| 1 |
735 |
jeremybenn |
/* Copyright (C) 2005, 2008, 2009, 2011 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 is a Linux specific implementation of a barrier synchronization
|
| 26 |
|
|
mechanism for libgomp. This type is private to the library. This
|
| 27 |
|
|
implementation uses atomic instructions and the futex syscall. */
|
| 28 |
|
|
|
| 29 |
|
|
#include <limits.h>
|
| 30 |
|
|
#include "wait.h"
|
| 31 |
|
|
|
| 32 |
|
|
|
| 33 |
|
|
void
|
| 34 |
|
|
gomp_barrier_wait_end (gomp_barrier_t *bar, gomp_barrier_state_t state)
|
| 35 |
|
|
{
|
| 36 |
|
|
if (__builtin_expect ((state & 1) != 0, 0))
|
| 37 |
|
|
{
|
| 38 |
|
|
/* Next time we'll be awaiting TOTAL threads again. */
|
| 39 |
|
|
bar->awaited = bar->total;
|
| 40 |
|
|
__atomic_store_n (&bar->generation, bar->generation + 4,
|
| 41 |
|
|
MEMMODEL_RELEASE);
|
| 42 |
|
|
futex_wake ((int *) &bar->generation, INT_MAX);
|
| 43 |
|
|
}
|
| 44 |
|
|
else
|
| 45 |
|
|
{
|
| 46 |
|
|
do
|
| 47 |
|
|
do_wait ((int *) &bar->generation, state);
|
| 48 |
|
|
while (__atomic_load_n (&bar->generation, MEMMODEL_ACQUIRE) == state);
|
| 49 |
|
|
}
|
| 50 |
|
|
}
|
| 51 |
|
|
|
| 52 |
|
|
void
|
| 53 |
|
|
gomp_barrier_wait (gomp_barrier_t *bar)
|
| 54 |
|
|
{
|
| 55 |
|
|
gomp_barrier_wait_end (bar, gomp_barrier_wait_start (bar));
|
| 56 |
|
|
}
|
| 57 |
|
|
|
| 58 |
|
|
/* Like gomp_barrier_wait, except that if the encountering thread
|
| 59 |
|
|
is not the last one to hit the barrier, it returns immediately.
|
| 60 |
|
|
The intended usage is that a thread which intends to gomp_barrier_destroy
|
| 61 |
|
|
this barrier calls gomp_barrier_wait, while all other threads
|
| 62 |
|
|
call gomp_barrier_wait_last. When gomp_barrier_wait returns,
|
| 63 |
|
|
the barrier can be safely destroyed. */
|
| 64 |
|
|
|
| 65 |
|
|
void
|
| 66 |
|
|
gomp_barrier_wait_last (gomp_barrier_t *bar)
|
| 67 |
|
|
{
|
| 68 |
|
|
gomp_barrier_state_t state = gomp_barrier_wait_start (bar);
|
| 69 |
|
|
if (state & 1)
|
| 70 |
|
|
gomp_barrier_wait_end (bar, state);
|
| 71 |
|
|
}
|
| 72 |
|
|
|
| 73 |
|
|
void
|
| 74 |
|
|
gomp_team_barrier_wake (gomp_barrier_t *bar, int count)
|
| 75 |
|
|
{
|
| 76 |
|
|
futex_wake ((int *) &bar->generation, count == 0 ? INT_MAX : count);
|
| 77 |
|
|
}
|
| 78 |
|
|
|
| 79 |
|
|
void
|
| 80 |
|
|
gomp_team_barrier_wait_end (gomp_barrier_t *bar, gomp_barrier_state_t state)
|
| 81 |
|
|
{
|
| 82 |
|
|
unsigned int generation, gen;
|
| 83 |
|
|
|
| 84 |
|
|
if (__builtin_expect ((state & 1) != 0, 0))
|
| 85 |
|
|
{
|
| 86 |
|
|
/* Next time we'll be awaiting TOTAL threads again. */
|
| 87 |
|
|
struct gomp_thread *thr = gomp_thread ();
|
| 88 |
|
|
struct gomp_team *team = thr->ts.team;
|
| 89 |
|
|
|
| 90 |
|
|
bar->awaited = bar->total;
|
| 91 |
|
|
if (__builtin_expect (team->task_count, 0))
|
| 92 |
|
|
{
|
| 93 |
|
|
gomp_barrier_handle_tasks (state);
|
| 94 |
|
|
state &= ~1;
|
| 95 |
|
|
}
|
| 96 |
|
|
else
|
| 97 |
|
|
{
|
| 98 |
|
|
__atomic_store_n (&bar->generation, state + 3, MEMMODEL_RELEASE);
|
| 99 |
|
|
futex_wake ((int *) &bar->generation, INT_MAX);
|
| 100 |
|
|
return;
|
| 101 |
|
|
}
|
| 102 |
|
|
}
|
| 103 |
|
|
|
| 104 |
|
|
generation = state;
|
| 105 |
|
|
do
|
| 106 |
|
|
{
|
| 107 |
|
|
do_wait ((int *) &bar->generation, generation);
|
| 108 |
|
|
gen = __atomic_load_n (&bar->generation, MEMMODEL_ACQUIRE);
|
| 109 |
|
|
if (__builtin_expect (gen & 1, 0))
|
| 110 |
|
|
{
|
| 111 |
|
|
gomp_barrier_handle_tasks (state);
|
| 112 |
|
|
gen = __atomic_load_n (&bar->generation, MEMMODEL_ACQUIRE);
|
| 113 |
|
|
}
|
| 114 |
|
|
if ((gen & 2) != 0)
|
| 115 |
|
|
generation |= 2;
|
| 116 |
|
|
}
|
| 117 |
|
|
while (gen != state + 4);
|
| 118 |
|
|
}
|
| 119 |
|
|
|
| 120 |
|
|
void
|
| 121 |
|
|
gomp_team_barrier_wait (gomp_barrier_t *bar)
|
| 122 |
|
|
{
|
| 123 |
|
|
gomp_team_barrier_wait_end (bar, gomp_barrier_wait_start (bar));
|
| 124 |
|
|
}
|