gcc/gcc43-libgomp-speedup.patch
Jakub Jelinek c358f1d2d3 4.3.0-5
2008-03-27 14:32:22 +00:00

2793 lines
90 KiB
Diff

2008-03-27 Jakub Jelinek <jakub@redhat.com>
* libgomp.h (struct gomp_team_state): Remove single_count field
ifndef HAVE_SYNC_BUILTINS.
(struct gomp_team): Likewise. Add work_share_list_free_lock
ifndef HAVE_SYNC_BUILTINS.
* team.c (gomp_new_team): If HAVE_SYNC_BUILTINS is not defined,
don't initialize single_count, but instead initialize
work_share_list_free_lock.
(free_team): Destroy work_share_list_free_lock ifndef
HAVE_SYNC_BUILTINS.
(gomp_team_start): Don't initialize ts.single_count ifndef
HAVE_SYNC_BUILTINS.
* work.c (alloc_work_share, free_work_share): Use
work_share_list_free_lock instead of atomic chaining ifndef
HAVE_SYNC_BUILTINS.
2008-03-26 Jakub Jelinek <jakub@redhat.com>
* loop.c (gomp_loop_init): Fix GFS_DYNAMIC ws->mode setting.
* testsuite/libgomp.c/loop-4.c: New test.
* libgomp.h (struct gomp_team_state): Add single_count field.
(struct gomp_team): Likewise.
* team.c (gomp_new_team): Clear single_count.
(gomp_team_start): Likewise.
* single.c (GOMP_single_start): Rewritten if HAVE_SYNC_BUILTINS.
2008-03-25 Jakub Jelinek <jakub@redhat.com>
* team.c (gomp_thread_start): Don't clear ts.static_trip here.
* loop.c (gomp_loop_static_start, gomp_loop_dynamic_start): Clear
ts.static_trip here.
* work.c (gomp_work_share_start): Don't clear ts.static_trip here.
2008-03-21 Jakub Jelinek <jakub@redhat.com>
* libgomp.h: Include ptrlock.h.
(struct gomp_work_share): Reshuffle fields. Add next_alloc,
next_ws, next_free and inline_ordered_team_ids fields, change
ordered_team_ids into pointer from flexible array member.
(struct gomp_team_state): Add last_work_share field, remove
work_share_generation.
(struct gomp_team): Remove work_share_lock, generation_mask,
oldest_live_gen, num_live_gen and init_work_shares fields, add
work work_share_list_alloc, work_share_list_free and work_share_chunk
fields. Change work_shares from pointer to pointers into an array.
(gomp_new_team): New prototype.
(gomp_team_start): Change type of last argument.
(gomp_new_work_share): Removed.
(gomp_init_work_share, gomp_fini_work_share): New prototypes.
(gomp_work_share_init_done): New static inline.
* team.c (gomp_thread_start): Clear ts.last_work_share, don't clear
ts.work_share_generation.
(new_team): Removed.
(gomp_new_team): New function.
(free_team): Free gomp_work_share blocks chained through next_alloc,
instead of freeing work_shares and destroying work_share_lock.
(gomp_team_start): Change last argument from ws to team, don't create
new team, set ts.work_share to &team->work_shares[0] and clear
ts.last_work_share. Don't clear ts.work_share_generation.
(gomp_team_end): Call gomp_fini_work_share.
* work.c (gomp_new_work_share): Removed.
(alloc_work_share, gomp_init_work_share, gomp_fini_work_share): New
functions.
(free_work_share): Add team argument. Call gomp_fini_work_share
and then either free ws if orphaned, or put it into
work_share_list_free list of the current team.
(gomp_work_share_start, gomp_work_share_end,
gomp_work_share_end_nowait): Rewritten.
* sections.c (GOMP_sections_start): Call gomp_work_share_init_done
after gomp_sections_init. If HAVE_SYNC_BUILTINS, call
gomp_iter_dynamic_next instead of the _locked variant and don't take
lock around it, otherwise acquire it before calling
gomp_iter_dynamic_next_locked.
(GOMP_sections_next): If HAVE_SYNC_BUILTINS, call
gomp_iter_dynamic_next instead of the _locked variant and don't take
lock around it.
(GOMP_parallel_sections_start): Call gomp_new_team instead of
gomp_new_work_share. Call gomp_sections_init on &team->work_shares[0].
Adjust gomp_team_start caller.
* loop.c (gomp_loop_static_start, gomp_loop_ordered_static_start): Call
gomp_work_share_init_done after gomp_loop_init. Don't unlock ws->lock.
(gomp_loop_dynamic_start, gomp_loop_guided_start): Call
gomp_work_share_init_done after gomp_loop_init. If HAVE_SYNC_BUILTINS,
don't unlock ws->lock, otherwise lock it.
(gomp_loop_ordered_dynamic_start, gomp_loop_ordered_guided_start): Call
gomp_work_share_init_done after gomp_loop_init. Lock ws->lock.
(gomp_parallel_loop_start): Call gomp_new_team instead of
gomp_new_work_share. Call gomp_loop_init on &team->work_shares[0].
Adjust gomp_team_start caller.
* single.c (GOMP_single_start, GOMP_single_copy_start): Call
gomp_work_share_init_done if gomp_work_share_start returned true.
Don't unlock ws->lock.
* parallel.c (GOMP_parallel_start): Call gomp_new_team and pass that
as last argument to gomp_team_start.
* config/linux/ptrlock.c: New file.
* config/linux/ptrlock.h: New file.
* config/posix/ptrlock.c: New file.
* config/posix/ptrlock.h: New file.
* Makefile.am (libgomp_la_SOURCES): Add ptrlock.c.
* Makefile.in: Regenerated.
* testsuite/Makefile.in: Regenerated.
2008-03-19 Jakub Jelinek <jakub@redhat.com>
* libgomp.h (gomp_active_wait_policy): Remove decl.
(gomp_throttled_spin_count_var, gomp_available_cpus,
gomp_managed_threads): New extern decls.
* team.c (gomp_team_start, gomp_team_end): If number of threads
changed, adjust atomically gomp_managed_threads.
* env.c (gomp_active_wait_policy, gomp_block_time_var): Remove.
(gomp_throttled_spin_count_var, gomp_available_cpus,
gomp_managed_threads): New variables.
(parse_millis): Removed.
(parse_spincount): New function.
(parse_wait_policy): Return -1/0/1 instead of setting
gomp_active_wait_policy.
(initialize_env): Call gomp_init_num_threads unconditionally.
Initialize gomp_available_cpus. Call parse_spincount instead
of parse_millis, initialize gomp_{,throttled_}spin_count_var
depending on presence and value of OMP_WAIT_POLICY and
GOMP_SPINCOUNT env vars.
* config/linux/wait.h (do_wait): Use gomp_throttled_spin_count_var
instead of gomp_spin_count_var if gomp_managed_threads >
gomp_available_cpus.
* config/linux/wait.h: Include errno.h.
(FUTEX_WAIT, FUTEX_WAKE, FUTEX_PRIVATE_FLAG): Define.
(gomp_futex_wake, gomp_futex_wait): New extern decls.
* config/linux/mutex.c (gomp_futex_wake, gomp_futex_wait): New
variables.
* config/linux/powerpc/futex.h (FUTEX_WAIT, FUTEX_WAKE): Remove.
(sys_futex0): Return error code.
(futex_wake, futex_wait): If ENOSYS was returned, clear
FUTEX_PRIVATE_FLAG in gomp_futex_wa{ke,it} and retry.
* config/linux/alpha/futex.h (FUTEX_WAIT, FUTEX_WAKE): Remove.
(futex_wake, futex_wait): If ENOSYS was returned, clear
FUTEX_PRIVATE_FLAG in gomp_futex_wa{ke,it} and retry.
* config/linux/x86/futex.h (FUTEX_WAIT, FUTEX_WAKE): Remove.
(sys_futex0): Return error code.
(futex_wake, futex_wait): If ENOSYS was returned, clear
FUTEX_PRIVATE_FLAG in gomp_futex_wa{ke,it} and retry.
* config/linux/s390/futex.h (FUTEX_WAIT, FUTEX_WAKE): Remove.
(sys_futex0): Return error code.
(futex_wake, futex_wait): If ENOSYS was returned, clear
FUTEX_PRIVATE_FLAG in gomp_futex_wa{ke,it} and retry.
* config/linux/ia64/futex.h (FUTEX_WAIT, FUTEX_WAKE): Remove.
(sys_futex0): Return error code.
(futex_wake, futex_wait): If ENOSYS was returned, clear
FUTEX_PRIVATE_FLAG in gomp_futex_wa{ke,it} and retry.
* config/linux/sparc/futex.h (FUTEX_WAIT, FUTEX_WAKE): Remove.
(sys_futex0): Return error code.
(futex_wake, futex_wait): If ENOSYS was returned, clear
FUTEX_PRIVATE_FLAG in gomp_futex_wa{ke,it} and retry.
2008-03-18 Jakub Jelinek <jakub@redhat.com>
* libgomp.h (struct gomp_work_share): Add mode field. Put lock and
next into a different cache line from most of the write-once fields.
* loop.c: Include limits.h.
(gomp_loop_init): For GFS_DYNAMIC, multiply ws->chunk_size by incr.
If adding ws->chunk_size nthreads + 1 times after end won't
overflow, set ws->mode to 1.
* iter.c (gomp_iter_dynamic_next_locked): Don't multiply
ws->chunk_size by incr.
(gomp_iter_dynamic_next): Likewise. If ws->mode, use more efficient
code.
* work.c: Include stddef.h.
(gomp_new_work_share): Use offsetof rather than sizeof.
2008-03-17 Jakub Jelinek <jakub@redhat.com>
* libgomp.h (struct gomp_team): Change ordered_release field
into gomp_sem_t ** from flexible array member. Add implicit_task
and initial_work_shares fields.
(gomp_new_task): Removed.
(gomp_init_task): New prototype.
* team.c (new_team): Allocate implicit_task for each thread
and initial work_shares together with gomp_team allocation.
(free_team): Only free work_shares if it is not init_work_shares.
(gomp_team_start): Use gomp_init_task instead of gomp_new_task,
set thr->task to the corresponding implicit_task array entry.
* task.c (gomp_new_task): Removed.
(gomp_init_task): New function.
(gomp_end_task): Don't free the task.
(GOMP_task): Allocate struct gomp_task on the stack, call
gomp_init_task rather than gomp_new_task.
* work.c (gomp_work_share_start): If work_shares ==
init_work_shares, gomp_malloc + memcpy rather than gomp_realloc.
2008-03-15 Jakub Jelinek <jakub@redhat.com>
Ulrich Drepper <drepper@redhat.com>
* config/linux/bar.h (gomp_barrier_state_t): Rewritten.
(gomp_barrier_state_t): Change to unsigned int.
(gomp_barrier_init, gomp_barrier_reinit, gomp_barrier_destroy,
gomp_barrier_wait_start, gomp_barrier_last_thread): Rewritten.
(gomp_barrier_wait_last): Prototype rather than inline.
* config/linux/bar.c (gomp_barrier_wait_end): Rewritten.
(gomp_barrier_wait_last): New function.
2008-03-15 Jakub Jelinek <jakub@redhat.com>
* team.c (gomp_thread_start): Use gomp_barrier_wait_last instead
of gomp_barrier_wait.
* env.c (gomp_block_time_var, gomp_spin_count_var): New variables.
(parse_millis): New function.
(initialize_env): Handle GOMP_BLOCKTIME env var.
* libgomp.h (struct gomp_team): Move close to the end of the struct.
(gomp_spin_count_var): New extern var decl.
* work.c (gomp_work_share_end): Use gomp_barrier_state_t bstate
var instead of bool last, call gomp_barrier_last_thread to check
for last thread, pass bstate to gomp_barrier_wait_end.
* config/linux/wait.h: New file.
* config/linux/mutex.c: Include wait.h instead of libgomp.h and
futex.h.
(gomp_mutex_lock_slow): Call do_wait instead of futex_wait.
* config/linux/bar.c: Include wait.h instead of libgomp.h and
futex.h.
(gomp_barrier_wait_end): Change second argument to
gomp_barrier_state_t. Call do_wait instead of futex_wait.
* config/linux/sem.c: Include wait.h instead of libgomp.h and
futex.h.
(gomp_sem_wait_slow): Call do_wait instead of futex_wait.
* config/linux/lock.c: Include wait.h instead of libgomp.h and
futex.h.
(gomp_set_nest_lock_25): Call do_wait instead of futex_wait.
* config/linux/affinity.c: Assume HAVE_SYNC_BUILTINS.
* config/linux/bar.h (gomp_barrier_state_t): New typedef.
(gomp_barrier_wait_end): Change second argument to
gomp_barrier_state_t.
(gomp_barrier_wait_start): Return gomp_barrier_state_t.
(gomp_barrier_last_thread, gomp_barrier_wait_last): New static
inlines.
* config/linux/powerpc/futex.h (cpu_relax, atomic_write_barrier): New
static inlines.
* config/linux/alpha/futex.h (cpu_relax, atomic_write_barrier):
Likewise.
* config/linux/x86/futex.h (cpu_relax, atomic_write_barrier):
Likewise.
* config/linux/s390/futex.h (cpu_relax, atomic_write_barrier):
Likewise.
* config/linux/ia64/futex.h (cpu_relax, atomic_write_barrier):
Likewise.
* config/linux/sparc/futex.h (cpu_relax, atomic_write_barrier):
Likewise.
* config/posix/bar.c (gomp_barrier_wait_end): Change second argument
to gomp_barrier_state_t.
* config/posix/bar.h (gomp_barrier_state_t): New typedef.
(gomp_barrier_wait_end): Change second argument to
gomp_barrier_state_t.
(gomp_barrier_wait_start): Return gomp_barrier_state_t.
(gomp_barrier_last_thread, gomp_barrier_wait_last): New static
inlines.
--- libgomp/parallel.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/parallel.c 2008-03-26 15:32:06.000000000 +0100
@@ -68,7 +68,7 @@ void
GOMP_parallel_start (void (*fn) (void *), void *data, unsigned num_threads)
{
num_threads = gomp_resolve_num_threads (num_threads);
- gomp_team_start (fn, data, num_threads, NULL);
+ gomp_team_start (fn, data, num_threads, gomp_new_team (num_threads));
}
void
--- libgomp/sections.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/sections.c 2008-03-26 15:33:06.000000000 +0100
@@ -59,14 +59,24 @@ GOMP_sections_start (unsigned count)
long s, e, ret;
if (gomp_work_share_start (false))
- gomp_sections_init (thr->ts.work_share, count);
+ {
+ gomp_sections_init (thr->ts.work_share, count);
+ gomp_work_share_init_done ();
+ }
+#ifdef HAVE_SYNC_BUILTINS
+ if (gomp_iter_dynamic_next (&s, &e))
+ ret = s;
+ else
+ ret = 0;
+#else
+ gomp_mutex_lock (&thr->ts.work_share->lock);
if (gomp_iter_dynamic_next_locked (&s, &e))
ret = s;
else
ret = 0;
-
gomp_mutex_unlock (&thr->ts.work_share->lock);
+#endif
return ret;
}
@@ -83,15 +93,23 @@ GOMP_sections_start (unsigned count)
unsigned
GOMP_sections_next (void)
{
- struct gomp_thread *thr = gomp_thread ();
long s, e, ret;
+#ifdef HAVE_SYNC_BUILTINS
+ if (gomp_iter_dynamic_next (&s, &e))
+ ret = s;
+ else
+ ret = 0;
+#else
+ struct gomp_thread *thr = gomp_thread ();
+
gomp_mutex_lock (&thr->ts.work_share->lock);
if (gomp_iter_dynamic_next_locked (&s, &e))
ret = s;
else
ret = 0;
gomp_mutex_unlock (&thr->ts.work_share->lock);
+#endif
return ret;
}
@@ -103,15 +121,15 @@ void
GOMP_parallel_sections_start (void (*fn) (void *), void *data,
unsigned num_threads, unsigned count)
{
- struct gomp_work_share *ws;
+ struct gomp_team *team;
num_threads = gomp_resolve_num_threads (num_threads);
if (gomp_dyn_var && num_threads > count)
num_threads = count;
- ws = gomp_new_work_share (false, num_threads);
- gomp_sections_init (ws, count);
- gomp_team_start (fn, data, num_threads, ws);
+ team = gomp_new_team (num_threads);
+ gomp_sections_init (&team->work_shares[0], count);
+ gomp_team_start (fn, data, num_threads, team);
}
/* The GOMP_section_end* routines are called after the thread is told
--- libgomp/env.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/env.c 2008-03-26 16:40:26.000000000 +0100
@@ -44,6 +44,11 @@ enum gomp_schedule_type gomp_run_sched_v
unsigned long gomp_run_sched_chunk = 1;
unsigned short *gomp_cpu_affinity;
size_t gomp_cpu_affinity_len;
+#ifndef HAVE_SYNC_BUILTINS
+gomp_mutex_t gomp_remaining_threads_lock;
+#endif
+unsigned long gomp_available_cpus = 1, gomp_managed_threads = 1;
+unsigned long long gomp_spin_count_var, gomp_throttled_spin_count_var;
/* Parse the OMP_SCHEDULE environment variable. */
@@ -147,6 +152,79 @@ parse_unsigned_long (const char *name, u
return false;
}
+/* Parse the GOMP_SPINCOUNT environment varible. Return true if one was
+ present and it was successfully parsed. */
+
+static bool
+parse_spincount (const char *name, unsigned long long *pvalue)
+{
+ char *env, *end;
+ unsigned long long value, mult = 1;
+
+ env = getenv (name);
+ if (env == NULL)
+ return false;
+
+ while (isspace ((unsigned char) *env))
+ ++env;
+ if (*env == '\0')
+ goto invalid;
+
+ if (strncasecmp (env, "infinite", 8) == 0
+ || strncasecmp (env, "infinity", 8) == 0)
+ {
+ value = ~0ULL;
+ end = env + 8;
+ goto check_tail;
+ }
+
+ errno = 0;
+ value = strtoull (env, &end, 10);
+ if (errno)
+ goto invalid;
+
+ while (isspace ((unsigned char) *end))
+ ++end;
+ if (*end != '\0')
+ {
+ switch (tolower (*end))
+ {
+ case 'k':
+ mult = 1000LL;
+ break;
+ case 'm':
+ mult = 1000LL * 1000LL;
+ break;
+ case 'g':
+ mult = 1000LL * 1000LL * 1000LL;
+ break;
+ case 't':
+ mult = 1000LL * 1000LL * 1000LL * 1000LL;
+ break;
+ default:
+ goto invalid;
+ }
+ ++end;
+ check_tail:
+ while (isspace ((unsigned char) *end))
+ ++end;
+ if (*end != '\0')
+ goto invalid;
+ }
+
+ if (value > ~0ULL / mult)
+ value = ~0ULL;
+ else
+ value *= mult;
+
+ *pvalue = value;
+ return true;
+
+ invalid:
+ gomp_error ("Invalid value for environment variable %s", name);
+ return false;
+}
+
/* Parse a boolean value for environment variable NAME and store the
result in VALUE. */
@@ -281,10 +359,25 @@ initialize_env (void)
parse_schedule ();
parse_boolean ("OMP_DYNAMIC", &gomp_dyn_var);
parse_boolean ("OMP_NESTED", &gomp_nest_var);
+ gomp_init_num_threads ();
+ gomp_available_cpus = gomp_nthreads_var;
if (!parse_unsigned_long ("OMP_NUM_THREADS", &gomp_nthreads_var))
- gomp_init_num_threads ();
+ gomp_nthreads_var = gomp_available_cpus;
if (parse_affinity ())
gomp_init_affinity ();
+ if (!parse_spincount ("GOMP_SPINCOUNT", &gomp_spin_count_var))
+ {
+ /* Using a rough estimation of 100000 spins per msec,
+ use 200 msec blocking.
+ Depending on the CPU speed, this can be e.g. 5 times longer
+ or 5 times shorter. */
+ gomp_spin_count_var = 20000000LL;
+ }
+ /* gomp_throttled_spin_count_var is used when there are more libgomp
+ managed threads than available CPUs. Use very short spinning. */
+ gomp_throttled_spin_count_var = 100LL;
+ if (gomp_throttled_spin_count_var > gomp_spin_count_var)
+ gomp_throttled_spin_count_var = gomp_spin_count_var;
/* Not strictly environment related, but ordering constructors is tricky. */
pthread_attr_init (&gomp_thread_attr);
--- libgomp/libgomp.h.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/libgomp.h 2008-03-27 12:21:51.000000000 +0100
@@ -50,6 +50,7 @@
#include "sem.h"
#include "mutex.h"
#include "bar.h"
+#include "ptrlock.h"
/* This structure contains the data to control one work-sharing construct,
@@ -70,6 +71,8 @@ struct gomp_work_share
If this is a SECTIONS construct, this value will always be DYNAMIC. */
enum gomp_schedule_type sched;
+ int mode;
+
/* This is the chunk_size argument to the SCHEDULE clause. */
long chunk_size;
@@ -81,17 +84,38 @@ struct gomp_work_share
is always 1. */
long incr;
- /* This lock protects the update of the following members. */
- gomp_mutex_t lock;
+ /* 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;
- 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;
+ /* This is the number of threads that have registered themselves in
+ the circular queue ordered_team_ids. */
+ unsigned ordered_num_used;
- /* This is the returned data structure for SINGLE COPYPRIVATE. */
- void *copyprivate;
- };
+ /* 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
@@ -99,27 +123,28 @@ struct gomp_work_share
of the team to exit the work share construct must deallocate it. */
unsigned threads_completed;
- /* 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;
+ 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;
- /* This is the number of threads that have registered themselves in
- the circular queue ordered_team_ids. */
- unsigned ordered_num_used;
+ /* This is the returned data structure for SINGLE COPYPRIVATE. */
+ void *copyprivate;
+ };
- /* This is the team_id of the currently acknoledged 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;
+ 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;
+ };
- /* 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[];
+ /* 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
@@ -133,21 +158,24 @@ struct gomp_team_state
/* 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. This value is NULL when there
- is no construct being processed. */
+ 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;
- /* The work share "generation" is a number that increases by one for
- each work share construct encountered in the dynamic flow of the
- program. It is used to find the control data for the work share
- when encountering it for the first time. This particular number
- reflects the generation of the work_share member of this struct. */
- unsigned work_share_generation;
+#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
@@ -163,41 +191,53 @@ struct gomp_team_state
struct gomp_team
{
- /* This lock protects access to the following work shares data structures. */
- gomp_mutex_t work_share_lock;
-
- /* This is a dynamically sized array containing pointers to the control
- structs for all "live" work share constructs. Here "live" means that
- the construct has been encountered by at least one thread, and not
- completed by all threads. */
- struct gomp_work_share **work_shares;
-
- /* The work_shares array is indexed by "generation & generation_mask".
- The mask will be 2**N - 1, where 2**N is the size of the array. */
- unsigned generation_mask;
-
- /* These two values define the bounds of the elements of the work_shares
- array that are currently in use. */
- unsigned oldest_live_gen;
- unsigned num_live_gen;
-
/* 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 barrier is used for most synchronization of the team. */
- gomp_barrier_t barrier;
-
/* 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 array contains pointers to the release semaphore of the threads
- in the team. */
+ /* 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];
+
+ /* This is an array with pointers to the release semaphore
+ of the threads in the team. */
gomp_sem_t *ordered_release[];
};
@@ -242,6 +282,11 @@ extern bool gomp_dyn_var;
extern bool gomp_nest_var;
extern enum gomp_schedule_type gomp_run_sched_var;
extern unsigned long gomp_run_sched_chunk;
+#ifndef HAVE_SYNC_BUILTINS
+extern gomp_mutex_t gomp_remaining_threads_lock;
+#endif
+extern unsigned long long gomp_spin_count_var, gomp_throttled_spin_count_var;
+extern unsigned long gomp_available_cpus, gomp_managed_threads;
/* The attributes to be used during thread creation. */
extern pthread_attr_t gomp_thread_attr;
@@ -306,17 +351,27 @@ extern unsigned gomp_dynamic_max_threads
/* team.c */
+extern struct gomp_team *gomp_new_team (unsigned);
extern void gomp_team_start (void (*) (void *), void *, unsigned,
- struct gomp_work_share *);
+ struct gomp_team *);
extern void gomp_team_end (void);
/* work.c */
-extern struct gomp_work_share * gomp_new_work_share (bool, unsigned);
+extern void gomp_init_work_share (struct gomp_work_share *, bool, unsigned);
+extern void gomp_fini_work_share (struct gomp_work_share *);
extern bool gomp_work_share_start (bool);
extern void gomp_work_share_end (void);
extern void gomp_work_share_end_nowait (void);
+static inline void
+gomp_work_share_init_done (void)
+{
+ struct gomp_thread *thr = gomp_thread ();
+ if (__builtin_expect (thr->ts.last_work_share != NULL, 1))
+ gomp_ptrlock_set (&thr->ts.last_work_share->next_ws, thr->ts.work_share);
+}
+
#ifdef HAVE_ATTRIBUTE_VISIBILITY
# pragma GCC visibility pop
#endif
--- libgomp/iter.c.jj 2008-03-26 14:48:34.000000000 +0100
+++ libgomp/iter.c 2008-03-26 15:11:23.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -154,7 +154,7 @@ gomp_iter_dynamic_next_locked (long *pst
if (start == ws->end)
return false;
- chunk = ws->chunk_size * ws->incr;
+ chunk = ws->chunk_size;
left = ws->end - start;
if (ws->incr < 0)
{
@@ -186,11 +186,38 @@ gomp_iter_dynamic_next (long *pstart, lo
struct gomp_work_share *ws = thr->ts.work_share;
long start, end, nend, chunk, incr;
- start = ws->next;
end = ws->end;
incr = ws->incr;
- chunk = ws->chunk_size * incr;
+ chunk = ws->chunk_size;
+
+ if (__builtin_expect (ws->mode, 1))
+ {
+ long tmp = __sync_fetch_and_add (&ws->next, chunk);
+ if (incr > 0)
+ {
+ if (tmp >= end)
+ return false;
+ nend = tmp + chunk;
+ if (nend > end)
+ nend = end;
+ *pstart = tmp;
+ *pend = nend;
+ return true;
+ }
+ else
+ {
+ if (tmp <= end)
+ return false;
+ nend = tmp + chunk;
+ if (nend < end)
+ nend = end;
+ *pstart = tmp;
+ *pend = nend;
+ return true;
+ }
+ }
+ start = ws->next;
while (1)
{
long left = end - start;
--- libgomp/work.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/work.c 2008-03-27 12:21:51.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -29,39 +29,138 @@
of threads. */
#include "libgomp.h"
+#include <stddef.h>
#include <stdlib.h>
#include <string.h>
-/* Create a new work share structure. */
+/* Allocate a new work share structure, preferably from current team's
+ free gomp_work_share cache. */
-struct gomp_work_share *
-gomp_new_work_share (bool ordered, unsigned nthreads)
+static struct gomp_work_share *
+alloc_work_share (struct gomp_team *team)
{
struct gomp_work_share *ws;
- size_t size;
+ unsigned int i;
- size = sizeof (*ws);
- if (ordered)
- size += nthreads * sizeof (ws->ordered_team_ids[0]);
+ /* This is called in a critical section. */
+ if (team->work_share_list_alloc != NULL)
+ {
+ ws = team->work_share_list_alloc;
+ team->work_share_list_alloc = ws->next_free;
+ return ws;
+ }
- ws = gomp_malloc_cleared (size);
- gomp_mutex_init (&ws->lock);
- ws->ordered_owner = -1;
+#ifdef HAVE_SYNC_BUILTINS
+ ws = team->work_share_list_free;
+ /* We need atomic read from work_share_list_free,
+ as free_work_share can be called concurrently. */
+ __asm ("" : "+r" (ws));
+
+ if (ws && ws->next_free)
+ {
+ struct gomp_work_share *next = ws->next_free;
+ ws->next_free = NULL;
+ team->work_share_list_alloc = next->next_free;
+ return next;
+ }
+#else
+ gomp_mutex_lock (&team->work_share_list_free_lock);
+ ws = team->work_share_list_free;
+ if (ws)
+ {
+ team->work_share_list_alloc = ws->next_free;
+ team->work_share_list_free = NULL;
+ gomp_mutex_unlock (&team->work_share_list_free_lock);
+ return ws;
+ }
+ gomp_mutex_unlock (&team->work_share_list_free_lock);
+#endif
+ team->work_share_chunk *= 2;
+ ws = gomp_malloc (team->work_share_chunk * sizeof (struct gomp_work_share));
+ ws->next_alloc = team->work_shares[0].next_alloc;
+ team->work_shares[0].next_alloc = ws;
+ team->work_share_list_alloc = &ws[1];
+ for (i = 1; i < team->work_share_chunk - 1; i++)
+ ws[i].next_free = &ws[i + 1];
+ ws[i].next_free = NULL;
return ws;
}
+/* Initialize an already allocated struct gomp_work_share.
+ This shouldn't touch the next_alloc field. */
+
+void
+gomp_init_work_share (struct gomp_work_share *ws, bool ordered,
+ unsigned nthreads)
+{
+ gomp_mutex_init (&ws->lock);
+ if (__builtin_expect (ordered, 0))
+ {
+#define INLINE_ORDERED_TEAM_IDS_CNT \
+ ((sizeof (struct gomp_work_share) \
+ - offsetof (struct gomp_work_share, inline_ordered_team_ids)) \
+ / sizeof (((struct gomp_work_share *) 0)->inline_ordered_team_ids[0]))
+
+ if (nthreads > INLINE_ORDERED_TEAM_IDS_CNT)
+ ws->ordered_team_ids
+ = gomp_malloc (nthreads * sizeof (*ws->ordered_team_ids));
+ else
+ ws->ordered_team_ids = ws->inline_ordered_team_ids;
+ memset (ws->ordered_team_ids, '\0',
+ nthreads * sizeof (*ws->ordered_team_ids));
+ ws->ordered_num_used = 0;
+ ws->ordered_owner = -1;
+ ws->ordered_cur = 0;
+ }
+ else
+ ws->ordered_team_ids = NULL;
+ gomp_ptrlock_init (&ws->next_ws, NULL);
+ ws->threads_completed = 0;
+}
-/* Free a work share structure. */
+/* Do any needed destruction of gomp_work_share fields before it
+ is put back into free gomp_work_share cache or freed. */
-static void
-free_work_share (struct gomp_work_share *ws)
+void
+gomp_fini_work_share (struct gomp_work_share *ws)
{
gomp_mutex_destroy (&ws->lock);
- free (ws);
+ if (ws->ordered_team_ids != ws->inline_ordered_team_ids)
+ free (ws->ordered_team_ids);
+ gomp_ptrlock_destroy (&ws->next_ws);
}
+/* Free a work share struct, if not orphaned, put it into current
+ team's free gomp_work_share cache. */
+
+static inline void
+free_work_share (struct gomp_team *team, struct gomp_work_share *ws)
+{
+ gomp_fini_work_share (ws);
+ if (__builtin_expect (team == NULL, 0))
+ free (ws);
+ else
+ {
+ struct gomp_work_share *next_ws;
+#ifdef HAVE_SYNC_BUILTINS
+ do
+ {
+ next_ws = team->work_share_list_free;
+ ws->next_free = next_ws;
+ }
+ while (!__sync_bool_compare_and_swap (&team->work_share_list_free,
+ next_ws, ws));
+#else
+ gomp_mutex_lock (&team->work_share_list_free_lock);
+ next_ws = team->work_share_list_free;
+ ws->next_free = next_ws;
+ team->work_share_list_free = ws;
+ gomp_mutex_unlock (&team->work_share_list_free_lock);
+#endif
+ }
+}
/* The current thread is ready to begin the next work sharing construct.
In all cases, thr->ts.work_share is updated to point to the new
@@ -74,71 +173,34 @@ gomp_work_share_start (bool ordered)
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_work_share *ws;
- unsigned ws_index, ws_gen;
/* Work sharing constructs can be orphaned. */
if (team == NULL)
{
- ws = gomp_new_work_share (ordered, 1);
+ ws = gomp_malloc (sizeof (*ws));
+ gomp_init_work_share (ws, ordered, 1);
thr->ts.work_share = ws;
- thr->ts.static_trip = 0;
- gomp_mutex_lock (&ws->lock);
- return true;
+ return ws;
}
- gomp_mutex_lock (&team->work_share_lock);
-
- /* This thread is beginning its next generation. */
- ws_gen = ++thr->ts.work_share_generation;
-
- /* If this next generation is not newer than any other generation in
- the team, then simply reference the existing construct. */
- if (ws_gen - team->oldest_live_gen < team->num_live_gen)
+ ws = thr->ts.work_share;
+ thr->ts.last_work_share = ws;
+ ws = gomp_ptrlock_get (&ws->next_ws);
+ if (ws == NULL)
{
- ws_index = ws_gen & team->generation_mask;
- ws = team->work_shares[ws_index];
+ /* This thread encountered a new ws first. */
+ struct gomp_work_share *ws = alloc_work_share (team);
+ gomp_init_work_share (ws, ordered, team->nthreads);
thr->ts.work_share = ws;
- thr->ts.static_trip = 0;
-
- gomp_mutex_lock (&ws->lock);
- gomp_mutex_unlock (&team->work_share_lock);
-
- return false;
+ return true;
}
-
- /* Resize the work shares queue if we've run out of space. */
- if (team->num_live_gen++ == team->generation_mask)
+ else
{
- team->work_shares = gomp_realloc (team->work_shares,
- 2 * team->num_live_gen
- * sizeof (*team->work_shares));
-
- /* Unless oldest_live_gen is zero, the sequence of live elements
- wraps around the end of the array. If we do nothing, we break
- lookup of the existing elements. Fix that by unwrapping the
- data from the front to the end. */
- if (team->oldest_live_gen > 0)
- memcpy (team->work_shares + team->num_live_gen,
- team->work_shares,
- (team->oldest_live_gen & team->generation_mask)
- * sizeof (*team->work_shares));
-
- team->generation_mask = team->generation_mask * 2 + 1;
- }
-
- ws_index = ws_gen & team->generation_mask;
- ws = gomp_new_work_share (ordered, team->nthreads);
- thr->ts.work_share = ws;
- thr->ts.static_trip = 0;
- team->work_shares[ws_index] = ws;
-
- gomp_mutex_lock (&ws->lock);
- gomp_mutex_unlock (&team->work_share_lock);
-
- return true;
+ thr->ts.work_share = ws;
+ return false;
+ }
}
-
/* The current thread is done with its current work sharing construct.
This version does imply a barrier at the end of the work-share. */
@@ -147,36 +209,28 @@ gomp_work_share_end (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
- struct gomp_work_share *ws = thr->ts.work_share;
- bool last;
-
- thr->ts.work_share = NULL;
+ gomp_barrier_state_t bstate;
/* Work sharing constructs can be orphaned. */
if (team == NULL)
{
- free_work_share (ws);
+ free_work_share (NULL, thr->ts.work_share);
+ thr->ts.work_share = NULL;
return;
}
- last = gomp_barrier_wait_start (&team->barrier);
+ bstate = gomp_barrier_wait_start (&team->barrier);
- if (last)
+ if (gomp_barrier_last_thread (bstate))
{
- unsigned ws_index;
-
- ws_index = thr->ts.work_share_generation & team->generation_mask;
- team->work_shares[ws_index] = NULL;
- team->oldest_live_gen++;
- team->num_live_gen = 0;
-
- free_work_share (ws);
+ if (__builtin_expect (thr->ts.last_work_share != NULL, 1))
+ free_work_share (team, thr->ts.last_work_share);
}
- gomp_barrier_wait_end (&team->barrier, last);
+ gomp_barrier_wait_end (&team->barrier, bstate);
+ thr->ts.last_work_share = NULL;
}
-
/* The current thread is done with its current work sharing construct.
This version does NOT imply a barrier at the end of the work-share. */
@@ -188,15 +242,17 @@ gomp_work_share_end_nowait (void)
struct gomp_work_share *ws = thr->ts.work_share;
unsigned completed;
- thr->ts.work_share = NULL;
-
/* Work sharing constructs can be orphaned. */
if (team == NULL)
{
- free_work_share (ws);
+ free_work_share (NULL, ws);
+ thr->ts.work_share = NULL;
return;
}
+ if (__builtin_expect (thr->ts.last_work_share == NULL, 0))
+ return;
+
#ifdef HAVE_SYNC_BUILTINS
completed = __sync_add_and_fetch (&ws->threads_completed, 1);
#else
@@ -206,18 +262,6 @@ gomp_work_share_end_nowait (void)
#endif
if (completed == team->nthreads)
- {
- unsigned ws_index;
-
- gomp_mutex_lock (&team->work_share_lock);
-
- ws_index = thr->ts.work_share_generation & team->generation_mask;
- team->work_shares[ws_index] = NULL;
- team->oldest_live_gen++;
- team->num_live_gen--;
-
- gomp_mutex_unlock (&team->work_share_lock);
-
- free_work_share (ws);
- }
+ free_work_share (team, thr->ts.last_work_share);
+ thr->ts.last_work_share = NULL;
}
--- libgomp/single.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/single.c 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -37,10 +37,24 @@
bool
GOMP_single_start (void)
{
+#ifdef HAVE_SYNC_BUILTINS
+ struct gomp_thread *thr = gomp_thread ();
+ struct gomp_team *team = thr->ts.team;
+ unsigned long single_count;
+
+ if (__builtin_expect (team == NULL, 0))
+ return true;
+
+ single_count = thr->ts.single_count++;
+ return __sync_bool_compare_and_swap (&team->single_count, single_count,
+ single_count + 1L);
+#else
bool ret = gomp_work_share_start (false);
- gomp_mutex_unlock (&gomp_thread ()->ts.work_share->lock);
+ if (ret)
+ gomp_work_share_init_done ();
gomp_work_share_end_nowait ();
return ret;
+#endif
}
/* This routine is called when first encountering a SINGLE construct that
@@ -57,10 +71,12 @@ GOMP_single_copy_start (void)
void *ret;
first = gomp_work_share_start (false);
- gomp_mutex_unlock (&thr->ts.work_share->lock);
if (first)
- ret = NULL;
+ {
+ gomp_work_share_init_done ();
+ ret = NULL;
+ }
else
{
gomp_barrier_wait (&thr->ts.team->barrier);
--- libgomp/loop.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/loop.c 2008-03-26 18:47:04.000000000 +0100
@@ -27,8 +27,9 @@
/* This file handles the LOOP (FOR/DO) construct. */
-#include "libgomp.h"
+#include <limits.h>
#include <stdlib.h>
+#include "libgomp.h"
/* Initialize the given work share construct from the given arguments. */
@@ -44,6 +45,39 @@ gomp_loop_init (struct gomp_work_share *
? start : end;
ws->incr = incr;
ws->next = start;
+ if (sched == GFS_DYNAMIC)
+ {
+ ws->chunk_size *= incr;
+
+#ifdef HAVE_SYNC_BUILTINS
+ {
+ /* For dynamic scheduling prepare things to make each iteration
+ faster. */
+ struct gomp_thread *thr = gomp_thread ();
+ struct gomp_team *team = thr->ts.team;
+ long nthreads = team ? team->nthreads : 1;
+
+ if (__builtin_expect (incr > 0, 1))
+ {
+ /* Cheap overflow protection. */
+ if (__builtin_expect ((nthreads | ws->chunk_size)
+ >= 1UL << (sizeof (long)
+ * __CHAR_BIT__ / 2 - 1), 0))
+ ws->mode = 0;
+ else
+ ws->mode = ws->end < (LONG_MAX
+ - (nthreads + 1) * ws->chunk_size);
+ }
+ /* Cheap overflow protection. */
+ else if (__builtin_expect ((nthreads | -ws->chunk_size)
+ >= 1UL << (sizeof (long)
+ * __CHAR_BIT__ / 2 - 1), 0))
+ ws->mode = 0;
+ else
+ ws->mode = ws->end > (nthreads + 1) * -ws->chunk_size - LONG_MAX;
+ }
+#endif
+ }
}
/* The *_start routines are called when first encountering a loop construct
@@ -68,10 +102,13 @@ gomp_loop_static_start (long start, long
{
struct gomp_thread *thr = gomp_thread ();
+ thr->ts.static_trip = 0;
if (gomp_work_share_start (false))
- gomp_loop_init (thr->ts.work_share, start, end, incr,
- GFS_STATIC, chunk_size);
- gomp_mutex_unlock (&thr->ts.work_share->lock);
+ {
+ gomp_loop_init (thr->ts.work_share, start, end, incr,
+ GFS_STATIC, chunk_size);
+ gomp_work_share_init_done ();
+ }
return !gomp_iter_static_next (istart, iend);
}
@@ -84,13 +121,16 @@ gomp_loop_dynamic_start (long start, lon
bool ret;
if (gomp_work_share_start (false))
- gomp_loop_init (thr->ts.work_share, start, end, incr,
- GFS_DYNAMIC, chunk_size);
+ {
+ gomp_loop_init (thr->ts.work_share, start, end, incr,
+ GFS_DYNAMIC, chunk_size);
+ gomp_work_share_init_done ();
+ }
#ifdef HAVE_SYNC_BUILTINS
- gomp_mutex_unlock (&thr->ts.work_share->lock);
ret = gomp_iter_dynamic_next (istart, iend);
#else
+ gomp_mutex_lock (&thr->ts.work_share->lock);
ret = gomp_iter_dynamic_next_locked (istart, iend);
gomp_mutex_unlock (&thr->ts.work_share->lock);
#endif
@@ -106,13 +146,16 @@ gomp_loop_guided_start (long start, long
bool ret;
if (gomp_work_share_start (false))
- gomp_loop_init (thr->ts.work_share, start, end, incr,
- GFS_GUIDED, chunk_size);
+ {
+ gomp_loop_init (thr->ts.work_share, start, end, incr,
+ GFS_GUIDED, chunk_size);
+ gomp_work_share_init_done ();
+ }
#ifdef HAVE_SYNC_BUILTINS
- gomp_mutex_unlock (&thr->ts.work_share->lock);
ret = gomp_iter_guided_next (istart, iend);
#else
+ gomp_mutex_lock (&thr->ts.work_share->lock);
ret = gomp_iter_guided_next_locked (istart, iend);
gomp_mutex_unlock (&thr->ts.work_share->lock);
#endif
@@ -149,13 +192,14 @@ gomp_loop_ordered_static_start (long sta
{
struct gomp_thread *thr = gomp_thread ();
+ thr->ts.static_trip = 0;
if (gomp_work_share_start (true))
{
gomp_loop_init (thr->ts.work_share, start, end, incr,
GFS_STATIC, chunk_size);
gomp_ordered_static_init ();
+ gomp_work_share_init_done ();
}
- gomp_mutex_unlock (&thr->ts.work_share->lock);
return !gomp_iter_static_next (istart, iend);
}
@@ -168,8 +212,14 @@ gomp_loop_ordered_dynamic_start (long st
bool ret;
if (gomp_work_share_start (true))
- gomp_loop_init (thr->ts.work_share, start, end, incr,
- GFS_DYNAMIC, chunk_size);
+ {
+ gomp_loop_init (thr->ts.work_share, start, end, incr,
+ GFS_DYNAMIC, chunk_size);
+ gomp_mutex_lock (&thr->ts.work_share->lock);
+ gomp_work_share_init_done ();
+ }
+ else
+ gomp_mutex_lock (&thr->ts.work_share->lock);
ret = gomp_iter_dynamic_next_locked (istart, iend);
if (ret)
@@ -187,8 +237,14 @@ gomp_loop_ordered_guided_start (long sta
bool ret;
if (gomp_work_share_start (true))
- gomp_loop_init (thr->ts.work_share, start, end, incr,
- GFS_GUIDED, chunk_size);
+ {
+ gomp_loop_init (thr->ts.work_share, start, end, incr,
+ GFS_GUIDED, chunk_size);
+ gomp_mutex_lock (&thr->ts.work_share->lock);
+ gomp_work_share_init_done ();
+ }
+ else
+ gomp_mutex_lock (&thr->ts.work_share->lock);
ret = gomp_iter_guided_next_locked (istart, iend);
if (ret)
@@ -375,12 +431,12 @@ gomp_parallel_loop_start (void (*fn) (vo
long incr, enum gomp_schedule_type sched,
long chunk_size)
{
- struct gomp_work_share *ws;
+ struct gomp_team *team;
num_threads = gomp_resolve_num_threads (num_threads);
- ws = gomp_new_work_share (false, num_threads);
- gomp_loop_init (ws, start, end, incr, sched, chunk_size);
- gomp_team_start (fn, data, num_threads, ws);
+ team = gomp_new_team (num_threads);
+ gomp_loop_init (&team->work_shares[0], start, end, incr, sched, chunk_size);
+ gomp_team_start (fn, data, num_threads, team);
}
void
--- libgomp/Makefile.in.jj 2008-01-10 20:53:47.000000000 +0100
+++ libgomp/Makefile.in 2008-03-26 18:51:01.000000000 +0100
@@ -83,7 +83,7 @@ libgomp_la_LIBADD =
am_libgomp_la_OBJECTS = alloc.lo barrier.lo critical.lo env.lo \
error.lo iter.lo loop.lo ordered.lo parallel.lo sections.lo \
single.lo team.lo work.lo lock.lo mutex.lo proc.lo sem.lo \
- bar.lo time.lo fortran.lo affinity.lo
+ bar.lo ptrlock.lo time.lo fortran.lo affinity.lo
libgomp_la_OBJECTS = $(am_libgomp_la_OBJECTS)
DEFAULT_INCLUDES = -I. -I$(srcdir) -I.
depcomp = $(SHELL) $(top_srcdir)/../depcomp
@@ -292,7 +292,7 @@ libgomp_version_info = -version-info $(l
libgomp_la_LDFLAGS = $(libgomp_version_info) $(libgomp_version_script)
libgomp_la_SOURCES = alloc.c barrier.c critical.c env.c error.c iter.c \
loop.c ordered.c parallel.c sections.c single.c team.c work.c \
- lock.c mutex.c proc.c sem.c bar.c time.c fortran.c affinity.c
+ lock.c mutex.c proc.c sem.c bar.c ptrlock.c time.c fortran.c affinity.c
nodist_noinst_HEADERS = libgomp_f.h
nodist_libsubinclude_HEADERS = omp.h
@@ -434,6 +434,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ordered.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/parallel.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/proc.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ptrlock.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sections.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sem.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/single.Plo@am__quote@
--- libgomp/testsuite/libgomp.c/loop-4.c.jj 2008-03-26 18:47:04.000000000 +0100
+++ libgomp/testsuite/libgomp.c/loop-4.c 2008-03-26 18:47:04.000000000 +0100
@@ -0,0 +1,28 @@
+/* { dg-do run } */
+
+extern void abort (void);
+
+int
+main (void)
+{
+ int e = 0;
+#pragma omp parallel num_threads (4) reduction(+:e)
+ {
+ long i;
+ #pragma omp for schedule(dynamic,1)
+ for (i = __LONG_MAX__ - 30001; i <= __LONG_MAX__ - 10001; i += 10000)
+ if (i != __LONG_MAX__ - 30001
+ && i != __LONG_MAX__ - 20001
+ && i != __LONG_MAX__ - 10001)
+ e = 1;
+ #pragma omp for schedule(dynamic,1)
+ for (i = -__LONG_MAX__ + 30000; i >= -__LONG_MAX__ + 10000; i -= 10000)
+ if (i != -__LONG_MAX__ + 30000
+ && i != -__LONG_MAX__ + 20000
+ && i != -__LONG_MAX__ + 10000)
+ e = 1;
+ }
+ if (e)
+ abort ();
+ return 0;
+}
--- libgomp/Makefile.am.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/Makefile.am 2008-03-26 15:15:19.000000000 +0100
@@ -31,7 +31,7 @@ libgomp_la_LDFLAGS = $(libgomp_version_i
libgomp_la_SOURCES = alloc.c barrier.c critical.c env.c error.c iter.c \
loop.c ordered.c parallel.c sections.c single.c team.c work.c \
- lock.c mutex.c proc.c sem.c bar.c time.c fortran.c affinity.c
+ lock.c mutex.c proc.c sem.c bar.c ptrlock.c time.c fortran.c affinity.c
nodist_noinst_HEADERS = libgomp_f.h
nodist_libsubinclude_HEADERS = omp.h
--- libgomp/team.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/team.c 2008-03-27 12:22:26.000000000 +0100
@@ -94,7 +94,7 @@ gomp_thread_start (void *xdata)
{
gomp_barrier_wait (&thr->ts.team->barrier);
local_fn (local_data);
- gomp_barrier_wait (&thr->ts.team->barrier);
+ gomp_barrier_wait_last (&thr->ts.team->barrier);
}
else
{
@@ -114,11 +114,10 @@ gomp_thread_start (void *xdata)
thr->data = NULL;
thr->ts.team = NULL;
thr->ts.work_share = NULL;
+ thr->ts.last_work_share = NULL;
thr->ts.team_id = 0;
- thr->ts.work_share_generation = 0;
- thr->ts.static_trip = 0;
- gomp_barrier_wait (&team->barrier);
+ gomp_barrier_wait_last (&team->barrier);
gomp_barrier_wait (&gomp_threads_dock);
local_fn = thr->fn;
@@ -133,21 +132,29 @@ gomp_thread_start (void *xdata)
/* Create a new team data structure. */
-static struct gomp_team *
-new_team (unsigned nthreads, struct gomp_work_share *work_share)
+struct gomp_team *
+gomp_new_team (unsigned nthreads)
{
struct gomp_team *team;
size_t size;
+ int i;
size = sizeof (*team) + nthreads * sizeof (team->ordered_release[0]);
team = gomp_malloc (size);
- gomp_mutex_init (&team->work_share_lock);
- team->work_shares = gomp_malloc (4 * sizeof (struct gomp_work_share *));
- team->generation_mask = 3;
- team->oldest_live_gen = work_share == NULL;
- team->num_live_gen = work_share != NULL;
- team->work_shares[0] = work_share;
+ team->work_share_chunk = 8;
+#ifdef HAVE_SYNC_BUILTINS
+ team->single_count = 0;
+#else
+ gomp_mutex_init (&team->work_share_list_free_lock);
+#endif
+ gomp_init_work_share (&team->work_shares[0], false, nthreads);
+ team->work_shares[0].next_alloc = NULL;
+ team->work_share_list_free = NULL;
+ team->work_share_list_alloc = &team->work_shares[1];
+ for (i = 1; i < 7; i++)
+ team->work_shares[i].next_free = &team->work_shares[i + 1];
+ team->work_shares[i].next_free = NULL;
team->nthreads = nthreads;
gomp_barrier_init (&team->barrier, nthreads);
@@ -164,10 +171,22 @@ new_team (unsigned nthreads, struct gomp
static void
free_team (struct gomp_team *team)
{
- free (team->work_shares);
- gomp_mutex_destroy (&team->work_share_lock);
+ if (__builtin_expect (team->work_shares[0].next_alloc != NULL, 0))
+ {
+ struct gomp_work_share *ws = team->work_shares[0].next_alloc;
+ do
+ {
+ struct gomp_work_share *next_ws = ws->next_alloc;
+ free (ws);
+ ws = next_ws;
+ }
+ while (ws != NULL);
+ }
gomp_barrier_destroy (&team->barrier);
gomp_sem_destroy (&team->master_release);
+#ifndef HAVE_SYNC_BUILTINS
+ gomp_mutex_destroy (&team->work_share_list_free_lock);
+#endif
free (team);
}
@@ -176,11 +195,10 @@ free_team (struct gomp_team *team)
void
gomp_team_start (void (*fn) (void *), void *data, unsigned nthreads,
- struct gomp_work_share *work_share)
+ struct gomp_team *team)
{
struct gomp_thread_start_data *start_data;
struct gomp_thread *thr, *nthr;
- struct gomp_team *team;
bool nested;
unsigned i, n, old_threads_used = 0;
pthread_attr_t thread_attr, *attr;
@@ -188,17 +206,18 @@ gomp_team_start (void (*fn) (void *), vo
thr = gomp_thread ();
nested = thr->ts.team != NULL;
- team = new_team (nthreads, work_share);
-
/* Always save the previous state, even if this isn't a nested team.
In particular, we should save any work share state from an outer
orphaned work share construct. */
team->prev_ts = thr->ts;
thr->ts.team = team;
- thr->ts.work_share = work_share;
thr->ts.team_id = 0;
- thr->ts.work_share_generation = 0;
+ thr->ts.work_share = &team->work_shares[0];
+ thr->ts.last_work_share = NULL;
+#ifdef HAVE_SYNC_BUILTINS
+ thr->ts.single_count = 0;
+#endif
thr->ts.static_trip = 0;
if (nthreads == 1)
@@ -241,9 +260,12 @@ gomp_team_start (void (*fn) (void *), vo
{
nthr = gomp_threads[i];
nthr->ts.team = team;
- nthr->ts.work_share = work_share;
+ nthr->ts.work_share = &team->work_shares[0];
+ nthr->ts.last_work_share = NULL;
nthr->ts.team_id = i;
- nthr->ts.work_share_generation = 0;
+#ifdef HAVE_SYNC_BUILTINS
+ nthr->ts.single_count = 0;
+#endif
nthr->ts.static_trip = 0;
nthr->fn = fn;
nthr->data = data;
@@ -266,8 +288,24 @@ gomp_team_start (void (*fn) (void *), vo
}
}
+ if (__builtin_expect (nthreads > old_threads_used, 0))
+ {
+ long diff = (long) nthreads - (long) old_threads_used;
+
+ if (old_threads_used == 0)
+ --diff;
+
+#ifdef HAVE_SYNC_BUILTINS
+ __sync_fetch_and_add (&gomp_managed_threads, diff);
+#else
+ gomp_mutex_lock (&gomp_remaining_threads_lock);
+ gomp_managed_threads += diff;
+ gomp_mutex_unlock (&gomp_remaining_threads_lock);
+#endif
+ }
+
attr = &gomp_thread_attr;
- if (gomp_cpu_affinity != NULL)
+ if (__builtin_expect (gomp_cpu_affinity != NULL, 0))
{
size_t stacksize;
pthread_attr_init (&thread_attr);
@@ -287,9 +325,12 @@ gomp_team_start (void (*fn) (void *), vo
int err;
start_data->ts.team = team;
- start_data->ts.work_share = work_share;
+ start_data->ts.work_share = &team->work_shares[0];
+ start_data->ts.last_work_share = NULL;
start_data->ts.team_id = i;
- start_data->ts.work_share_generation = 0;
+#ifdef HAVE_SYNC_BUILTINS
+ start_data->ts.single_count = 0;
+#endif
start_data->ts.static_trip = 0;
start_data->fn = fn;
start_data->fn_data = data;
@@ -303,7 +344,7 @@ gomp_team_start (void (*fn) (void *), vo
gomp_fatal ("Thread creation failed: %s", strerror (err));
}
- if (gomp_cpu_affinity != NULL)
+ if (__builtin_expect (gomp_cpu_affinity != NULL, 0))
pthread_attr_destroy (&thread_attr);
do_release:
@@ -313,8 +354,20 @@ gomp_team_start (void (*fn) (void *), vo
that should arrive back at the end of this team. The extra
threads should be exiting. Note that we arrange for this test
to never be true for nested teams. */
- if (nthreads < old_threads_used)
- gomp_barrier_reinit (&gomp_threads_dock, nthreads);
+ if (__builtin_expect (nthreads < old_threads_used, 0))
+ {
+ long diff = (long) nthreads - (long) old_threads_used;
+
+ gomp_barrier_reinit (&gomp_threads_dock, nthreads);
+
+#ifdef HAVE_SYNC_BUILTINS
+ __sync_fetch_and_add (&gomp_managed_threads, diff);
+#else
+ gomp_mutex_lock (&gomp_remaining_threads_lock);
+ gomp_managed_threads += diff;
+ gomp_mutex_unlock (&gomp_remaining_threads_lock);
+#endif
+ }
}
@@ -329,8 +382,21 @@ gomp_team_end (void)
gomp_barrier_wait (&team->barrier);
+ gomp_fini_work_share (thr->ts.work_share);
+
thr->ts = team->prev_ts;
+ if (__builtin_expect (thr->ts.team != NULL, 0))
+ {
+#ifdef HAVE_SYNC_BUILTINS
+ __sync_fetch_and_add (&gomp_managed_threads, 1L - team->nthreads);
+#else
+ gomp_mutex_lock (&gomp_remaining_threads_lock);
+ gomp_managed_threads -= team->nthreads - 1L;
+ gomp_mutex_unlock (&gomp_remaining_threads_lock);
+#endif
+ }
+
free_team (team);
}
--- libgomp/config/posix/bar.h.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/config/posix/bar.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -46,18 +46,32 @@ typedef struct
unsigned total;
unsigned arrived;
} gomp_barrier_t;
+typedef bool gomp_barrier_state_t;
extern void gomp_barrier_init (gomp_barrier_t *, unsigned);
extern void gomp_barrier_reinit (gomp_barrier_t *, unsigned);
extern void gomp_barrier_destroy (gomp_barrier_t *);
extern void gomp_barrier_wait (gomp_barrier_t *);
-extern void gomp_barrier_wait_end (gomp_barrier_t *, bool);
+extern void gomp_barrier_wait_end (gomp_barrier_t *, gomp_barrier_state_t);
-static inline bool gomp_barrier_wait_start (gomp_barrier_t *bar)
+static inline gomp_barrier_state_t
+gomp_barrier_wait_start (gomp_barrier_t *bar)
{
gomp_mutex_lock (&bar->mutex1);
return ++bar->arrived == bar->total;
}
+static inline bool
+gomp_barrier_last_thread (gomp_barrier_state_t state)
+{
+ return state;
+}
+
+static inline void
+gomp_barrier_wait_last (gomp_barrier_t *bar)
+{
+ gomp_barrier_wait (bar);
+}
+
#endif /* GOMP_BARRIER_H */
--- libgomp/config/posix/ptrlock.h.jj 2008-03-26 15:11:32.000000000 +0100
+++ libgomp/config/posix/ptrlock.h 2008-03-26 15:11:32.000000000 +0100
@@ -0,0 +1,69 @@
+/* Copyright (C) 2008 Free Software Foundation, Inc.
+ Contributed by Jakub Jelinek <jakub@redhat.com>.
+
+ This file is part of the GNU OpenMP Library (libgomp).
+
+ Libgomp is free software; you can redistribute it and/or modify it
+ under the terms of the GNU Lesser General Public License as published by
+ the Free Software Foundation; either version 2.1 of the License, 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 Lesser General Public License for
+ more details.
+
+ You should have received a copy of the GNU Lesser General Public License
+ along with libgomp; see the file COPYING.LIB. If not, write to the
+ Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
+ MA 02110-1301, USA. */
+
+/* As a special exception, if you link this library with other files, some
+ of which are compiled with GCC, to produce an executable, this library
+ does not by itself cause the resulting executable to be covered by the
+ GNU General Public License. This exception does not however invalidate
+ any other reasons why the executable file might be covered by the GNU
+ General Public License. */
+
+/* This is a Linux specific implementation of a mutex synchronization
+ mechanism for libgomp. This type is private to the library. This
+ implementation uses atomic instructions and the futex syscall. */
+
+#ifndef GOMP_PTRLOCK_H
+#define GOMP_PTRLOCK_H 1
+
+typedef struct { void *ptr; gomp_mutex_t lock; } gomp_ptrlock_t;
+
+static inline void gomp_ptrlock_init (gomp_ptrlock_t *ptrlock, void *ptr)
+{
+ ptrlock->ptr = ptr;
+ gomp_mutex_init (&ptrlock->lock);
+}
+
+static inline void *gomp_ptrlock_get (gomp_ptrlock_t *ptrlock)
+{
+ if (ptrlock->ptr != NULL)
+ return ptrlock->ptr;
+
+ gomp_mutex_lock (&ptrlock->lock);
+ if (ptrlock->ptr != NULL)
+ {
+ gomp_mutex_unlock (&ptrlock->lock);
+ return ptrlock->ptr;
+ }
+
+ return NULL;
+}
+
+static inline void gomp_ptrlock_set (gomp_ptrlock_t *ptrlock, void *ptr)
+{
+ ptrlock->ptr = ptr;
+ gomp_mutex_unlock (&ptrlock->lock);
+}
+
+static inline void gomp_ptrlock_destroy (gomp_ptrlock_t *ptrlock)
+{
+ gomp_mutex_destroy (&ptrlock->lock);
+}
+
+#endif /* GOMP_PTRLOCK_H */
--- libgomp/config/posix/ptrlock.c.jj 2008-03-26 15:11:32.000000000 +0100
+++ libgomp/config/posix/ptrlock.c 2008-03-26 15:11:32.000000000 +0100
@@ -0,0 +1 @@
+/* Everything is in the header. */
--- libgomp/config/posix/bar.c.jj 2007-12-07 14:41:01.000000000 +0100
+++ libgomp/config/posix/bar.c 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -70,7 +70,7 @@ gomp_barrier_reinit (gomp_barrier_t *bar
}
void
-gomp_barrier_wait_end (gomp_barrier_t *bar, bool last)
+gomp_barrier_wait_end (gomp_barrier_t *bar, gomp_barrier_state_t last)
{
unsigned int n;
--- libgomp/config/linux/alpha/futex.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/alpha/futex.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -30,8 +30,6 @@
#ifndef SYS_futex
#define SYS_futex 394
#endif
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
static inline void
@@ -45,7 +43,7 @@ futex_wait (int *addr, int val)
sc_0 = SYS_futex;
sc_16 = (long) addr;
- sc_17 = FUTEX_WAIT;
+ sc_17 = gomp_futex_wait;
sc_18 = val;
sc_19 = 0;
__asm volatile ("callsys"
@@ -53,6 +51,20 @@ futex_wait (int *addr, int val)
: "0"(sc_0), "r" (sc_16), "r"(sc_17), "r"(sc_18), "1"(sc_19)
: "$1", "$2", "$3", "$4", "$5", "$6", "$7", "$8",
"$22", "$23", "$24", "$25", "$27", "$28", "memory");
+ if (__builtin_expect (sc_19, 0) && sc_0 == ENOSYS)
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sc_0 = SYS_futex;
+ sc_17 &= ~FUTEX_PRIVATE_FLAG;
+ sc_19 = 0;
+ __asm volatile ("callsys"
+ : "=r" (sc_0), "=r"(sc_19)
+ : "0"(sc_0), "r" (sc_16), "r"(sc_17), "r"(sc_18),
+ "1"(sc_19)
+ : "$1", "$2", "$3", "$4", "$5", "$6", "$7", "$8",
+ "$22", "$23", "$24", "$25", "$27", "$28", "memory");
+ }
}
static inline void
@@ -66,11 +78,35 @@ futex_wake (int *addr, int count)
sc_0 = SYS_futex;
sc_16 = (long) addr;
- sc_17 = FUTEX_WAKE;
+ sc_17 = gomp_futex_wake;
sc_18 = count;
__asm volatile ("callsys"
: "=r" (sc_0), "=r"(sc_19)
: "0"(sc_0), "r" (sc_16), "r"(sc_17), "r"(sc_18)
: "$1", "$2", "$3", "$4", "$5", "$6", "$7", "$8",
"$22", "$23", "$24", "$25", "$27", "$28", "memory");
+ if (__builtin_expect (sc_19, 0) && sc_0 == ENOSYS)
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sc_0 = SYS_futex;
+ sc_17 &= ~FUTEX_PRIVATE_FLAG;
+ __asm volatile ("callsys"
+ : "=r" (sc_0), "=r"(sc_19)
+ : "0"(sc_0), "r" (sc_16), "r"(sc_17), "r"(sc_18)
+ : "$1", "$2", "$3", "$4", "$5", "$6", "$7", "$8",
+ "$22", "$23", "$24", "$25", "$27", "$28", "memory");
+ }
+}
+
+static inline void
+cpu_relax (void)
+{
+ __asm volatile ("" : : : "memory");
+}
+
+static inline void
+atomic_write_barrier (void)
+{
+ __asm volatile ("wmb" : : : "memory");
}
--- libgomp/config/linux/affinity.c.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/affinity.c 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2006, 2007 Free Software Foundation, Inc.
+/* Copyright (C) 2006, 2007, 2008 Free Software Foundation, Inc.
Contributed by Jakub Jelinek <jakub@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -38,9 +38,6 @@
#ifdef HAVE_PTHREAD_AFFINITY_NP
static unsigned int affinity_counter;
-#ifndef HAVE_SYNC_BUILTINS
-static gomp_mutex_t affinity_lock;
-#endif
void
gomp_init_affinity (void)
@@ -76,9 +73,6 @@ gomp_init_affinity (void)
CPU_SET (gomp_cpu_affinity[0], &cpuset);
pthread_setaffinity_np (pthread_self (), sizeof (cpuset), &cpuset);
affinity_counter = 1;
-#ifndef HAVE_SYNC_BUILTINS
- gomp_mutex_init (&affinity_lock);
-#endif
}
void
@@ -87,13 +81,7 @@ gomp_init_thread_affinity (pthread_attr_
unsigned int cpu;
cpu_set_t cpuset;
-#ifdef HAVE_SYNC_BUILTINS
cpu = __sync_fetch_and_add (&affinity_counter, 1);
-#else
- gomp_mutex_lock (&affinity_lock);
- cpu = affinity_counter++;
- gomp_mutex_unlock (&affinity_lock);
-#endif
cpu %= gomp_cpu_affinity_len;
CPU_ZERO (&cpuset);
CPU_SET (gomp_cpu_affinity[cpu], &cpuset);
--- libgomp/config/linux/bar.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/bar.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -36,40 +36,49 @@
typedef struct
{
- gomp_mutex_t mutex;
- unsigned total;
- unsigned arrived;
- int generation;
+ /* Make sure total/generation is in a mostly read cacheline, while
+ awaited in a separate cacheline. */
+ unsigned total __attribute__((aligned (64)));
+ unsigned generation;
+ unsigned awaited __attribute__((aligned (64)));
} gomp_barrier_t;
+typedef unsigned int gomp_barrier_state_t;
static inline void gomp_barrier_init (gomp_barrier_t *bar, unsigned count)
{
- gomp_mutex_init (&bar->mutex);
bar->total = count;
- bar->arrived = 0;
+ bar->awaited = count;
bar->generation = 0;
}
static inline void gomp_barrier_reinit (gomp_barrier_t *bar, unsigned count)
{
- gomp_mutex_lock (&bar->mutex);
+ __sync_fetch_and_add (&bar->awaited, count - bar->total);
bar->total = count;
- gomp_mutex_unlock (&bar->mutex);
}
static inline void gomp_barrier_destroy (gomp_barrier_t *bar)
{
- /* Before destroying, make sure all threads have left the barrier. */
- gomp_mutex_lock (&bar->mutex);
}
extern void gomp_barrier_wait (gomp_barrier_t *);
-extern void gomp_barrier_wait_end (gomp_barrier_t *, bool);
+extern void gomp_barrier_wait_last (gomp_barrier_t *);
+extern void gomp_barrier_wait_end (gomp_barrier_t *, gomp_barrier_state_t);
-static inline bool gomp_barrier_wait_start (gomp_barrier_t *bar)
+static inline gomp_barrier_state_t
+gomp_barrier_wait_start (gomp_barrier_t *bar)
{
- gomp_mutex_lock (&bar->mutex);
- return ++bar->arrived == bar->total;
+ unsigned int ret = bar->generation;
+ /* Do we need any barrier here or is __sync_add_and_fetch acting
+ as the needed LoadLoad barrier already? */
+ ret += __sync_add_and_fetch (&bar->awaited, -1) == 0;
+ return ret;
+}
+
+static inline bool
+gomp_barrier_last_thread (gomp_barrier_state_t state)
+{
+ return state & 1;
}
#endif /* GOMP_BARRIER_H */
--- libgomp/config/linux/ptrlock.h.jj 2008-03-26 15:11:32.000000000 +0100
+++ libgomp/config/linux/ptrlock.h 2008-03-26 15:11:32.000000000 +0100
@@ -0,0 +1,65 @@
+/* Copyright (C) 2008 Free Software Foundation, Inc.
+ Contributed by Jakub Jelinek <jakub@redhat.com>.
+
+ This file is part of the GNU OpenMP Library (libgomp).
+
+ Libgomp is free software; you can redistribute it and/or modify it
+ under the terms of the GNU Lesser General Public License as published by
+ the Free Software Foundation; either version 2.1 of the License, 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 Lesser General Public License for
+ more details.
+
+ You should have received a copy of the GNU Lesser General Public License
+ along with libgomp; see the file COPYING.LIB. If not, write to the
+ Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
+ MA 02110-1301, USA. */
+
+/* As a special exception, if you link this library with other files, some
+ of which are compiled with GCC, to produce an executable, this library
+ does not by itself cause the resulting executable to be covered by the
+ GNU General Public License. This exception does not however invalidate
+ any other reasons why the executable file might be covered by the GNU
+ General Public License. */
+
+/* This is a Linux specific implementation of a mutex synchronization
+ mechanism for libgomp. This type is private to the library. This
+ implementation uses atomic instructions and the futex syscall. */
+
+#ifndef GOMP_PTRLOCK_H
+#define GOMP_PTRLOCK_H 1
+
+typedef void *gomp_ptrlock_t;
+
+static inline void gomp_ptrlock_init (gomp_ptrlock_t *ptrlock, void *ptr)
+{
+ *ptrlock = ptr;
+}
+
+extern void *gomp_ptrlock_get_slow (gomp_ptrlock_t *ptrlock);
+static inline void *gomp_ptrlock_get (gomp_ptrlock_t *ptrlock)
+{
+ if ((uintptr_t) *ptrlock > 2)
+ return *ptrlock;
+
+ if (__sync_bool_compare_and_swap (ptrlock, NULL, (uintptr_t) 1))
+ return NULL;
+
+ return gomp_ptrlock_get_slow (ptrlock);
+}
+
+extern void gomp_ptrlock_set_slow (gomp_ptrlock_t *ptrlock, void *ptr);
+static inline void gomp_ptrlock_set (gomp_ptrlock_t *ptrlock, void *ptr)
+{
+ if (!__sync_bool_compare_and_swap (ptrlock, (uintptr_t) 1, ptr))
+ gomp_ptrlock_set_slow (ptrlock, ptr);
+}
+
+static inline void gomp_ptrlock_destroy (gomp_ptrlock_t *ptrlock)
+{
+}
+
+#endif /* GOMP_PTRLOCK_H */
--- libgomp/config/linux/lock.c.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/lock.c 2008-03-26 15:11:32.000000000 +0100
@@ -29,11 +29,10 @@
primitives. This implementation uses atomic instructions and the futex
syscall. */
-#include "libgomp.h"
#include <string.h>
#include <unistd.h>
#include <sys/syscall.h>
-#include "futex.h"
+#include "wait.h"
/* The internal gomp_mutex_t and the external non-recursive omp_lock_t
@@ -137,7 +136,7 @@ omp_set_nest_lock (omp_nest_lock_t *lock
return;
}
- futex_wait (&lock->owner, otid);
+ do_wait (&lock->owner, otid);
}
}
--- libgomp/config/linux/ptrlock.c.jj 2008-03-26 15:11:32.000000000 +0100
+++ libgomp/config/linux/ptrlock.c 2008-03-26 15:11:32.000000000 +0100
@@ -0,0 +1,70 @@
+/* Copyright (C) 2008 Free Software Foundation, Inc.
+ Contributed by Jakub Jelinek <jakub@redhat.com>.
+
+ This file is part of the GNU OpenMP Library (libgomp).
+
+ Libgomp is free software; you can redistribute it and/or modify it
+ under the terms of the GNU Lesser General Public License as published by
+ the Free Software Foundation; either version 2.1 of the License, 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 Lesser General Public License for
+ more details.
+
+ You should have received a copy of the GNU Lesser General Public License
+ along with libgomp; see the file COPYING.LIB. If not, write to the
+ Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
+ MA 02110-1301, USA. */
+
+/* As a special exception, if you link this library with other files, some
+ of which are compiled with GCC, to produce an executable, this library
+ does not by itself cause the resulting executable to be covered by the
+ GNU General Public License. This exception does not however invalidate
+ any other reasons why the executable file might be covered by the GNU
+ General Public License. */
+
+/* This is a Linux specific implementation of a mutex synchronization
+ mechanism for libgomp. This type is private to the library. This
+ implementation uses atomic instructions and the futex syscall. */
+
+#include <endian.h>
+#include <limits.h>
+#include "wait.h"
+
+void *
+gomp_ptrlock_get_slow (gomp_ptrlock_t *ptrlock)
+{
+ int *intptr;
+ __sync_bool_compare_and_swap (ptrlock, 1, 2);
+
+ /* futex works on ints, not pointers.
+ But a valid work share pointer will be at least
+ 8 byte aligned, so it is safe to assume the low
+ 32-bits of the pointer won't contain values 1 or 2. */
+ __asm volatile ("" : "=r" (intptr) : "0" (ptrlock));
+#if __BYTE_ORDER == __BIG_ENDIAN
+ if (sizeof (*ptrlock) > sizeof (int))
+ intptr += (sizeof (*ptrlock) / sizeof (int)) - 1;
+#endif
+ do
+ do_wait (intptr, 2);
+ while (*intptr == 2);
+ __asm volatile ("" : : : "memory");
+ return *ptrlock;
+}
+
+void
+gomp_ptrlock_set_slow (gomp_ptrlock_t *ptrlock, void *ptr)
+{
+ int *intptr;
+
+ *ptrlock = ptr;
+ __asm volatile ("" : "=r" (intptr) : "0" (ptrlock));
+#if __BYTE_ORDER == __BIG_ENDIAN
+ if (sizeof (*ptrlock) > sizeof (int))
+ intptr += (sizeof (*ptrlock) / sizeof (int)) - 1;
+#endif
+ futex_wake (intptr, INT_MAX);
+}
--- libgomp/config/linux/x86/futex.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/x86/futex.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -27,9 +27,6 @@
/* Provide target-specific access to the futex system call. */
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-
#ifdef __LP64__
# ifndef SYS_futex
# define SYS_futex 202
@@ -38,14 +35,26 @@
static inline void
futex_wait (int *addr, int val)
{
- register long r10 __asm__("%r10") = 0;
+ register long r10 __asm__("%r10");
long res;
+ r10 = 0;
__asm volatile ("syscall"
: "=a" (res)
- : "0"(SYS_futex), "D" (addr), "S"(FUTEX_WAIT),
- "d"(val), "r"(r10)
+ : "0" (SYS_futex), "D" (addr), "S" (gomp_futex_wait),
+ "d" (val), "r" (r10)
: "r11", "rcx", "memory");
+ if (__builtin_expect (res == -ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ r10 = 0;
+ __asm volatile ("syscall"
+ : "=a" (res)
+ : "0" (SYS_futex), "D" (addr), "S" (gomp_futex_wait),
+ "d" (val), "r" (r10)
+ : "r11", "rcx", "memory");
+ }
}
static inline void
@@ -55,8 +64,19 @@ futex_wake (int *addr, int count)
__asm volatile ("syscall"
: "=a" (res)
- : "0"(SYS_futex), "D" (addr), "S"(FUTEX_WAKE), "d"(count)
+ : "0" (SYS_futex), "D" (addr), "S" (gomp_futex_wake),
+ "d" (count)
: "r11", "rcx", "memory");
+ if (__builtin_expect (res == -ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ __asm volatile ("syscall"
+ : "=a" (res)
+ : "0" (SYS_futex), "D" (addr), "S" (gomp_futex_wake),
+ "d" (count)
+ : "r11", "rcx", "memory");
+ }
}
#else
# ifndef SYS_futex
@@ -65,7 +85,7 @@ futex_wake (int *addr, int count)
# ifdef __PIC__
-static inline void
+static inline long
sys_futex0 (int *addr, int op, int val)
{
long res;
@@ -77,11 +97,12 @@ sys_futex0 (int *addr, int op, int val)
: "0"(SYS_futex), "r" (addr), "c"(op),
"d"(val), "S"(0)
: "memory");
+ return res;
}
# else
-static inline void
+static inline long
sys_futex0 (int *addr, int op, int val)
{
long res;
@@ -91,6 +112,7 @@ sys_futex0 (int *addr, int op, int val)
: "0"(SYS_futex), "b" (addr), "c"(op),
"d"(val), "S"(0)
: "memory");
+ return res;
}
# endif /* __PIC__ */
@@ -98,13 +120,37 @@ sys_futex0 (int *addr, int op, int val)
static inline void
futex_wait (int *addr, int val)
{
- sys_futex0 (addr, FUTEX_WAIT, val);
+ long res = sys_futex0 (addr, gomp_futex_wait, val);
+ if (__builtin_expect (res == -ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wait, val);
+ }
}
static inline void
futex_wake (int *addr, int count)
{
- sys_futex0 (addr, FUTEX_WAKE, count);
+ long res = sys_futex0 (addr, gomp_futex_wake, count);
+ if (__builtin_expect (res == -ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wake, count);
+ }
}
#endif /* __LP64__ */
+
+static inline void
+cpu_relax (void)
+{
+ __asm volatile ("rep; nop" : : : "memory");
+}
+
+static inline void
+atomic_write_barrier (void)
+{
+ __sync_synchronize ();
+}
--- libgomp/config/linux/wait.h.jj 2008-03-26 15:11:32.000000000 +0100
+++ libgomp/config/linux/wait.h 2008-03-26 15:11:32.000000000 +0100
@@ -0,0 +1,68 @@
+/* Copyright (C) 2008 Free Software Foundation, Inc.
+ Contributed by Jakub Jelinek <jakub@redhat.com>.
+
+ This file is part of the GNU OpenMP Library (libgomp).
+
+ Libgomp is free software; you can redistribute it and/or modify it
+ under the terms of the GNU Lesser General Public License as published by
+ the Free Software Foundation; either version 2.1 of the License, 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 Lesser General Public License for
+ more details.
+
+ You should have received a copy of the GNU Lesser General Public License
+ along with libgomp; see the file COPYING.LIB. If not, write to the
+ Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
+ MA 02110-1301, USA. */
+
+/* As a special exception, if you link this library with other files, some
+ of which are compiled with GCC, to produce an executable, this library
+ does not by itself cause the resulting executable to be covered by the
+ GNU General Public License. This exception does not however invalidate
+ any other reasons why the executable file might be covered by the GNU
+ General Public License. */
+
+/* This is a Linux specific implementation of a mutex synchronization
+ mechanism for libgomp. This type is private to the library. This
+ implementation uses atomic instructions and the futex syscall. */
+
+#ifndef GOMP_WAIT_H
+#define GOMP_WAIT_H 1
+
+#include "libgomp.h"
+#include <errno.h>
+
+#define FUTEX_WAIT 0
+#define FUTEX_WAKE 1
+#define FUTEX_PRIVATE_FLAG 128L
+
+#ifdef HAVE_ATTRIBUTE_VISIBILITY
+# pragma GCC visibility push(hidden)
+#endif
+
+extern long int gomp_futex_wait, gomp_futex_wake;
+
+#include "futex.h"
+
+static inline void do_wait (int *addr, int val)
+{
+ unsigned long long i, count = gomp_spin_count_var;
+
+ if (__builtin_expect (gomp_managed_threads > gomp_available_cpus, 0))
+ count = gomp_throttled_spin_count_var;
+ for (i = 0; i < count; i++)
+ if (__builtin_expect (*addr != val, 0))
+ return;
+ else
+ cpu_relax ();
+ futex_wait (addr, val);
+}
+
+#ifdef HAVE_ATTRIBUTE_VISIBILITY
+# pragma GCC visibility pop
+#endif
+
+#endif /* GOMP_WAIT_H */
--- libgomp/config/linux/sparc/futex.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/sparc/futex.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Jakub Jelinek <jakub@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -28,10 +28,8 @@
/* Provide target-specific access to the futex system call. */
#include <sys/syscall.h>
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-static inline void
+static inline long
sys_futex0 (int *addr, int op, int val)
{
register long int g1 __asm__ ("g1");
@@ -47,9 +45,9 @@ sys_futex0 (int *addr, int op, int val)
o3 = 0;
#ifdef __arch64__
-# define SYSCALL_STRING "ta\t0x6d"
+# define SYSCALL_STRING "ta\t0x6d; bcs,a,pt %%xcc, 1f; sub %%g0, %%o0, %%o0; 1:"
#else
-# define SYSCALL_STRING "ta\t0x10"
+# define SYSCALL_STRING "ta\t0x10; bcs,a 1f; sub %%g0, %%o0, %%o0; 1:"
#endif
__asm volatile (SYSCALL_STRING
@@ -65,16 +63,49 @@ sys_futex0 (int *addr, int op, int val)
"f48", "f50", "f52", "f54", "f56", "f58", "f60", "f62",
#endif
"cc", "memory");
+ return o0;
}
static inline void
futex_wait (int *addr, int val)
{
- sys_futex0 (addr, FUTEX_WAIT, val);
+ long err = sys_futex0 (addr, gomp_futex_wait, val);
+ if (__builtin_expect (err == ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wait, val);
+ }
}
static inline void
futex_wake (int *addr, int count)
{
- sys_futex0 (addr, FUTEX_WAKE, count);
+ long err = sys_futex0 (addr, gomp_futex_wake, count);
+ if (__builtin_expect (err == ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wake, count);
+ }
+}
+
+static inline void
+cpu_relax (void)
+{
+#if defined __arch64__ || defined __sparc_v9__
+ __asm volatile ("membar #LoadLoad" : : : "memory");
+#else
+ __asm volatile ("" : : : "memory");
+#endif
+}
+
+static inline void
+atomic_write_barrier (void)
+{
+#if defined __arch64__ || defined __sparc_v9__
+ __asm volatile ("membar #WriteWrite" : : : "memory");
+#else
+ __sync_synchronize ();
+#endif
}
--- libgomp/config/linux/ia64/futex.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/ia64/futex.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -29,23 +29,24 @@
#include <sys/syscall.h>
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-static inline void
-sys_futex0(int *addr, int op, int val)
+static inline long
+sys_futex0(int *addr, long op, int val)
{
register long out0 asm ("out0") = (long) addr;
register long out1 asm ("out1") = op;
register long out2 asm ("out2") = val;
register long out3 asm ("out3") = 0;
+ register long r8 asm ("r8");
+ register long r10 asm ("r10");
register long r15 asm ("r15") = SYS_futex;
__asm __volatile ("break 0x100000"
- : "=r"(r15), "=r"(out0), "=r"(out1), "=r"(out2), "=r"(out3)
+ : "=r"(r15), "=r"(out0), "=r"(out1), "=r"(out2), "=r"(out3),
+ "=r"(r8), "=r"(r10)
: "r"(r15), "r"(out0), "r"(out1), "r"(out2), "r"(out3)
- : "memory", "r8", "r10", "out4", "out5", "out6", "out7",
+ : "memory", "out4", "out5", "out6", "out7",
/* Non-stacked integer registers, minus r8, r10, r15. */
"r2", "r3", "r9", "r11", "r12", "r13", "r14", "r16", "r17", "r18",
"r19", "r20", "r21", "r22", "r23", "r24", "r25", "r26", "r27",
@@ -56,16 +57,41 @@ sys_futex0(int *addr, int op, int val)
"f6", "f7", "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15",
/* Branch registers. */
"b6");
+ return r8 & r10;
}
static inline void
futex_wait (int *addr, int val)
{
- sys_futex0 (addr, FUTEX_WAIT, val);
+ long err = sys_futex0 (addr, gomp_futex_wait, val);
+ if (__builtin_expect (err == ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wait, val);
+ }
}
static inline void
futex_wake (int *addr, int count)
{
- sys_futex0 (addr, FUTEX_WAKE, count);
+ long err = sys_futex0 (addr, gomp_futex_wake, count);
+ if (__builtin_expect (err == ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wake, count);
+ }
+}
+
+static inline void
+cpu_relax (void)
+{
+ __asm volatile ("hint @pause" : : : "memory");
+}
+
+static inline void
+atomic_write_barrier (void)
+{
+ __sync_synchronize ();
}
--- libgomp/config/linux/s390/futex.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/s390/futex.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Jakub Jelinek <jakub@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -28,10 +28,8 @@
/* Provide target-specific access to the futex system call. */
#include <sys/syscall.h>
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-static inline void
+static inline long
sys_futex0 (int *addr, int op, int val)
{
register long int gpr2 __asm__ ("2");
@@ -49,16 +47,41 @@ sys_futex0 (int *addr, int op, int val)
: "i" (SYS_futex),
"0" (gpr2), "d" (gpr3), "d" (gpr4), "d" (gpr5)
: "memory");
+ return gpr2;
}
static inline void
futex_wait (int *addr, int val)
{
- sys_futex0 (addr, FUTEX_WAIT, val);
+ long err = sys_futex0 (addr, gomp_futex_wait, val);
+ if (__builtin_expect (err == -ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wait, val);
+ }
}
static inline void
futex_wake (int *addr, int count)
{
- sys_futex0 (addr, FUTEX_WAKE, count);
+ long err = sys_futex0 (addr, gomp_futex_wake, count);
+ if (__builtin_expect (err == -ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wake, count);
+ }
+}
+
+static inline void
+cpu_relax (void)
+{
+ __asm volatile ("" : : : "memory");
+}
+
+static inline void
+atomic_write_barrier (void)
+{
+ __sync_synchronize ();
}
--- libgomp/config/linux/mutex.c.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/mutex.c 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -29,9 +29,10 @@
mechanism for libgomp. This type is private to the library. This
implementation uses atomic instructions and the futex syscall. */
-#include "libgomp.h"
-#include "futex.h"
+#include "wait.h"
+long int gomp_futex_wake = FUTEX_WAKE | FUTEX_PRIVATE_FLAG;
+long int gomp_futex_wait = FUTEX_WAIT | FUTEX_PRIVATE_FLAG;
void
gomp_mutex_lock_slow (gomp_mutex_t *mutex)
@@ -40,7 +41,7 @@ gomp_mutex_lock_slow (gomp_mutex_t *mute
{
int oldval = __sync_val_compare_and_swap (mutex, 1, 2);
if (oldval != 0)
- futex_wait (mutex, 2);
+ do_wait (mutex, 2);
}
while (!__sync_bool_compare_and_swap (mutex, 0, 2));
}
--- libgomp/config/linux/sem.c.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/sem.c 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -29,8 +29,7 @@
mechanism for libgomp. This type is private to the library. This
implementation uses atomic instructions and the futex syscall. */
-#include "libgomp.h"
-#include "futex.h"
+#include "wait.h"
void
@@ -44,7 +43,7 @@ gomp_sem_wait_slow (gomp_sem_t *sem)
if (__sync_bool_compare_and_swap (sem, val, val - 1))
return;
}
- futex_wait (sem, -1);
+ do_wait (sem, -1);
}
}
--- libgomp/config/linux/powerpc/futex.h.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/powerpc/futex.h 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -28,10 +28,8 @@
/* Provide target-specific access to the futex system call. */
#include <sys/syscall.h>
-#define FUTEX_WAIT 0
-#define FUTEX_WAKE 1
-static inline void
+static inline long
sys_futex0 (int *addr, int op, int val)
{
register long int r0 __asm__ ("r0");
@@ -50,21 +48,48 @@ sys_futex0 (int *addr, int op, int val)
doesn't. It doesn't much matter for us. In the interest of unity,
go ahead and clobber it always. */
- __asm volatile ("sc"
+ __asm volatile ("sc; mfcr %0"
: "=r"(r0), "=r"(r3), "=r"(r4), "=r"(r5), "=r"(r6)
: "r"(r0), "r"(r3), "r"(r4), "r"(r5), "r"(r6)
: "r7", "r8", "r9", "r10", "r11", "r12",
"cr0", "ctr", "memory");
+ if (__builtin_expect (r0 & (1 << 28), 0))
+ return r3;
+ return 0;
}
static inline void
futex_wait (int *addr, int val)
{
- sys_futex0 (addr, FUTEX_WAIT, val);
+ long err = sys_futex0 (addr, gomp_futex_wait, val);
+ if (__builtin_expect (err == ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wait, val);
+ }
}
static inline void
futex_wake (int *addr, int count)
{
- sys_futex0 (addr, FUTEX_WAKE, count);
+ long err = sys_futex0 (addr, gomp_futex_wake, count);
+ if (__builtin_expect (err == ENOSYS, 0))
+ {
+ gomp_futex_wait &= ~FUTEX_PRIVATE_FLAG;
+ gomp_futex_wake &= ~FUTEX_PRIVATE_FLAG;
+ sys_futex0 (addr, gomp_futex_wake, count);
+ }
+}
+
+static inline void
+cpu_relax (void)
+{
+ __asm volatile ("" : : : "memory");
+}
+
+static inline void
+atomic_write_barrier (void)
+{
+ __asm volatile ("eieio" : : : "memory");
}
--- libgomp/config/linux/bar.c.jj 2007-12-07 14:41:00.000000000 +0100
+++ libgomp/config/linux/bar.c 2008-03-26 15:11:32.000000000 +0100
@@ -1,4 +1,4 @@
-/* Copyright (C) 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2005, 2008 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU OpenMP Library (libgomp).
@@ -29,32 +29,29 @@
mechanism for libgomp. This type is private to the library. This
implementation uses atomic instructions and the futex syscall. */
-#include "libgomp.h"
-#include "futex.h"
#include <limits.h>
+#include "wait.h"
void
-gomp_barrier_wait_end (gomp_barrier_t *bar, bool last)
+gomp_barrier_wait_end (gomp_barrier_t *bar, gomp_barrier_state_t state)
{
- if (last)
+ if (__builtin_expect ((state & 1) != 0, 0))
{
- bar->generation++;
- futex_wake (&bar->generation, INT_MAX);
+ /* Next time we'll be awaiting TOTAL threads again. */
+ bar->awaited = bar->total;
+ atomic_write_barrier ();
+ bar->generation += 2;
+ futex_wake ((int *) &bar->generation, INT_MAX);
}
else
{
- unsigned int generation = bar->generation;
-
- gomp_mutex_unlock (&bar->mutex);
+ unsigned int generation = state;
do
- futex_wait (&bar->generation, generation);
+ do_wait ((int *) &bar->generation, generation);
while (bar->generation == generation);
}
-
- if (__sync_add_and_fetch (&bar->arrived, -1) == 0)
- gomp_mutex_unlock (&bar->mutex);
}
void
@@ -62,3 +59,18 @@ gomp_barrier_wait (gomp_barrier_t *barri
{
gomp_barrier_wait_end (barrier, gomp_barrier_wait_start (barrier));
}
+
+/* Like gomp_barrier_wait, except that if the encountering thread
+ is not the last one to hit the barrier, it returns immediately.
+ The intended usage is that a thread which intends to gomp_barrier_destroy
+ this barrier calls gomp_barrier_wait, while all other threads
+ call gomp_barrier_wait_last. When gomp_barrier_wait returns,
+ the barrier can be safely destroyed. */
+
+void
+gomp_barrier_wait_last (gomp_barrier_t *barrier)
+{
+ gomp_barrier_state_t state = gomp_barrier_wait_start (barrier);
+ if (state & 1)
+ gomp_barrier_wait_end (barrier, state);
+}