1 |
735 |
jeremybenn |
/* Copyright (C) 2005, 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
|
2 |
|
|
Contributed by Richard Henderson <rth@redhat.com>.
|
3 |
|
|
|
4 |
|
|
This file is part of the GNU OpenMP Library (libgomp).
|
5 |
|
|
|
6 |
|
|
Libgomp is free software; you can redistribute it and/or modify it
|
7 |
|
|
under the terms of the GNU General Public License as published by
|
8 |
|
|
the Free Software Foundation; either version 3, or (at your option)
|
9 |
|
|
any later version.
|
10 |
|
|
|
11 |
|
|
Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
|
12 |
|
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
13 |
|
|
FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
14 |
|
|
more details.
|
15 |
|
|
|
16 |
|
|
Under Section 7 of GPL version 3, you are granted additional
|
17 |
|
|
permissions described in the GCC Runtime Library Exception, version
|
18 |
|
|
3.1, as published by the Free Software Foundation.
|
19 |
|
|
|
20 |
|
|
You should have received a copy of the GNU General Public License and
|
21 |
|
|
a copy of the GCC Runtime Library Exception along with this program;
|
22 |
|
|
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
23 |
|
|
<http://www.gnu.org/licenses/>. */
|
24 |
|
|
|
25 |
|
|
/* This file handles the (bare) PARALLEL construct. */
|
26 |
|
|
|
27 |
|
|
#include "libgomp.h"
|
28 |
|
|
#include <limits.h>
|
29 |
|
|
|
30 |
|
|
|
31 |
|
|
/* Determine the number of threads to be launched for a PARALLEL construct.
|
32 |
|
|
This algorithm is explicitly described in OpenMP 3.0 section 2.4.1.
|
33 |
|
|
SPECIFIED is a combination of the NUM_THREADS clause and the IF clause.
|
34 |
|
|
If the IF clause is false, SPECIFIED is forced to 1. When NUM_THREADS
|
35 |
|
|
is not present, SPECIFIED is 0. */
|
36 |
|
|
|
37 |
|
|
unsigned
|
38 |
|
|
gomp_resolve_num_threads (unsigned specified, unsigned count)
|
39 |
|
|
{
|
40 |
|
|
struct gomp_thread *thread = gomp_thread();
|
41 |
|
|
struct gomp_task_icv *icv;
|
42 |
|
|
unsigned threads_requested, max_num_threads, num_threads;
|
43 |
|
|
unsigned long remaining;
|
44 |
|
|
|
45 |
|
|
icv = gomp_icv (false);
|
46 |
|
|
|
47 |
|
|
if (specified == 1)
|
48 |
|
|
return 1;
|
49 |
|
|
else if (thread->ts.active_level >= 1 && !icv->nest_var)
|
50 |
|
|
return 1;
|
51 |
|
|
else if (thread->ts.active_level >= gomp_max_active_levels_var)
|
52 |
|
|
return 1;
|
53 |
|
|
|
54 |
|
|
/* If NUM_THREADS not specified, use nthreads_var. */
|
55 |
|
|
if (specified == 0)
|
56 |
|
|
threads_requested = icv->nthreads_var;
|
57 |
|
|
else
|
58 |
|
|
threads_requested = specified;
|
59 |
|
|
|
60 |
|
|
max_num_threads = threads_requested;
|
61 |
|
|
|
62 |
|
|
/* If dynamic threads are enabled, bound the number of threads
|
63 |
|
|
that we launch. */
|
64 |
|
|
if (icv->dyn_var)
|
65 |
|
|
{
|
66 |
|
|
unsigned dyn = gomp_dynamic_max_threads ();
|
67 |
|
|
if (dyn < max_num_threads)
|
68 |
|
|
max_num_threads = dyn;
|
69 |
|
|
|
70 |
|
|
/* Optimization for parallel sections. */
|
71 |
|
|
if (count && count < max_num_threads)
|
72 |
|
|
max_num_threads = count;
|
73 |
|
|
}
|
74 |
|
|
|
75 |
|
|
/* ULONG_MAX stands for infinity. */
|
76 |
|
|
if (__builtin_expect (gomp_thread_limit_var == ULONG_MAX, 1)
|
77 |
|
|
|| max_num_threads == 1)
|
78 |
|
|
return max_num_threads;
|
79 |
|
|
|
80 |
|
|
#ifdef HAVE_SYNC_BUILTINS
|
81 |
|
|
do
|
82 |
|
|
{
|
83 |
|
|
remaining = gomp_remaining_threads_count;
|
84 |
|
|
num_threads = max_num_threads;
|
85 |
|
|
if (num_threads > remaining)
|
86 |
|
|
num_threads = remaining + 1;
|
87 |
|
|
}
|
88 |
|
|
while (__sync_val_compare_and_swap (&gomp_remaining_threads_count,
|
89 |
|
|
remaining, remaining - num_threads + 1)
|
90 |
|
|
!= remaining);
|
91 |
|
|
#else
|
92 |
|
|
gomp_mutex_lock (&gomp_remaining_threads_lock);
|
93 |
|
|
num_threads = max_num_threads;
|
94 |
|
|
remaining = gomp_remaining_threads_count;
|
95 |
|
|
if (num_threads > remaining)
|
96 |
|
|
num_threads = remaining + 1;
|
97 |
|
|
gomp_remaining_threads_count -= num_threads - 1;
|
98 |
|
|
gomp_mutex_unlock (&gomp_remaining_threads_lock);
|
99 |
|
|
#endif
|
100 |
|
|
|
101 |
|
|
return num_threads;
|
102 |
|
|
}
|
103 |
|
|
|
104 |
|
|
void
|
105 |
|
|
GOMP_parallel_start (void (*fn) (void *), void *data, unsigned num_threads)
|
106 |
|
|
{
|
107 |
|
|
num_threads = gomp_resolve_num_threads (num_threads, 0);
|
108 |
|
|
gomp_team_start (fn, data, num_threads, gomp_new_team (num_threads));
|
109 |
|
|
}
|
110 |
|
|
|
111 |
|
|
void
|
112 |
|
|
GOMP_parallel_end (void)
|
113 |
|
|
{
|
114 |
|
|
if (__builtin_expect (gomp_thread_limit_var != ULONG_MAX, 0))
|
115 |
|
|
{
|
116 |
|
|
struct gomp_thread *thr = gomp_thread ();
|
117 |
|
|
struct gomp_team *team = thr->ts.team;
|
118 |
|
|
if (team && team->nthreads > 1)
|
119 |
|
|
{
|
120 |
|
|
#ifdef HAVE_SYNC_BUILTINS
|
121 |
|
|
__sync_fetch_and_add (&gomp_remaining_threads_count,
|
122 |
|
|
1UL - team->nthreads);
|
123 |
|
|
#else
|
124 |
|
|
gomp_mutex_lock (&gomp_remaining_threads_lock);
|
125 |
|
|
gomp_remaining_threads_count -= team->nthreads - 1;
|
126 |
|
|
gomp_mutex_unlock (&gomp_remaining_threads_lock);
|
127 |
|
|
#endif
|
128 |
|
|
}
|
129 |
|
|
}
|
130 |
|
|
gomp_team_end ();
|
131 |
|
|
}
|
132 |
|
|
|
133 |
|
|
|
134 |
|
|
/* The public OpenMP API for thread and team related inquiries. */
|
135 |
|
|
|
136 |
|
|
int
|
137 |
|
|
omp_get_num_threads (void)
|
138 |
|
|
{
|
139 |
|
|
struct gomp_team *team = gomp_thread ()->ts.team;
|
140 |
|
|
return team ? team->nthreads : 1;
|
141 |
|
|
}
|
142 |
|
|
|
143 |
|
|
int
|
144 |
|
|
omp_get_thread_num (void)
|
145 |
|
|
{
|
146 |
|
|
return gomp_thread ()->ts.team_id;
|
147 |
|
|
}
|
148 |
|
|
|
149 |
|
|
/* This wasn't right for OpenMP 2.5. Active region used to be non-zero
|
150 |
|
|
when the IF clause doesn't evaluate to false, starting with OpenMP 3.0
|
151 |
|
|
it is non-zero with more than one thread in the team. */
|
152 |
|
|
|
153 |
|
|
int
|
154 |
|
|
omp_in_parallel (void)
|
155 |
|
|
{
|
156 |
|
|
return gomp_thread ()->ts.active_level > 0;
|
157 |
|
|
}
|
158 |
|
|
|
159 |
|
|
int
|
160 |
|
|
omp_get_level (void)
|
161 |
|
|
{
|
162 |
|
|
return gomp_thread ()->ts.level;
|
163 |
|
|
}
|
164 |
|
|
|
165 |
|
|
int
|
166 |
|
|
omp_get_ancestor_thread_num (int level)
|
167 |
|
|
{
|
168 |
|
|
struct gomp_team_state *ts = &gomp_thread ()->ts;
|
169 |
|
|
if (level < 0 || level > ts->level)
|
170 |
|
|
return -1;
|
171 |
|
|
for (level = ts->level - level; level > 0; --level)
|
172 |
|
|
ts = &ts->team->prev_ts;
|
173 |
|
|
return ts->team_id;
|
174 |
|
|
}
|
175 |
|
|
|
176 |
|
|
int
|
177 |
|
|
omp_get_team_size (int level)
|
178 |
|
|
{
|
179 |
|
|
struct gomp_team_state *ts = &gomp_thread ()->ts;
|
180 |
|
|
if (level < 0 || level > ts->level)
|
181 |
|
|
return -1;
|
182 |
|
|
for (level = ts->level - level; level > 0; --level)
|
183 |
|
|
ts = &ts->team->prev_ts;
|
184 |
|
|
if (ts->team == NULL)
|
185 |
|
|
return 1;
|
186 |
|
|
else
|
187 |
|
|
return ts->team->nthreads;
|
188 |
|
|
}
|
189 |
|
|
|
190 |
|
|
int
|
191 |
|
|
omp_get_active_level (void)
|
192 |
|
|
{
|
193 |
|
|
return gomp_thread ()->ts.active_level;
|
194 |
|
|
}
|
195 |
|
|
|
196 |
|
|
ialias (omp_get_num_threads)
|
197 |
|
|
ialias (omp_get_thread_num)
|
198 |
|
|
ialias (omp_in_parallel)
|
199 |
|
|
ialias (omp_get_level)
|
200 |
|
|
ialias (omp_get_ancestor_thread_num)
|
201 |
|
|
ialias (omp_get_team_size)
|
202 |
|
|
ialias (omp_get_active_level)
|