-
Notifications
You must be signed in to change notification settings - Fork 0
/
parallel.c
302 lines (264 loc) · 8.24 KB
/
parallel.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
/* 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 handles the (bare) PARALLEL construct. */
#include "libgomp.h"
#include <limits.h>
/* Determine the number of threads to be launched for a PARALLEL construct.
This algorithm is explicitly described in OpenMP 3.0 section 2.4.1.
SPECIFIED is a combination of the NUM_THREADS clause and the IF clause.
If the IF clause is false, SPECIFIED is forced to 1. When NUM_THREADS
is not present, SPECIFIED is 0. */
unsigned
gomp_resolve_num_threads (unsigned specified, unsigned count)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task_icv *icv;
unsigned threads_requested, max_num_threads, num_threads;
unsigned long busy;
struct gomp_thread_pool *pool;
icv = gomp_icv (false);
if (specified == 1)
return 1;
else if (thr->ts.active_level >= 1 && !icv->nest_var)
return 1;
else if (thr->ts.active_level >= gomp_max_active_levels_var)
return 1;
/* If NUM_THREADS not specified, use nthreads_var. */
if (specified == 0)
threads_requested = icv->nthreads_var;
else
threads_requested = specified;
max_num_threads = threads_requested;
/* If dynamic threads are enabled, bound the number of threads
that we launch. */
if (icv->dyn_var)
{
unsigned dyn = gomp_dynamic_max_threads ();
if (dyn < max_num_threads)
max_num_threads = dyn;
/* Optimization for parallel sections. */
if (count && count < max_num_threads)
max_num_threads = count;
}
/* UINT_MAX stands for infinity. */
if (__builtin_expect (icv->thread_limit_var == UINT_MAX, 1)
|| max_num_threads == 1)
return max_num_threads;
/* The threads_busy counter lives in thread_pool, if there
isn't a thread_pool yet, there must be just one thread
in the contention group. If thr->team is NULL, this isn't
nested parallel, so there is just one thread in the
contention group as well, no need to handle it atomically. */
pool = thr->thread_pool;
if (thr->ts.team == NULL || pool == NULL)
{
num_threads = max_num_threads;
if (num_threads > icv->thread_limit_var)
num_threads = icv->thread_limit_var;
if (pool)
pool->threads_busy = num_threads;
return num_threads;
}
#ifdef HAVE_SYNC_BUILTINS
do
{
busy = pool->threads_busy;
num_threads = max_num_threads;
if (icv->thread_limit_var - busy + 1 < num_threads)
num_threads = icv->thread_limit_var - busy + 1;
}
while (__sync_val_compare_and_swap (&pool->threads_busy,
busy, busy + num_threads - 1)
!= busy);
#else
gomp_mutex_lock (&gomp_managed_threads_lock);
num_threads = max_num_threads;
busy = pool->threads_busy;
if (icv->thread_limit_var - busy + 1 < num_threads)
num_threads = icv->thread_limit_var - busy + 1;
pool->threads_busy += num_threads - 1;
gomp_mutex_unlock (&gomp_managed_threads_lock);
#endif
return num_threads;
}
void
GOMP_parallel_start (void (*fn) (void *), void *data, unsigned num_threads)
{
num_threads = gomp_resolve_num_threads (num_threads, 0);
gomp_team_start (fn, data, num_threads, 0, gomp_new_team (num_threads));
}
void
GOMP_parallel_end (void)
{
struct gomp_task_icv *icv = gomp_icv (false);
if (__builtin_expect (icv->thread_limit_var != UINT_MAX, 0))
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
unsigned int nthreads = team ? team->nthreads : 1;
gomp_team_end ();
if (nthreads > 1)
{
/* If not nested, there is just one thread in the
contention group left, no need for atomicity. */
if (thr->ts.team == NULL)
thr->thread_pool->threads_busy = 1;
else
{
#ifdef HAVE_SYNC_BUILTINS
__sync_fetch_and_add (&thr->thread_pool->threads_busy,
1UL - nthreads);
#else
gomp_mutex_lock (&gomp_managed_threads_lock);
thr->thread_pool->threads_busy -= nthreads - 1;
gomp_mutex_unlock (&gomp_managed_threads_lock);
#endif
}
}
}
else
gomp_team_end ();
}
ialias (GOMP_parallel_end)
void
GOMP_parallel (void (*fn) (void *), void *data, unsigned num_threads, unsigned int flags)
{
num_threads = gomp_resolve_num_threads (num_threads, 0);
gomp_team_start (fn, data, num_threads, flags, gomp_new_team (num_threads));
fn (data);
ialias_call (GOMP_parallel_end) ();
}
bool
GOMP_cancellation_point (int which)
{
if (!gomp_cancel_var)
return false;
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
if (which & (GOMP_CANCEL_LOOP | GOMP_CANCEL_SECTIONS))
{
if (team == NULL)
return false;
return team->work_share_cancelled != 0;
}
else if (which & GOMP_CANCEL_TASKGROUP)
{
if (thr->task->taskgroup && thr->task->taskgroup->cancelled)
return true;
/* FALLTHRU into the GOMP_CANCEL_PARALLEL case,
as #pragma omp cancel parallel also cancels all explicit
tasks. */
}
if (team)
return gomp_team_barrier_cancelled (&team->barrier);
return false;
}
ialias (GOMP_cancellation_point)
bool
GOMP_cancel (int which, bool do_cancel)
{
if (!gomp_cancel_var)
return false;
if (!do_cancel)
return ialias_call (GOMP_cancellation_point) (which);
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
if (which & (GOMP_CANCEL_LOOP | GOMP_CANCEL_SECTIONS))
{
/* In orphaned worksharing region, all we want to cancel
is current thread. */
if (team != NULL)
team->work_share_cancelled = 1;
return true;
}
else if (which & GOMP_CANCEL_TASKGROUP)
{
if (thr->task->taskgroup && !thr->task->taskgroup->cancelled)
{
gomp_mutex_lock (&team->task_lock);
thr->task->taskgroup->cancelled = true;
gomp_mutex_unlock (&team->task_lock);
}
return true;
}
team->team_cancelled = 1;
gomp_team_barrier_cancel (team);
return true;
}
/* The public OpenMP API for thread and team related inquiries. */
int
omp_get_num_threads (void)
{
struct gomp_team *team = gomp_thread ()->ts.team;
return team ? team->nthreads : 1;
}
int
omp_get_thread_num (void)
{
return gomp_thread ()->ts.team_id;
}
/* This wasn't right for OpenMP 2.5. Active region used to be non-zero
when the IF clause doesn't evaluate to false, starting with OpenMP 3.0
it is non-zero with more than one thread in the team. */
int
omp_in_parallel (void)
{
return gomp_thread ()->ts.active_level > 0;
}
int
omp_get_level (void)
{
return gomp_thread ()->ts.level;
}
int
omp_get_ancestor_thread_num (int level)
{
struct gomp_team_state *ts = &gomp_thread ()->ts;
if (level < 0 || level > ts->level)
return -1;
for (level = ts->level - level; level > 0; --level)
ts = &ts->team->prev_ts;
return ts->team_id;
}
int
omp_get_team_size (int level)
{
struct gomp_team_state *ts = &gomp_thread ()->ts;
if (level < 0 || level > ts->level)
return -1;
for (level = ts->level - level; level > 0; --level)
ts = &ts->team->prev_ts;
if (ts->team == NULL)
return 1;
else
return ts->team->nthreads;
}
int
omp_get_active_level (void)
{
return gomp_thread ()->ts.active_level;
}
ialias (omp_get_num_threads)
ialias (omp_get_thread_num)
ialias (omp_in_parallel)
ialias (omp_get_level)
ialias (omp_get_ancestor_thread_num)
ialias (omp_get_team_size)
ialias (omp_get_active_level)