool parent_depends_on;
/* Dependencies provided and/or needed for this task. DEPEND_COUNT
is the number of items available. */
struct gomp_task_depend_entry depend[];
};
#if _LIBGOMP_TASK_SWITCH_AUDITING_
#include "task-switch-auditing.h"
#endif
/* This structure describes a single #pragma omp taskgroup. */
struct gomp_taskgroup
{
struct gomp_taskgroup *prev;
/* Queue of TIED tasks that belong in this taskgroup. */
struct priority_queue tied_taskgroup_queue;
/* Queue of UNTIED tasks that belong in this taskgroup. */
struct priority_queue untied_taskgroup_queue;
bool in_taskgroup_wait;
bool cancelled;
gomp_sem_t taskgroup_sem;
size_t num_children;
};
/* Various state of OpenMP async offloading tasks. */
enum gomp_target_task_state
{
GOMP_TARGET_TASK_DATA,
GOMP_TARGET_TASK_BEFORE_MAP,
GOMP_TARGET_TASK_FALLBACK,
GOMP_TARGET_TASK_READY_TO_RUN,
GOMP_TARGET_TASK_RUNNING,
GOMP_TARGET_TASK_FINISHED
};
/* This structure describes a target task. */
struct gomp_target_task
{
struct gomp_device_descr *devicep;
void (*fn) (void *);
size_t mapnum;
size_t *sizes;
unsigned short *kinds;
unsigned int flags;
enum gomp_target_task_state state;
struct target_mem_desc *tgt;
struct gomp_task *task;
struct gomp_team *team;
/* Device-specific target arguments. */
void **args;
void *hostaddrs[];
};
/* This structure describes a "team" of threads. These are the threads
that are spawned by a PARALLEL constructs, as well as the work sharing
constructs that the team encounters. */
struct gomp_team
{
/* This is the number of threads in the current team. */
unsigned nthreads;
/* This is number of gomp_work_share structs that have been allocated
as a block last time. */
unsigned work_share_chunk;
/* This is the saved team state that applied to a master thread before
the current thread was created. */
struct gomp_team_state prev_ts;
/* This semaphore should be used by the master thread instead of its
"native" semaphore in the thread structure. Required for nested
parallels, as the master is a member of two teams. */
gomp_sem_t master_release;
/* This points to an array with pointers to the release semaphore
of the threads in the team. */
gomp_sem_t **ordered_release;
#if _LIBGOMP_TEAM_TIMING_
struct gomp_team_time team_time;
#endif
struct gomp_task_time_table **prio_task_time;
#if _LIBGOMP_TASK_GRANULARITY_
struct gomp_task_granularity_table **task_granularity_table;
#endif
#if _LIBGOMP_TASK_SWITCH_AUDITING_
struct gomp_task_switch_audit **task_switch_audit;
#endif
#if _LIBGOMP_LIBGOMP_TIMING_
struct gomp_libgomp_time **libgomp_time;
#endif
#if _LIBGOMP_TEAM_LOCK_TIMING_
struct gomp_team_lock_time **team_lock_time;
#endif
/* List of work shares on which gomp_fini_work_share hasn't been
called yet. If the team hasn't been cancelled, this should be
equal to each thr->ts.work_share, but otherwise it can be a possibly
long list of workshares. */
struct gomp_work_share *work_shares_to_free;
/* List of gomp_work_share structs chained through next_free fields.
This is populated and taken off only by the first thread in the
team encountering a new work sharing construct, in a critical
section. */
struct gomp_work_share *work_share_list_alloc;
/* List of gomp_work_share structs freed by free_work_share. New
entries are atomically added to the start of the list, and
alloc_work_share can safely only move all but the first entry
to work_share_list alloc, as free_work_share can happen concurrently
with alloc_work_share. */
struct gomp_work_share *work_share_list_free;
#ifdef HAVE_SYNC_BUILTINS
/* Number of simple single regions encountered by threads in this
team. */
unsigned long single_count;
#else
/* Mutex protecting addition of workshares to work_share_list_free. */
gomp_mutex_t work_share_list_free_lock;
#endif
/* This barrier is used for most synchronization of the team. */
gomp_barrier_t barrier;
/* Initial work shares, to avoid allocating any gomp_work_share
structs in the common case. */
struct gomp_work_share work_shares[8];
gomp_mutex_t task_lock;
/* Scheduled TIED tasks. */
struct priority_queue tied_task_queue;
/* Scheduled UNTIED tasks. */
struct priority_queue untied_task_queue;
/* Number of all GOMP_TASK_{WAITING,TIED} tasks in the team. */
unsigned int task_count;
/* Number of GOMP_TASK_WAITING tasks currently waiting to be scheduled. */
unsigned int task_queued_count;
/* Number of GOMP_TASK_{WAITING,TIED} tasks currently running
directly in gomp_barrier_handle_tasks; tasks spawned
from e.g. GOMP_taskwait or GOMP_taskgroup_end don't count, even when
that is called from a task run from gomp_barrier_handle_tasks.
task_running_count should be always <= team->nthreads,
and if current task isn't in_tied_task, then it will be
even < team->nthreads. */
unsigned int task_running_count;
int work_share_cancelled;
int team_cancelled;
/* This structure maintains a global and a certain number of per-thread
list of available "gomp_task_state" structures for every team. Each
one represents an execution-context ready to be used by whatever task. */
struct gomp_task_state_list_group task_state_pool;
/* This array contains structures for implicit tasks. */
struct gomp_task implicit_task[];
};
/* This structure contains all data that is private to libgomp and is
allocated per thread. */
struct gomp_thread
{
/* This is the function that the thread should run upon launch. */
void (*fn) (void *data);
void *data;
/* This is the current team state for this thread. The ts.team member
is NULL only if the thread is idle. */
struct gomp_team_state ts;
/* This is the task that the thread is currently executing. */
struct gomp_task *task;
/* This is the last task that is tied to thread, if any. */
struct gomp_task *last_tied_task;
/* This points to an already extracted and cached task's state
data structure, so to allow fast retriving when a task is
ready to run and must be associated to a state. */
struct gomp_task_state *cached_state;
/* This queue maintains a reference to all those TIED tasks
suspended by this thread. This structure does not keep track
of the order in which the tasks are suspended since this
information can be retrieved by following the TIED tasks
chain that starts from the last_tied_task pointer. */
struct priority_queue tied_suspended;
/* This queue maintains a reference to all those UNTIED tasks
suspended by this thread that are not yet resumed by any
other thread in the team. */
struct priority_queue untied_suspended;
/* Indicate whether this thread is currently holding the
team's lock. It is used as a hint to unlock after a
task switch is occurred. */
bool hold_team_lock;
#if defined HAVE_TLS || defined USE_EMUTLS
/* Indicate whether this thread is currently executing any
function that belongs to the GOMP library. It is used to
filter out all those threads that potentially would receive
IPIs from concurrent ones but they don't relly need it in
that they are not busy in doing tasks for the application. */
bool in_libgomp;
#endif
/* Indicate whether this thread is currently set as non
preemptable. It is used to avoid eventual task-switch
when, for instance, the task is executing within a
criical section. */
unsigned int non_preemptable;
/* This semaphore is used for ordered loops. */
gomp_sem_t release;
/* Place this thread is bound to plus one, or zero if not bound
to any place. */
unsigned int place;
struct gomp_task_time_table *prio_task_time;
#if _LIBGOMP_TASK_GRANULARITY_
struct gomp_task_granularity_table *task_granularity_table;
#endif
#if _LIBGOMP_TASK_SWITCH_AUDITING_
struct gomp_task_switch_audit *task_switch_audit;
#endif
#if _LIBGOMP_LIBGOMP_TIMING_
struct gomp_libgomp_time *libgomp_time;
#endif
#if _LIBGOMP_TEAM_LOCK_TIMING_
struct gomp_team_lock_time *team_lock_time;
#endif
/* Points to a thread local list of available free "gomp_task_state"
structures for fast access, rather than pass from the global
structure maintaing the whole group of lists. */
struct gomp_task_state_list *local_task_state_list;
/* Points to the global group of per-thread lists of free
"gomp_task_state" structures. Having a local copy avoid further
reads in the team data structure, which fields are not likely
accessed in the most of the cases. */
struct gomp_task_state_list_group *global_task_state_group;
/* User pthread thread pool */
struct gomp_thread_pool *thread_pool;
};
struct gomp_thread_pool
{
/* This array manages threads spawned from the top level, which will
return to the idle loop once the current PARALLEL construct ends. */
struct gomp_thread **threads;
unsigned threads_size;
unsigned threads_used;
/* The last team is used for non-nested teams to delay their destruction to
make sure all the threads in the team move on to the pool's barrier before
the team's barrier is destroyed. */
struct gomp_team *last_team;
/* Number of threads running in this contention group. */
unsigned long threads_busy;
/* This barrier holds and releases threads waiting in thread pools. */
gomp_simple_barrier_t threads_dock;
};
enum gomp_cancel_kind
{
GOMP_CANCEL_PARALLEL = 1,
GOMP_CANCEL_LOOP = 2,
GOMP_CANCEL_FOR = GOMP_CANCEL_LOOP,
GOMP_CANCEL_DO = GOMP_CANCEL_LOOP,
GOMP_CANCEL_SECTIONS = 4,
GOMP_CANCEL_TASKGROUP = 8
};
/* ... and here is that TLS data. */
#if defined __nvptx__
extern struct gomp_thread *nvptx_thrs __attribute__((shared));
static inline struct gomp_thread *gomp_thread (void)
{
int tid;
asm ("mov.u32 %0, %%tid.y;" : "=r" (tid));
return nvptx_thrs + tid;
}
#elif defined HAVE_TLS || defined USE_EMUTLS
extern __thread struct gomp_thread *gomp_tls_ptr;
extern __thread struct gomp_thread gomp_tls_data;
static inline struct gomp_thread *gomp_thread (void)
{
return &gomp_tls_data;
}
#else
extern pthread_key_t gomp_tls_key;
static inline struct gomp_thread *gomp_thread (void)
{
return pthread_getspecific (gomp_tls_key);
}
#endif
extern struct gomp_task_icv *gomp_new_icv (void);
/* Here's how to access the current copy of the ICVs. */
static inline struct gomp_task_icv *gomp_icv (bool write)
{
struct gomp_task *task = gomp_thread ()->task;
if (task)
return &task->icv;
else if (write)
return gomp_new_icv ();
else
return &gomp_global_icv;
}
#ifdef LIBGOMP_USE_PTHREADS
/* The attributes to be used during thread creation. */
extern pthread_attr_t gomp_thread_attr;
extern pthread_key_t gomp_thread_destructor;
#endif
static inline __attribute__((always_inline)) unsigned int
gomp_count_1_bits(unsigned long long int i)
{
i = i - ((i >> 1) & 0x5555555555555555ULL);
i = (i & 0x3333333333333333ULL) + ((i >> 2) & 0x3333333333333333ULL);
i = (i + (i >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
i = i + (i >> 8);
i = i + (i >> 16);
i = i + (i >> 32);
return (unsigned int) (i & 0x7f);
}
#if defined HAVE_TLS || defined USE_EMUTLS
#ifndef RDTSC_32_LSB
#define RDTSC_32_LSB() ({ \
unsigned int cycles_low; \
asm volatile ( \
"RDTSC\n\t" \