aboutsummaryrefslogtreecommitdiffstats
path: root/runtime/src/kmp_csupport.cpp
diff options
context:
space:
mode:
authorDimitry Andric <dim@FreeBSD.org>2019-08-20 20:52:04 +0000
committerDimitry Andric <dim@FreeBSD.org>2019-08-20 20:52:04 +0000
commit1756886afc900127720de68088f9adff3108ef61 (patch)
treed4c46dc43cbf2f632d3c575a6d01716370a7b45b /runtime/src/kmp_csupport.cpp
parent4254a3821b2072ba01f8f7e3347ac862d324babc (diff)
downloadsrc-1756886afc900127720de68088f9adff3108ef61.tar.gz
src-1756886afc900127720de68088f9adff3108ef61.zip
Notes
Notes: svn path=/vendor/llvm-openmp/dist/; revision=351292 svn path=/vendor/llvm-openmp/openmp-release_900-r372316/; revision=352532; tag=vendor/llvm-openmp/openmp-release_900-r372316
Diffstat (limited to 'runtime/src/kmp_csupport.cpp')
-rw-r--r--runtime/src/kmp_csupport.cpp269
1 files changed, 146 insertions, 123 deletions
diff --git a/runtime/src/kmp_csupport.cpp b/runtime/src/kmp_csupport.cpp
index 61d4a9301120..c778c97022f5 100644
--- a/runtime/src/kmp_csupport.cpp
+++ b/runtime/src/kmp_csupport.cpp
@@ -4,10 +4,9 @@
//===----------------------------------------------------------------------===//
//
-// The LLVM Compiler Infrastructure
-//
-// This file is dual licensed under the MIT and the University of Illinois Open
-// Source Licenses. See LICENSE.txt for details.
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
@@ -245,8 +244,6 @@ void __kmpc_pop_num_threads(ident_t *loc, kmp_int32 global_tid) {
/* the num_threads are automatically popped */
}
-#if OMP_40_ENABLED
-
void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 global_tid,
kmp_int32 proc_bind) {
KA_TRACE(20, ("__kmpc_push_proc_bind: enter T#%d proc_bind=%d\n", global_tid,
@@ -255,8 +252,6 @@ void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 global_tid,
__kmp_push_proc_bind(loc, global_tid, (kmp_proc_bind_t)proc_bind);
}
-#endif /* OMP_40_ENABLED */
-
/*!
@ingroup PARALLEL
@param loc source location information
@@ -345,7 +340,6 @@ void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, kmpc_micro microtask, ...) {
#endif // KMP_STATS_ENABLED
}
-#if OMP_40_ENABLED
/*!
@ingroup PARALLEL
@param loc source location information
@@ -383,7 +377,15 @@ void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,
va_list ap;
va_start(ap, microtask);
+#if KMP_STATS_ENABLED
KMP_COUNT_BLOCK(OMP_TEAMS);
+ stats_state_e previous_state = KMP_GET_THREAD_STATE();
+ if (previous_state == stats_state_e::SERIAL_REGION) {
+ KMP_EXCHANGE_PARTITIONED_TIMER(OMP_teams_overhead);
+ } else {
+ KMP_PUSH_PARTITIONED_TIMER(OMP_teams_overhead);
+ }
+#endif
// remember teams entry point and nesting level
this_thr->th.th_teams_microtask = microtask;
@@ -426,12 +428,35 @@ void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,
#endif
);
+ // Pop current CG root off list
+ KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
+ kmp_cg_root_t *tmp = this_thr->th.th_cg_roots;
+ this_thr->th.th_cg_roots = tmp->up;
+ KA_TRACE(100, ("__kmpc_fork_teams: Thread %p popping node %p and moving up"
+ " to node %p. cg_nthreads was %d\n",
+ this_thr, tmp, this_thr->th.th_cg_roots, tmp->cg_nthreads));
+ KMP_DEBUG_ASSERT(tmp->cg_nthreads);
+ int i = tmp->cg_nthreads--;
+ if (i == 1) { // check is we are the last thread in CG (not always the case)
+ __kmp_free(tmp);
+ }
+ // Restore current task's thread_limit from CG root
+ KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
+ this_thr->th.th_current_task->td_icvs.thread_limit =
+ this_thr->th.th_cg_roots->cg_thread_limit;
+
this_thr->th.th_teams_microtask = NULL;
this_thr->th.th_teams_level = 0;
*(kmp_int64 *)(&this_thr->th.th_teams_size) = 0L;
va_end(ap);
+#if KMP_STATS_ENABLED
+ if (previous_state == stats_state_e::SERIAL_REGION) {
+ KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
+ } else {
+ KMP_POP_PARTITIONED_TIMER();
+ }
+#endif // KMP_STATS_ENABLED
}
-#endif /* OMP_40_ENABLED */
// I don't think this function should ever have been exported.
// The __kmpc_ prefix was misapplied. I'm fairly certain that no generated
@@ -485,16 +510,15 @@ void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid) {
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
this_thr = __kmp_threads[global_tid];
serial_team = this_thr->th.th_serial_team;
-#if OMP_45_ENABLED
kmp_task_team_t *task_team = this_thr->th.th_task_team;
-
// we need to wait for the proxy tasks before finishing the thread
if (task_team != NULL && task_team->tt.tt_found_proxy_tasks)
__kmp_task_team_wait(this_thr, serial_team USE_ITT_BUILD_ARG(NULL));
-#endif
KMP_MB();
KMP_DEBUG_ASSERT(serial_team);
@@ -549,9 +573,7 @@ void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid) {
serial_team->t.t_dispatch->th_disp_buffer->next;
__kmp_free(disp_buffer);
}
-#if OMP_50_ENABLED
this_thr->th.th_def_allocator = serial_team->t.t_def_allocator; // restore
-#endif
--serial_team->t.t_serialized;
if (serial_team->t.t_serialized == 0) {
@@ -667,7 +689,7 @@ void __kmpc_flush(ident_t *loc) {
// }
// and adding the yield here is good for at least a 10x speedup
// when running >2 threads per core (on the NAS LU benchmark).
- __kmp_yield(TRUE);
+ __kmp_yield();
#endif
#else
#error Unknown or unsupported architecture
@@ -696,11 +718,12 @@ void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid) {
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
if (__kmp_env_consistency_check) {
if (loc == 0) {
KMP_WARNING(ConstructIdentInvalid); // ??? What does it mean for the user?
}
-
__kmp_check_barrier(global_tid, ct_barrier, loc);
}
@@ -744,6 +767,8 @@ kmp_int32 __kmpc_master(ident_t *loc, kmp_int32 global_tid) {
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
if (KMP_MASTER_GTID(global_tid)) {
KMP_COUNT_BLOCK(OMP_MASTER);
KMP_PUSH_PARTITIONED_TIMER(OMP_master);
@@ -834,6 +859,8 @@ void __kmpc_ordered(ident_t *loc, kmp_int32 gtid) {
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
#if USE_ITT_BUILD
__kmp_itt_ordered_prep(gtid);
// TODO: ordered_wait_id
@@ -848,7 +875,7 @@ void __kmpc_ordered(ident_t *loc, kmp_int32 gtid) {
if (ompt_enabled.enabled) {
OMPT_STORE_RETURN_ADDRESS(gtid);
team = __kmp_team_from_gtid(gtid);
- lck = (ompt_wait_id_t)&team->t.t_ordered.dt.t_value;
+ lck = (ompt_wait_id_t)(uintptr_t)&team->t.t_ordered.dt.t_value;
/* OMPT state update */
th->th.ompt_thread_info.wait_id = lck;
th->th.ompt_thread_info.state = ompt_state_wait_ordered;
@@ -857,8 +884,8 @@ void __kmpc_ordered(ident_t *loc, kmp_int32 gtid) {
codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
- ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin,
- (ompt_wait_id_t)lck, codeptr_ra);
+ ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin, lck,
+ codeptr_ra);
}
}
#endif
@@ -877,7 +904,7 @@ void __kmpc_ordered(ident_t *loc, kmp_int32 gtid) {
/* OMPT event callback */
if (ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_ordered, (ompt_wait_id_t)lck, codeptr_ra);
+ ompt_mutex_ordered, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
}
}
#endif
@@ -917,7 +944,8 @@ void __kmpc_end_ordered(ident_t *loc, kmp_int32 gtid) {
if (ompt_enabled.ompt_callback_mutex_released) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
ompt_mutex_ordered,
- (ompt_wait_id_t)&__kmp_team_from_gtid(gtid)->t.t_ordered.dt.t_value,
+ (ompt_wait_id_t)(uintptr_t)&__kmp_team_from_gtid(gtid)
+ ->t.t_ordered.dt.t_value,
OMPT_LOAD_RETURN_ADDRESS(gtid));
}
#endif
@@ -965,24 +993,18 @@ __kmp_init_indirect_csptr(kmp_critical_name *crit, ident_t const *loc,
kmp_uint32 spins; \
KMP_FSYNC_PREPARE(l); \
KMP_INIT_YIELD(spins); \
- if (TCR_4(__kmp_nth) > \
- (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
- KMP_YIELD(TRUE); \
- } else { \
- KMP_YIELD_SPIN(spins); \
- } \
kmp_backoff_t backoff = __kmp_spin_backoff_params; \
- while ( \
- KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
- !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \
- __kmp_spin_backoff(&backoff); \
+ do { \
if (TCR_4(__kmp_nth) > \
(__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
KMP_YIELD(TRUE); \
} else { \
KMP_YIELD_SPIN(spins); \
} \
- } \
+ __kmp_spin_backoff(&backoff); \
+ } while ( \
+ KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
+ !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)); \
} \
KMP_FSYNC_ACQUIRED(l); \
}
@@ -1068,8 +1090,7 @@ __kmp_init_indirect_csptr(kmp_critical_name *crit, ident_t const *loc,
KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \
} \
KMP_MB(); \
- KMP_YIELD(TCR_4(__kmp_nth) > \
- (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)); \
+ KMP_YIELD_OVERSUB(); \
}
#endif // KMP_USE_FUTEX
@@ -1188,7 +1209,7 @@ void __kmpc_critical(ident_t *loc, kmp_int32 global_tid,
ti = __kmp_threads[global_tid]->th.ompt_thread_info;
/* OMPT state update */
prev_state = ti.state;
- ti.wait_id = (ompt_wait_id_t)lck;
+ ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
ti.state = ompt_state_wait_critical;
/* OMPT event callback */
@@ -1196,7 +1217,7 @@ void __kmpc_critical(ident_t *loc, kmp_int32 global_tid,
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
- (ompt_wait_id_t)crit, codeptr_ra);
+ (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
}
}
#endif
@@ -1216,7 +1237,7 @@ void __kmpc_critical(ident_t *loc, kmp_int32 global_tid,
/* OMPT event callback */
if (ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_critical, (ompt_wait_id_t)crit, codeptr_ra);
+ ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
}
}
#endif
@@ -1402,14 +1423,15 @@ void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid,
ti = __kmp_threads[global_tid]->th.ompt_thread_info;
/* OMPT state update */
prev_state = ti.state;
- ti.wait_id = (ompt_wait_id_t)lck;
+ ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
ti.state = ompt_state_wait_critical;
/* OMPT event callback */
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_critical, (unsigned int)hint,
- __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)crit, codeptr);
+ __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)(uintptr_t)lck,
+ codeptr);
}
}
#endif
@@ -1440,14 +1462,15 @@ void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid,
ti = __kmp_threads[global_tid]->th.ompt_thread_info;
/* OMPT state update */
prev_state = ti.state;
- ti.wait_id = (ompt_wait_id_t)lck;
+ ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
ti.state = ompt_state_wait_critical;
/* OMPT event callback */
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_critical, (unsigned int)hint,
- __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)crit, codeptr);
+ __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)(uintptr_t)lck,
+ codeptr);
}
}
#endif
@@ -1467,7 +1490,7 @@ void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid,
/* OMPT event callback */
if (ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_critical, (ompt_wait_id_t)crit, codeptr);
+ ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
}
#endif
@@ -1565,7 +1588,8 @@ void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid,
OMPT_STORE_RETURN_ADDRESS(global_tid);
if (ompt_enabled.ompt_callback_mutex_released) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_critical, (ompt_wait_id_t)crit, OMPT_LOAD_RETURN_ADDRESS(0));
+ ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck,
+ OMPT_LOAD_RETURN_ADDRESS(0));
}
#endif
@@ -1590,6 +1614,8 @@ kmp_int32 __kmpc_barrier_master(ident_t *loc, kmp_int32 global_tid) {
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
if (__kmp_env_consistency_check)
__kmp_check_barrier(global_tid, ct_barrier, loc);
@@ -1648,6 +1674,8 @@ kmp_int32 __kmpc_barrier_master_nowait(ident_t *loc, kmp_int32 global_tid) {
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
if (__kmp_env_consistency_check) {
if (loc == 0) {
KMP_WARNING(ConstructIdentInvalid); // ??? What does it mean for the user?
@@ -1843,7 +1871,7 @@ void ompc_set_nested(int flag) {
__kmp_save_internal_controls(thread);
- set__nested(thread, flag ? TRUE : FALSE);
+ set__max_active_levels(thread, flag ? __kmp_dflt_max_active_levels : 1);
}
void ompc_set_max_active_levels(int max_active_levels) {
@@ -1867,7 +1895,6 @@ int ompc_get_team_size(int level) {
return __kmp_get_team_size(__kmp_entry_gtid(), level);
}
-#if OMP_50_ENABLED
/* OpenMP 5.0 Affinity Format API */
void ompc_set_affinity_format(char const *format) {
@@ -1918,7 +1945,6 @@ size_t ompc_capture_affinity(char *buffer, size_t buf_size,
__kmp_str_buf_free(&capture_buf);
return num_required;
}
-#endif /* OMP_50_ENABLED */
void kmpc_set_stacksize(int arg) {
// __kmp_aux_set_stacksize initializes the library if needed
@@ -2189,8 +2215,8 @@ void __kmpc_init_lock_with_hint(ident_t *loc, kmp_int32 gtid, void **user_lock,
if (ompt_enabled.ompt_callback_lock_init) {
ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
ompt_mutex_lock, (omp_lock_hint_t)hint,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
}
@@ -2213,8 +2239,8 @@ void __kmpc_init_nest_lock_with_hint(ident_t *loc, kmp_int32 gtid,
if (ompt_enabled.ompt_callback_lock_init) {
ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
ompt_mutex_nest_lock, (omp_lock_hint_t)hint,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
}
@@ -2239,8 +2265,8 @@ void __kmpc_init_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_lock_init) {
ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
ompt_mutex_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2282,7 +2308,7 @@ void __kmpc_init_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_lock_init) {
ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
- (ompt_wait_id_t)user_lock, codeptr);
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2311,8 +2337,8 @@ void __kmpc_init_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_lock_init) {
ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
ompt_mutex_nest_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2357,7 +2383,7 @@ void __kmpc_init_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_lock_init) {
ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
- (ompt_wait_id_t)user_lock, codeptr);
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2393,7 +2419,7 @@ void __kmpc_destroy_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
lck = (kmp_user_lock_p)user_lock;
}
ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
- ompt_mutex_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
@@ -2421,7 +2447,7 @@ void __kmpc_destroy_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
codeptr = OMPT_GET_RETURN_ADDRESS(0);
if (ompt_enabled.ompt_callback_lock_destroy) {
ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
- ompt_mutex_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2461,7 +2487,7 @@ void __kmpc_destroy_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
codeptr = OMPT_GET_RETURN_ADDRESS(0);
if (ompt_enabled.ompt_callback_lock_destroy) {
ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
@@ -2493,7 +2519,7 @@ void __kmpc_destroy_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
codeptr = OMPT_GET_RETURN_ADDRESS(0);
if (ompt_enabled.ompt_callback_lock_destroy) {
ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2538,8 +2564,8 @@ void __kmpc_set_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
#if KMP_USE_INLINED_TAS
@@ -2560,7 +2586,7 @@ void __kmpc_set_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
if (ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2593,7 +2619,7 @@ void __kmpc_set_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
- (ompt_wait_id_t)lck, codeptr);
+ (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
#endif
@@ -2606,7 +2632,7 @@ void __kmpc_set_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
if (ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
#endif
@@ -2628,8 +2654,8 @@ void __kmpc_set_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_nest_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
}
#endif
@@ -2646,13 +2672,14 @@ void __kmpc_set_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquired) {
// lock_first
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
+ codeptr);
}
} else {
if (ompt_enabled.ompt_callback_nest_lock) {
// lock_next
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_scope_begin, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
}
}
@@ -2690,7 +2717,8 @@ void __kmpc_set_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_nest_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(), (ompt_wait_id_t)lck, codeptr);
+ __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
+ codeptr);
}
}
#endif
@@ -2707,13 +2735,13 @@ void __kmpc_set_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquired) {
// lock_first
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
} else {
if (ompt_enabled.ompt_callback_nest_lock) {
// lock_next
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_scope_begin, (ompt_wait_id_t)lck, codeptr);
+ ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
}
}
@@ -2749,7 +2777,7 @@ void __kmpc_unset_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
codeptr = OMPT_GET_RETURN_ADDRESS(0);
if (ompt_enabled.ompt_callback_mutex_released) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
@@ -2778,7 +2806,7 @@ void __kmpc_unset_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
codeptr = OMPT_GET_RETURN_ADDRESS(0);
if (ompt_enabled.ompt_callback_mutex_released) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
#endif
@@ -2810,7 +2838,7 @@ void __kmpc_unset_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
codeptr = OMPT_GET_RETURN_ADDRESS(0);
if (ompt_enabled.ompt_callback_mutex_released) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
#endif
@@ -2838,12 +2866,13 @@ void __kmpc_unset_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_released) {
// release_lock_last
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
+ codeptr);
}
} else if (ompt_enabled.ompt_callback_nest_lock) {
// release_lock_prev
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_scope_end, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_scope_end, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
}
#endif
@@ -2887,12 +2916,12 @@ void __kmpc_unset_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_released) {
// release_lock_last
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
} else if (ompt_enabled.ompt_callback_nest_lock) {
// release_lock_previous
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_mutex_scope_end, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
}
#endif
@@ -2929,12 +2958,12 @@ void __kmpc_unset_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_released) {
// release_lock_last
ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
} else if (ompt_enabled.ompt_callback_nest_lock) {
// release_lock_previous
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_mutex_scope_end, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
}
#endif
@@ -2960,8 +2989,8 @@ int __kmpc_test_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
#if KMP_USE_INLINED_TAS
@@ -2983,7 +3012,7 @@ int __kmpc_test_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
if (ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
return FTN_TRUE;
@@ -3024,7 +3053,7 @@ int __kmpc_test_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
- (ompt_wait_id_t)lck, codeptr);
+ (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
#endif
@@ -3039,7 +3068,7 @@ int __kmpc_test_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
#if OMPT_SUPPORT && OMPT_OPTIONAL
if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
#endif
@@ -3065,8 +3094,8 @@ int __kmpc_test_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_nest_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(user_lock), (ompt_wait_id_t)user_lock,
- codeptr);
+ __ompt_get_mutex_impl_type(user_lock),
+ (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
#endif
rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
@@ -3083,13 +3112,14 @@ int __kmpc_test_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquired) {
// lock_first
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
+ codeptr);
}
} else {
if (ompt_enabled.ompt_callback_nest_lock) {
// lock_next
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_scope_begin, (ompt_wait_id_t)user_lock, codeptr);
+ ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
}
}
}
@@ -3130,7 +3160,8 @@ int __kmpc_test_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
ompt_enabled.ompt_callback_mutex_acquire) {
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
ompt_mutex_nest_lock, omp_lock_hint_none,
- __ompt_get_mutex_impl_type(), (ompt_wait_id_t)lck, codeptr);
+ __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
+ codeptr);
}
#endif
@@ -3148,13 +3179,13 @@ int __kmpc_test_nest_lock(ident_t *loc, kmp_int32 gtid, void **user_lock) {
if (ompt_enabled.ompt_callback_mutex_acquired) {
// lock_first
ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
- ompt_mutex_nest_lock, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
} else {
if (ompt_enabled.ompt_callback_nest_lock) {
// lock_next
ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
- ompt_mutex_scope_begin, (ompt_wait_id_t)lck, codeptr);
+ ompt_mutex_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
}
}
}
@@ -3288,7 +3319,6 @@ __kmp_end_critical_section_reduce_block(ident_t *loc, kmp_int32 global_tid,
#endif // KMP_USE_DYNAMIC_LOCK
} // __kmp_end_critical_section_reduce_block
-#if OMP_40_ENABLED
static __forceinline int
__kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
int *task_state) {
@@ -3323,7 +3353,6 @@ __kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team, int task_state) {
th->th.th_task_team = team->t.t_task_team[task_state];
th->th.th_task_state = task_state;
}
-#endif
/* 2.a.i. Reduce Block without a terminating barrier */
/*!
@@ -3350,11 +3379,9 @@ __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
KMP_COUNT_BLOCK(REDUCE_nowait);
int retval = 0;
PACKED_REDUCTION_METHOD_T packed_reduction_method;
-#if OMP_40_ENABLED
kmp_info_t *th;
kmp_team_t *team;
int teams_swapped = 0, task_state;
-#endif
KA_TRACE(10, ("__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
// why do we need this initialization here at all?
@@ -3366,6 +3393,8 @@ __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
// check correctness of reduce block nesting
#if KMP_USE_DYNAMIC_LOCK
if (__kmp_env_consistency_check)
@@ -3375,10 +3404,8 @@ __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
__kmp_push_sync(global_tid, ct_reduce, loc, NULL);
#endif
-#if OMP_40_ENABLED
th = __kmp_thread_from_gtid(global_tid);
teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
-#endif // OMP_40_ENABLED
// packed_reduction_method value will be reused by __kmp_end_reduce* function,
// the value should be kept in a variable
@@ -3479,11 +3506,9 @@ __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
// should never reach this block
KMP_ASSERT(0); // "unexpected method"
}
-#if OMP_40_ENABLED
if (teams_swapped) {
__kmp_restore_swapped_teams(th, team, task_state);
}
-#endif
KA_TRACE(
10,
("__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
@@ -3569,11 +3594,9 @@ kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
KMP_COUNT_BLOCK(REDUCE_wait);
int retval = 0;
PACKED_REDUCTION_METHOD_T packed_reduction_method;
-#if OMP_40_ENABLED
kmp_info_t *th;
kmp_team_t *team;
int teams_swapped = 0, task_state;
-#endif
KA_TRACE(10, ("__kmpc_reduce() enter: called T#%d\n", global_tid));
@@ -3586,6 +3609,8 @@ kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
if (!TCR_4(__kmp_init_parallel))
__kmp_parallel_initialize();
+ __kmp_resume_if_soft_paused();
+
// check correctness of reduce block nesting
#if KMP_USE_DYNAMIC_LOCK
if (__kmp_env_consistency_check)
@@ -3595,10 +3620,8 @@ kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
__kmp_push_sync(global_tid, ct_reduce, loc, NULL);
#endif
-#if OMP_40_ENABLED
th = __kmp_thread_from_gtid(global_tid);
teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
-#endif // OMP_40_ENABLED
packed_reduction_method = __kmp_determine_reduction_method(
loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
@@ -3661,16 +3684,13 @@ kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars,
// should never reach this block
KMP_ASSERT(0); // "unexpected method"
}
-#if OMP_40_ENABLED
if (teams_swapped) {
__kmp_restore_swapped_teams(th, team, task_state);
}
-#endif
KA_TRACE(10,
("__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
global_tid, packed_reduction_method, retval));
-
return retval;
}
@@ -3688,18 +3708,14 @@ void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
kmp_critical_name *lck) {
PACKED_REDUCTION_METHOD_T packed_reduction_method;
-#if OMP_40_ENABLED
kmp_info_t *th;
kmp_team_t *team;
int teams_swapped = 0, task_state;
-#endif
KA_TRACE(10, ("__kmpc_end_reduce() enter: called T#%d\n", global_tid));
-#if OMP_40_ENABLED
th = __kmp_thread_from_gtid(global_tid);
teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
-#endif // OMP_40_ENABLED
packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
@@ -3707,7 +3723,6 @@ void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
// tool (it's a terminating barrier on constructs if NOWAIT not specified)
if (packed_reduction_method == critical_reduce_block) {
-
__kmp_end_critical_section_reduce_block(loc, global_tid, lck);
// TODO: implicit barrier: should be exposed
@@ -3788,11 +3803,9 @@ void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
// should never reach this block
KMP_ASSERT(0); // "unexpected method"
}
-#if OMP_40_ENABLED
if (teams_swapped) {
__kmp_restore_swapped_teams(th, team, task_state);
}
-#endif
if (__kmp_env_consistency_check)
__kmp_pop_sync(global_tid, ct_reduce, loc);
@@ -3838,7 +3851,6 @@ kmp_uint64 __kmpc_get_parent_taskid() {
} // __kmpc_get_parent_taskid
-#if OMP_45_ENABLED
/*!
@ingroup WORK_SHARING
@param loc source location information.
@@ -3932,8 +3944,8 @@ void __kmpc_doacross_init(ident_t *loc, int gtid, int num_dims,
// __kmp_dispatch_num_buffers)
if (idx != sh_buf->doacross_buf_idx) {
// Shared buffer is occupied, wait for it to be free
- __kmp_wait_yield_4((volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
- __kmp_eq_4, NULL);
+ __kmp_wait_4((volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
+ __kmp_eq_4, NULL);
}
#if KMP_32_BIT_ARCH
// Check if we are the first thread. After the CAS the first thread gets 0,
@@ -4150,15 +4162,26 @@ void __kmpc_doacross_fini(ident_t *loc, int gtid) {
pr_buf->th_doacross_info = NULL;
KA_TRACE(20, ("__kmpc_doacross_fini() exit: T#%d\n", gtid));
}
-#endif
-#if OMP_50_ENABLED
+/* omp_alloc/omp_free only defined for C/C++, not for Fortran */
+void *omp_alloc(size_t size, omp_allocator_handle_t allocator) {
+ return __kmpc_alloc(__kmp_entry_gtid(), size, allocator);
+}
+
+void omp_free(void *ptr, omp_allocator_handle_t allocator) {
+ __kmpc_free(__kmp_entry_gtid(), ptr, allocator);
+}
+
int __kmpc_get_target_offload(void) {
if (!__kmp_init_serial) {
__kmp_serial_initialize();
}
return __kmp_target_offload;
}
-#endif // OMP_50_ENABLED
-// end of file //
+int __kmpc_pause_resource(kmp_pause_status_t level) {
+ if (!__kmp_init_serial) {
+ return 1; // Can't pause if runtime is not initialized
+ }
+ return __kmp_pause_resource(level);
+}