17 #include "kmp_error.h" 21 #include "kmp_stats.h" 24 #include "ompt-specific.h" 27 #define MAX_MESSAGE 512 43 if ((env = getenv(
"KMP_INITIAL_THREAD_BIND")) != NULL &&
44 __kmp_str_match_true(env)) {
45 __kmp_middle_initialize();
46 KC_TRACE(10, (
"__kmpc_begin: middle initialization called\n"));
47 }
else if (__kmp_ignore_mppbeg() == FALSE) {
49 __kmp_internal_begin();
50 KC_TRACE(10, (
"__kmpc_begin: called\n"));
68 if (__kmp_ignore_mppend() == FALSE) {
69 KC_TRACE(10, (
"__kmpc_end: called\n"));
70 KA_TRACE(30, (
"__kmpc_end\n"));
72 __kmp_internal_end_thread(-1);
74 #if KMP_OS_WINDOWS && OMPT_SUPPORT 79 if (ompt_enabled.enabled)
80 __kmp_internal_end_library(__kmp_gtid_get_specific());
103 kmp_int32 gtid = __kmp_entry_gtid();
105 KC_TRACE(10, (
"__kmpc_global_thread_num: T#%d\n", gtid));
126 (
"__kmpc_global_num_threads: num_threads = %d\n", __kmp_all_nth));
128 return TCR_4(__kmp_all_nth);
138 KC_TRACE(10, (
"__kmpc_bound_thread_num: called\n"));
139 return __kmp_tid_from_gtid(__kmp_entry_gtid());
148 KC_TRACE(10, (
"__kmpc_bound_num_threads: called\n"));
150 return __kmp_entry_thread()->th.th_team->t.t_nproc;
170 if (__kmp_par_range == 0) {
177 semi2 = strchr(semi2,
';');
181 semi2 = strchr(semi2 + 1,
';');
185 if (__kmp_par_range_filename[0]) {
186 const char *name = semi2 - 1;
187 while ((name > loc->
psource) && (*name !=
'/') && (*name !=
';')) {
190 if ((*name ==
'/') || (*name ==
';')) {
193 if (strncmp(__kmp_par_range_filename, name, semi2 - name)) {
194 return __kmp_par_range < 0;
197 semi3 = strchr(semi2 + 1,
';');
198 if (__kmp_par_range_routine[0]) {
199 if ((semi3 != NULL) && (semi3 > semi2) &&
200 (strncmp(__kmp_par_range_routine, semi2 + 1, semi3 - semi2 - 1))) {
201 return __kmp_par_range < 0;
204 if (KMP_SSCANF(semi3 + 1,
"%d", &line_no) == 1) {
205 if ((line_no >= __kmp_par_range_lb) && (line_no <= __kmp_par_range_ub)) {
206 return __kmp_par_range > 0;
208 return __kmp_par_range < 0;
222 return __kmp_entry_thread()->th.th_root->r.r_active;
235 kmp_int32 num_threads) {
236 KA_TRACE(20, (
"__kmpc_push_num_threads: enter T#%d num_threads=%d\n",
237 global_tid, num_threads));
239 __kmp_push_num_threads(loc, global_tid, num_threads);
242 void __kmpc_pop_num_threads(
ident_t *loc, kmp_int32 global_tid) {
243 KA_TRACE(20, (
"__kmpc_pop_num_threads: enter\n"));
250 void __kmpc_push_proc_bind(
ident_t *loc, kmp_int32 global_tid,
251 kmp_int32 proc_bind) {
252 KA_TRACE(20, (
"__kmpc_push_proc_bind: enter T#%d proc_bind=%d\n", global_tid,
255 __kmp_push_proc_bind(loc, global_tid, (kmp_proc_bind_t)proc_bind);
271 int gtid = __kmp_entry_gtid();
273 #if (KMP_STATS_ENABLED) 277 if (previous_state == stats_state_e::SERIAL_REGION) {
278 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_parallel_overhead);
280 KMP_PUSH_PARTITIONED_TIMER(OMP_parallel_overhead);
293 va_start(ap, microtask);
296 omp_frame_t *ompt_frame;
297 if (ompt_enabled.enabled) {
298 kmp_info_t *master_th = __kmp_threads[gtid];
299 kmp_team_t *parent_team = master_th->th.th_team;
300 ompt_lw_taskteam_t *lwt = parent_team->t.ompt_serialized_team_info;
302 ompt_frame = &(lwt->ompt_task_info.frame);
304 int tid = __kmp_tid_from_gtid(gtid);
306 parent_team->t.t_implicit_task_taskdata[tid].ompt_task_info.frame);
308 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
309 OMPT_STORE_RETURN_ADDRESS(gtid);
313 #if INCLUDE_SSC_MARKS 316 __kmp_fork_call(loc, gtid, fork_context_intel, argc,
317 VOLATILE_CAST(microtask_t) microtask,
318 VOLATILE_CAST(launch_t) __kmp_invoke_task_func,
320 #
if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
326 #if INCLUDE_SSC_MARKS 329 __kmp_join_call(loc, gtid
339 #if KMP_STATS_ENABLED 340 if (previous_state == stats_state_e::SERIAL_REGION) {
341 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
343 KMP_POP_PARTITIONED_TIMER();
345 #endif // KMP_STATS_ENABLED 361 kmp_int32 num_teams, kmp_int32 num_threads) {
363 (
"__kmpc_push_num_teams: enter T#%d num_teams=%d num_threads=%d\n",
364 global_tid, num_teams, num_threads));
366 __kmp_push_num_teams(loc, global_tid, num_teams, num_threads);
381 int gtid = __kmp_entry_gtid();
382 kmp_info_t *this_thr = __kmp_threads[gtid];
384 va_start(ap, microtask);
389 this_thr->th.th_teams_microtask = microtask;
390 this_thr->th.th_teams_level =
391 this_thr->th.th_team->t.t_level;
394 kmp_team_t *parent_team = this_thr->th.th_team;
395 int tid = __kmp_tid_from_gtid(gtid);
396 if (ompt_enabled.enabled) {
397 parent_team->t.t_implicit_task_taskdata[tid]
398 .ompt_task_info.frame.enter_frame = OMPT_GET_FRAME_ADDRESS(1);
400 OMPT_STORE_RETURN_ADDRESS(gtid);
405 if (this_thr->th.th_teams_size.nteams == 0) {
406 __kmp_push_num_teams(loc, gtid, 0, 0);
408 KMP_DEBUG_ASSERT(this_thr->th.th_set_nproc >= 1);
409 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nteams >= 1);
410 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nth >= 1);
412 __kmp_fork_call(loc, gtid, fork_context_intel, argc,
413 VOLATILE_CAST(microtask_t)
415 VOLATILE_CAST(launch_t) __kmp_invoke_teams_master,
416 #
if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
422 __kmp_join_call(loc, gtid
429 this_thr->th.th_teams_microtask = NULL;
430 this_thr->th.th_teams_level = 0;
431 *(kmp_int64 *)(&this_thr->th.th_teams_size) = 0L;
440 int __kmpc_invoke_task_func(
int gtid) {
return __kmp_invoke_task_func(gtid); }
459 OMPT_STORE_RETURN_ADDRESS(global_tid);
461 __kmp_serialized_parallel(loc, global_tid);
472 kmp_internal_control_t *top;
473 kmp_info_t *this_thr;
474 kmp_team_t *serial_team;
477 (
"__kmpc_end_serialized_parallel: called by T#%d\n", global_tid));
485 if (!TCR_4(__kmp_init_parallel))
486 __kmp_parallel_initialize();
488 this_thr = __kmp_threads[global_tid];
489 serial_team = this_thr->th.th_serial_team;
492 kmp_task_team_t *task_team = this_thr->th.th_task_team;
495 if (task_team != NULL && task_team->tt.tt_found_proxy_tasks)
496 __kmp_task_team_wait(this_thr, serial_team USE_ITT_BUILD_ARG(NULL));
500 KMP_DEBUG_ASSERT(serial_team);
501 KMP_ASSERT(serial_team->t.t_serialized);
502 KMP_DEBUG_ASSERT(this_thr->th.th_team == serial_team);
503 KMP_DEBUG_ASSERT(serial_team != this_thr->th.th_root->r.r_root_team);
504 KMP_DEBUG_ASSERT(serial_team->t.t_threads);
505 KMP_DEBUG_ASSERT(serial_team->t.t_threads[0] == this_thr);
508 if (ompt_enabled.enabled &&
509 this_thr->th.ompt_thread_info.state != omp_state_overhead) {
510 OMPT_CUR_TASK_INFO(this_thr)->frame.exit_frame = NULL;
511 if (ompt_enabled.ompt_callback_implicit_task) {
512 ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
513 ompt_scope_end, NULL, OMPT_CUR_TASK_DATA(this_thr), 1,
514 OMPT_CUR_TASK_INFO(this_thr)->thread_num);
518 ompt_data_t *parent_task_data;
519 __ompt_get_task_info_internal(1, NULL, &parent_task_data, NULL, NULL, NULL);
521 if (ompt_enabled.ompt_callback_parallel_end) {
522 ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
523 &(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
524 ompt_parallel_invoker_program, OMPT_LOAD_RETURN_ADDRESS(global_tid));
526 __ompt_lw_taskteam_unlink(this_thr);
527 this_thr->th.ompt_thread_info.state = omp_state_overhead;
533 top = serial_team->t.t_control_stack_top;
534 if (top && top->serial_nesting_level == serial_team->t.t_serialized) {
535 copy_icvs(&serial_team->t.t_threads[0]->th.th_current_task->td_icvs, top);
536 serial_team->t.t_control_stack_top = top->next;
541 serial_team->t.t_level--;
544 KMP_DEBUG_ASSERT(serial_team->t.t_dispatch->th_disp_buffer);
546 dispatch_private_info_t *disp_buffer =
547 serial_team->t.t_dispatch->th_disp_buffer;
548 serial_team->t.t_dispatch->th_disp_buffer =
549 serial_team->t.t_dispatch->th_disp_buffer->next;
550 __kmp_free(disp_buffer);
553 this_thr->th.th_def_allocator = serial_team->t.t_def_allocator;
556 --serial_team->t.t_serialized;
557 if (serial_team->t.t_serialized == 0) {
561 #if KMP_ARCH_X86 || KMP_ARCH_X86_64 562 if (__kmp_inherit_fp_control && serial_team->t.t_fp_control_saved) {
563 __kmp_clear_x87_fpu_status_word();
564 __kmp_load_x87_fpu_control_word(&serial_team->t.t_x87_fpu_control_word);
565 __kmp_load_mxcsr(&serial_team->t.t_mxcsr);
569 this_thr->th.th_team = serial_team->t.t_parent;
570 this_thr->th.th_info.ds.ds_tid = serial_team->t.t_master_tid;
573 this_thr->th.th_team_nproc = serial_team->t.t_parent->t.t_nproc;
574 this_thr->th.th_team_master =
575 serial_team->t.t_parent->t.t_threads[0];
576 this_thr->th.th_team_serialized = this_thr->th.th_team->t.t_serialized;
579 this_thr->th.th_dispatch =
580 &this_thr->th.th_team->t.t_dispatch[serial_team->t.t_master_tid];
582 __kmp_pop_current_task_from_thread(this_thr);
584 KMP_ASSERT(this_thr->th.th_current_task->td_flags.executing == 0);
585 this_thr->th.th_current_task->td_flags.executing = 1;
587 if (__kmp_tasking_mode != tskm_immediate_exec) {
589 this_thr->th.th_task_team =
590 this_thr->th.th_team->t.t_task_team[this_thr->th.th_task_state];
592 (
"__kmpc_end_serialized_parallel: T#%d restoring task_team %p / " 594 global_tid, this_thr->th.th_task_team, this_thr->th.th_team));
597 if (__kmp_tasking_mode != tskm_immediate_exec) {
598 KA_TRACE(20, (
"__kmpc_end_serialized_parallel: T#%d decreasing nesting " 599 "depth of serial team %p to %d\n",
600 global_tid, serial_team, serial_team->t.t_serialized));
604 if (__kmp_env_consistency_check)
605 __kmp_pop_parallel(global_tid, NULL);
607 if (ompt_enabled.enabled)
608 this_thr->th.ompt_thread_info.state =
609 ((this_thr->th.th_team_serialized) ? omp_state_work_serial
610 : omp_state_work_parallel);
623 KC_TRACE(10, (
"__kmpc_flush: called\n"));
628 #if (KMP_ARCH_X86 || KMP_ARCH_X86_64) 642 if (!__kmp_cpuinfo.initialized) {
643 __kmp_query_cpuid(&__kmp_cpuinfo);
645 if (!__kmp_cpuinfo.sse2) {
650 #elif KMP_COMPILER_MSVC 653 __sync_synchronize();
654 #endif // KMP_COMPILER_ICC 657 #elif (KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || KMP_ARCH_MIPS64) 673 #error Unknown or unsupported architecture 676 #if OMPT_SUPPORT && OMPT_OPTIONAL 677 if (ompt_enabled.ompt_callback_flush) {
678 ompt_callbacks.ompt_callback(ompt_callback_flush)(
679 __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
694 KC_TRACE(10, (
"__kmpc_barrier: called T#%d\n", global_tid));
696 if (!TCR_4(__kmp_init_parallel))
697 __kmp_parallel_initialize();
699 if (__kmp_env_consistency_check) {
701 KMP_WARNING(ConstructIdentInvalid);
704 __kmp_check_barrier(global_tid, ct_barrier, loc);
708 omp_frame_t *ompt_frame;
709 if (ompt_enabled.enabled) {
710 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
711 if (ompt_frame->enter_frame == NULL)
712 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
713 OMPT_STORE_RETURN_ADDRESS(global_tid);
716 __kmp_threads[global_tid]->th.th_ident = loc;
724 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
725 #if OMPT_SUPPORT && OMPT_OPTIONAL 726 if (ompt_enabled.enabled) {
727 ompt_frame->enter_frame = NULL;
742 KC_TRACE(10, (
"__kmpc_master: called T#%d\n", global_tid));
744 if (!TCR_4(__kmp_init_parallel))
745 __kmp_parallel_initialize();
747 if (KMP_MASTER_GTID(global_tid)) {
749 KMP_PUSH_PARTITIONED_TIMER(OMP_master);
753 #if OMPT_SUPPORT && OMPT_OPTIONAL 755 if (ompt_enabled.ompt_callback_master) {
756 kmp_info_t *this_thr = __kmp_threads[global_tid];
757 kmp_team_t *team = this_thr->th.th_team;
759 int tid = __kmp_tid_from_gtid(global_tid);
760 ompt_callbacks.ompt_callback(ompt_callback_master)(
761 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
762 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
763 OMPT_GET_RETURN_ADDRESS(0));
768 if (__kmp_env_consistency_check) {
769 #if KMP_USE_DYNAMIC_LOCK 771 __kmp_push_sync(global_tid, ct_master, loc, NULL, 0);
773 __kmp_check_sync(global_tid, ct_master, loc, NULL, 0);
776 __kmp_push_sync(global_tid, ct_master, loc, NULL);
778 __kmp_check_sync(global_tid, ct_master, loc, NULL);
794 KC_TRACE(10, (
"__kmpc_end_master: called T#%d\n", global_tid));
796 KMP_DEBUG_ASSERT(KMP_MASTER_GTID(global_tid));
797 KMP_POP_PARTITIONED_TIMER();
799 #if OMPT_SUPPORT && OMPT_OPTIONAL 800 kmp_info_t *this_thr = __kmp_threads[global_tid];
801 kmp_team_t *team = this_thr->th.th_team;
802 if (ompt_enabled.ompt_callback_master) {
803 int tid = __kmp_tid_from_gtid(global_tid);
804 ompt_callbacks.ompt_callback(ompt_callback_master)(
805 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
806 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
807 OMPT_GET_RETURN_ADDRESS(0));
811 if (__kmp_env_consistency_check) {
813 KMP_WARNING(ThreadIdentInvalid);
815 if (KMP_MASTER_GTID(global_tid))
816 __kmp_pop_sync(global_tid, ct_master, loc);
830 KMP_DEBUG_ASSERT(__kmp_init_serial);
832 KC_TRACE(10, (
"__kmpc_ordered: called T#%d\n", gtid));
834 if (!TCR_4(__kmp_init_parallel))
835 __kmp_parallel_initialize();
838 __kmp_itt_ordered_prep(gtid);
842 th = __kmp_threads[gtid];
844 #if OMPT_SUPPORT && OMPT_OPTIONAL 848 if (ompt_enabled.enabled) {
849 OMPT_STORE_RETURN_ADDRESS(gtid);
850 team = __kmp_team_from_gtid(gtid);
851 lck = (omp_wait_id_t)&team->t.t_ordered.dt.t_value;
853 th->th.ompt_thread_info.wait_id = lck;
854 th->th.ompt_thread_info.state = omp_state_wait_ordered;
857 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
858 if (ompt_enabled.ompt_callback_mutex_acquire) {
859 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
860 ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin,
861 (omp_wait_id_t)lck, codeptr_ra);
866 if (th->th.th_dispatch->th_deo_fcn != 0)
867 (*th->th.th_dispatch->th_deo_fcn)(>id, &cid, loc);
869 __kmp_parallel_deo(>id, &cid, loc);
871 #if OMPT_SUPPORT && OMPT_OPTIONAL 872 if (ompt_enabled.enabled) {
874 th->th.ompt_thread_info.state = omp_state_work_parallel;
875 th->th.ompt_thread_info.wait_id = 0;
878 if (ompt_enabled.ompt_callback_mutex_acquired) {
879 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
880 ompt_mutex_ordered, (omp_wait_id_t)lck, codeptr_ra);
886 __kmp_itt_ordered_start(gtid);
901 KC_TRACE(10, (
"__kmpc_end_ordered: called T#%d\n", gtid));
904 __kmp_itt_ordered_end(gtid);
908 th = __kmp_threads[gtid];
910 if (th->th.th_dispatch->th_dxo_fcn != 0)
911 (*th->th.th_dispatch->th_dxo_fcn)(>id, &cid, loc);
913 __kmp_parallel_dxo(>id, &cid, loc);
915 #if OMPT_SUPPORT && OMPT_OPTIONAL 916 OMPT_STORE_RETURN_ADDRESS(gtid);
917 if (ompt_enabled.ompt_callback_mutex_released) {
918 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
920 (omp_wait_id_t)&__kmp_team_from_gtid(gtid)->t.t_ordered.dt.t_value,
921 OMPT_LOAD_RETURN_ADDRESS(gtid));
926 #if KMP_USE_DYNAMIC_LOCK 928 static __forceinline
void 929 __kmp_init_indirect_csptr(kmp_critical_name *crit,
ident_t const *loc,
930 kmp_int32 gtid, kmp_indirect_locktag_t tag) {
934 kmp_indirect_lock_t **lck;
935 lck = (kmp_indirect_lock_t **)crit;
936 kmp_indirect_lock_t *ilk = __kmp_allocate_indirect_lock(&idx, gtid, tag);
937 KMP_I_LOCK_FUNC(ilk, init)(ilk->lock);
938 KMP_SET_I_LOCK_LOCATION(ilk, loc);
939 KMP_SET_I_LOCK_FLAGS(ilk, kmp_lf_critical_section);
941 (
"__kmp_init_indirect_csptr: initialized indirect lock #%d\n", tag));
943 __kmp_itt_critical_creating(ilk->lock, loc);
945 int status = KMP_COMPARE_AND_STORE_PTR(lck,
nullptr, ilk);
948 __kmp_itt_critical_destroyed(ilk->lock);
954 KMP_DEBUG_ASSERT(*lck != NULL);
958 #define KMP_ACQUIRE_TAS_LOCK(lock, gtid) \ 960 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \ 961 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \ 962 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \ 963 if (KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \ 964 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \ 966 KMP_FSYNC_PREPARE(l); \ 967 KMP_INIT_YIELD(spins); \ 968 if (TCR_4(__kmp_nth) > \ 969 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \ 972 KMP_YIELD_SPIN(spins); \ 974 kmp_backoff_t backoff = __kmp_spin_backoff_params; \ 976 KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \ 977 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \ 978 __kmp_spin_backoff(&backoff); \ 979 if (TCR_4(__kmp_nth) > \ 980 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \ 983 KMP_YIELD_SPIN(spins); \ 987 KMP_FSYNC_ACQUIRED(l); \ 991 #define KMP_TEST_TAS_LOCK(lock, gtid, rc) \ 993 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \ 994 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \ 995 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \ 996 rc = KMP_ATOMIC_LD_RLX(&l->lk.poll) == tas_free && \ 997 __kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy); \ 1001 #define KMP_RELEASE_TAS_LOCK(lock, gtid) \ 1002 { KMP_ATOMIC_ST_REL(&((kmp_tas_lock_t *)lock)->lk.poll, KMP_LOCK_FREE(tas)); } 1006 #include <sys/syscall.h> 1009 #define FUTEX_WAIT 0 1012 #define FUTEX_WAKE 1 1016 #define KMP_ACQUIRE_FUTEX_LOCK(lock, gtid) \ 1018 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \ 1019 kmp_int32 gtid_code = (gtid + 1) << 1; \ 1021 KMP_FSYNC_PREPARE(ftx); \ 1022 kmp_int32 poll_val; \ 1023 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( \ 1024 &(ftx->lk.poll), KMP_LOCK_FREE(futex), \ 1025 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { \ 1026 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; \ 1028 if (!KMP_COMPARE_AND_STORE_RET32(&(ftx->lk.poll), poll_val, \ 1030 KMP_LOCK_BUSY(1, futex))) { \ 1033 poll_val |= KMP_LOCK_BUSY(1, futex); \ 1036 if ((rc = syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAIT, poll_val, \ 1037 NULL, NULL, 0)) != 0) { \ 1042 KMP_FSYNC_ACQUIRED(ftx); \ 1046 #define KMP_TEST_FUTEX_LOCK(lock, gtid, rc) \ 1048 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \ 1049 if (KMP_COMPARE_AND_STORE_ACQ32(&(ftx->lk.poll), KMP_LOCK_FREE(futex), \ 1050 KMP_LOCK_BUSY(gtid + 1 << 1, futex))) { \ 1051 KMP_FSYNC_ACQUIRED(ftx); \ 1059 #define KMP_RELEASE_FUTEX_LOCK(lock, gtid) \ 1061 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \ 1063 KMP_FSYNC_RELEASING(ftx); \ 1064 kmp_int32 poll_val = \ 1065 KMP_XCHG_FIXED32(&(ftx->lk.poll), KMP_LOCK_FREE(futex)); \ 1066 if (KMP_LOCK_STRIP(poll_val) & 1) { \ 1067 syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAKE, \ 1068 KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \ 1071 KMP_YIELD(TCR_4(__kmp_nth) > \ 1072 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)); \ 1075 #endif // KMP_USE_FUTEX 1077 #else // KMP_USE_DYNAMIC_LOCK 1079 static kmp_user_lock_p __kmp_get_critical_section_ptr(kmp_critical_name *crit,
1082 kmp_user_lock_p *lck_pp = (kmp_user_lock_p *)crit;
1085 kmp_user_lock_p lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1092 lck = __kmp_user_lock_allocate(&idx, gtid, kmp_lf_critical_section);
1093 __kmp_init_user_lock_with_checks(lck);
1094 __kmp_set_user_lock_location(lck, loc);
1096 __kmp_itt_critical_creating(lck);
1107 int status = KMP_COMPARE_AND_STORE_PTR(lck_pp, 0, lck);
1112 __kmp_itt_critical_destroyed(lck);
1116 __kmp_destroy_user_lock_with_checks(lck);
1117 __kmp_user_lock_free(&idx, gtid, lck);
1118 lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1119 KMP_DEBUG_ASSERT(lck != NULL);
1125 #endif // KMP_USE_DYNAMIC_LOCK 1138 kmp_critical_name *crit) {
1139 #if KMP_USE_DYNAMIC_LOCK 1140 #if OMPT_SUPPORT && OMPT_OPTIONAL 1141 OMPT_STORE_RETURN_ADDRESS(global_tid);
1142 #endif // OMPT_SUPPORT 1143 __kmpc_critical_with_hint(loc, global_tid, crit, omp_lock_hint_none);
1146 #if OMPT_SUPPORT && OMPT_OPTIONAL 1147 omp_state_t prev_state = omp_state_undefined;
1148 ompt_thread_info_t ti;
1150 kmp_user_lock_p lck;
1152 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1156 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1157 KMP_CHECK_USER_LOCK_INIT();
1159 if ((__kmp_user_lock_kind == lk_tas) &&
1160 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1161 lck = (kmp_user_lock_p)crit;
1164 else if ((__kmp_user_lock_kind == lk_futex) &&
1165 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1166 lck = (kmp_user_lock_p)crit;
1170 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
1173 if (__kmp_env_consistency_check)
1174 __kmp_push_sync(global_tid, ct_critical, loc, lck);
1182 __kmp_itt_critical_acquiring(lck);
1184 #if OMPT_SUPPORT && OMPT_OPTIONAL 1185 OMPT_STORE_RETURN_ADDRESS(gtid);
1186 void *codeptr_ra = NULL;
1187 if (ompt_enabled.enabled) {
1188 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1190 prev_state = ti.state;
1191 ti.wait_id = (omp_wait_id_t)lck;
1192 ti.state = omp_state_wait_critical;
1195 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1196 if (ompt_enabled.ompt_callback_mutex_acquire) {
1197 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1198 ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
1199 (omp_wait_id_t)crit, codeptr_ra);
1205 __kmp_acquire_user_lock_with_checks(lck, global_tid);
1208 __kmp_itt_critical_acquired(lck);
1210 #if OMPT_SUPPORT && OMPT_OPTIONAL 1211 if (ompt_enabled.enabled) {
1213 ti.state = prev_state;
1217 if (ompt_enabled.ompt_callback_mutex_acquired) {
1218 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1219 ompt_mutex_critical, (omp_wait_id_t)crit, codeptr_ra);
1223 KMP_POP_PARTITIONED_TIMER();
1225 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1226 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1227 #endif // KMP_USE_DYNAMIC_LOCK 1230 #if KMP_USE_DYNAMIC_LOCK 1233 static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
1235 #define KMP_TSX_LOCK(seq) lockseq_##seq 1237 #define KMP_TSX_LOCK(seq) __kmp_user_lock_seq 1240 #if KMP_ARCH_X86 || KMP_ARCH_X86_64 1241 #define KMP_CPUINFO_RTM (__kmp_cpuinfo.rtm) 1243 #define KMP_CPUINFO_RTM 0 1247 if (hint & kmp_lock_hint_hle)
1248 return KMP_TSX_LOCK(hle);
1249 if (hint & kmp_lock_hint_rtm)
1250 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm) : __kmp_user_lock_seq;
1251 if (hint & kmp_lock_hint_adaptive)
1252 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
1255 if ((hint & omp_lock_hint_contended) && (hint & omp_lock_hint_uncontended))
1256 return __kmp_user_lock_seq;
1257 if ((hint & omp_lock_hint_speculative) &&
1258 (hint & omp_lock_hint_nonspeculative))
1259 return __kmp_user_lock_seq;
1262 if (hint & omp_lock_hint_contended)
1263 return lockseq_queuing;
1266 if ((hint & omp_lock_hint_uncontended) && !(hint & omp_lock_hint_speculative))
1270 if (hint & omp_lock_hint_speculative)
1271 return KMP_TSX_LOCK(hle);
1273 return __kmp_user_lock_seq;
1276 #if OMPT_SUPPORT && OMPT_OPTIONAL 1277 #if KMP_USE_DYNAMIC_LOCK 1278 static kmp_mutex_impl_t
1279 __ompt_get_mutex_impl_type(
void *user_lock, kmp_indirect_lock_t *ilock = 0) {
1281 switch (KMP_EXTRACT_D_TAG(user_lock)) {
1286 return kmp_mutex_impl_queuing;
1289 return kmp_mutex_impl_spin;
1292 return kmp_mutex_impl_speculative;
1295 return ompt_mutex_impl_unknown;
1297 ilock = KMP_LOOKUP_I_LOCK(user_lock);
1300 switch (ilock->type) {
1302 case locktag_adaptive:
1304 return kmp_mutex_impl_speculative;
1306 case locktag_nested_tas:
1307 return kmp_mutex_impl_spin;
1309 case locktag_nested_futex:
1311 case locktag_ticket:
1312 case locktag_queuing:
1314 case locktag_nested_ticket:
1315 case locktag_nested_queuing:
1316 case locktag_nested_drdpa:
1317 return kmp_mutex_impl_queuing;
1319 return ompt_mutex_impl_unknown;
1324 static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
1325 switch (__kmp_user_lock_kind) {
1327 return kmp_mutex_impl_spin;
1334 return kmp_mutex_impl_queuing;
1339 return kmp_mutex_impl_speculative;
1342 return ompt_mutex_impl_unknown;
1345 #endif // KMP_USE_DYNAMIC_LOCK 1346 #endif // OMPT_SUPPORT && OMPT_OPTIONAL 1361 void __kmpc_critical_with_hint(
ident_t *loc, kmp_int32 global_tid,
1362 kmp_critical_name *crit, uint32_t hint) {
1364 kmp_user_lock_p lck;
1365 #if OMPT_SUPPORT && OMPT_OPTIONAL 1366 omp_state_t prev_state = omp_state_undefined;
1367 ompt_thread_info_t ti;
1369 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(global_tid);
1371 codeptr = OMPT_GET_RETURN_ADDRESS(0);
1374 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1376 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
1378 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1380 kmp_dyna_lockseq_t lckseq = __kmp_map_hint_to_lock(hint);
1381 if (KMP_IS_D_LOCK(lckseq)) {
1382 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
1383 KMP_GET_D_TAG(lckseq));
1385 __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lckseq));
1391 if (KMP_EXTRACT_D_TAG(lk) != 0) {
1392 lck = (kmp_user_lock_p)lk;
1393 if (__kmp_env_consistency_check) {
1394 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1395 __kmp_map_hint_to_lock(hint));
1398 __kmp_itt_critical_acquiring(lck);
1400 #if OMPT_SUPPORT && OMPT_OPTIONAL 1401 if (ompt_enabled.enabled) {
1402 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1404 prev_state = ti.state;
1405 ti.wait_id = (omp_wait_id_t)lck;
1406 ti.state = omp_state_wait_critical;
1409 if (ompt_enabled.ompt_callback_mutex_acquire) {
1410 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1411 ompt_mutex_critical, (
unsigned int)hint,
1412 __ompt_get_mutex_impl_type(crit), (omp_wait_id_t)crit, codeptr);
1416 #if KMP_USE_INLINED_TAS 1417 if (__kmp_user_lock_seq == lockseq_tas && !__kmp_env_consistency_check) {
1418 KMP_ACQUIRE_TAS_LOCK(lck, global_tid);
1420 #elif KMP_USE_INLINED_FUTEX 1421 if (__kmp_user_lock_seq == lockseq_futex && !__kmp_env_consistency_check) {
1422 KMP_ACQUIRE_FUTEX_LOCK(lck, global_tid);
1426 KMP_D_LOCK_FUNC(lk,
set)(lk, global_tid);
1429 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
1431 if (__kmp_env_consistency_check) {
1432 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1433 __kmp_map_hint_to_lock(hint));
1436 __kmp_itt_critical_acquiring(lck);
1438 #if OMPT_SUPPORT && OMPT_OPTIONAL 1439 if (ompt_enabled.enabled) {
1440 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1442 prev_state = ti.state;
1443 ti.wait_id = (omp_wait_id_t)lck;
1444 ti.state = omp_state_wait_critical;
1447 if (ompt_enabled.ompt_callback_mutex_acquire) {
1448 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1449 ompt_mutex_critical, (
unsigned int)hint,
1450 __ompt_get_mutex_impl_type(0, ilk), (omp_wait_id_t)crit, codeptr);
1454 KMP_I_LOCK_FUNC(ilk,
set)(lck, global_tid);
1456 KMP_POP_PARTITIONED_TIMER();
1459 __kmp_itt_critical_acquired(lck);
1461 #if OMPT_SUPPORT && OMPT_OPTIONAL 1462 if (ompt_enabled.enabled) {
1464 ti.state = prev_state;
1468 if (ompt_enabled.ompt_callback_mutex_acquired) {
1469 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1470 ompt_mutex_critical, (omp_wait_id_t)crit, codeptr);
1475 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1476 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1479 #endif // KMP_USE_DYNAMIC_LOCK 1491 kmp_critical_name *crit) {
1492 kmp_user_lock_p lck;
1494 KC_TRACE(10, (
"__kmpc_end_critical: called T#%d\n", global_tid));
1496 #if KMP_USE_DYNAMIC_LOCK 1497 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
1498 lck = (kmp_user_lock_p)crit;
1499 KMP_ASSERT(lck != NULL);
1500 if (__kmp_env_consistency_check) {
1501 __kmp_pop_sync(global_tid, ct_critical, loc);
1504 __kmp_itt_critical_releasing(lck);
1506 #if KMP_USE_INLINED_TAS 1507 if (__kmp_user_lock_seq == lockseq_tas && !__kmp_env_consistency_check) {
1508 KMP_RELEASE_TAS_LOCK(lck, global_tid);
1510 #elif KMP_USE_INLINED_FUTEX 1511 if (__kmp_user_lock_seq == lockseq_futex && !__kmp_env_consistency_check) {
1512 KMP_RELEASE_FUTEX_LOCK(lck, global_tid);
1516 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
1519 kmp_indirect_lock_t *ilk =
1520 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
1521 KMP_ASSERT(ilk != NULL);
1523 if (__kmp_env_consistency_check) {
1524 __kmp_pop_sync(global_tid, ct_critical, loc);
1527 __kmp_itt_critical_releasing(lck);
1529 KMP_I_LOCK_FUNC(ilk, unset)(lck, global_tid);
1532 #else // KMP_USE_DYNAMIC_LOCK 1534 if ((__kmp_user_lock_kind == lk_tas) &&
1535 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1536 lck = (kmp_user_lock_p)crit;
1539 else if ((__kmp_user_lock_kind == lk_futex) &&
1540 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1541 lck = (kmp_user_lock_p)crit;
1545 lck = (kmp_user_lock_p)TCR_PTR(*((kmp_user_lock_p *)crit));
1548 KMP_ASSERT(lck != NULL);
1550 if (__kmp_env_consistency_check)
1551 __kmp_pop_sync(global_tid, ct_critical, loc);
1554 __kmp_itt_critical_releasing(lck);
1558 __kmp_release_user_lock_with_checks(lck, global_tid);
1560 #endif // KMP_USE_DYNAMIC_LOCK 1562 #if OMPT_SUPPORT && OMPT_OPTIONAL 1565 OMPT_STORE_RETURN_ADDRESS(global_tid);
1566 if (ompt_enabled.ompt_callback_mutex_released) {
1567 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1568 ompt_mutex_critical, (omp_wait_id_t)crit, OMPT_LOAD_RETURN_ADDRESS(0));
1572 KMP_POP_PARTITIONED_TIMER();
1573 KA_TRACE(15, (
"__kmpc_end_critical: done T#%d\n", global_tid));
1588 KC_TRACE(10, (
"__kmpc_barrier_master: called T#%d\n", global_tid));
1590 if (!TCR_4(__kmp_init_parallel))
1591 __kmp_parallel_initialize();
1593 if (__kmp_env_consistency_check)
1594 __kmp_check_barrier(global_tid, ct_barrier, loc);
1597 omp_frame_t *ompt_frame;
1598 if (ompt_enabled.enabled) {
1599 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1600 if (ompt_frame->enter_frame == NULL)
1601 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
1602 OMPT_STORE_RETURN_ADDRESS(global_tid);
1606 __kmp_threads[global_tid]->th.th_ident = loc;
1608 status = __kmp_barrier(bs_plain_barrier, global_tid, TRUE, 0, NULL, NULL);
1609 #if OMPT_SUPPORT && OMPT_OPTIONAL 1610 if (ompt_enabled.enabled) {
1611 ompt_frame->enter_frame = NULL;
1615 return (status != 0) ? 0 : 1;
1628 KC_TRACE(10, (
"__kmpc_end_barrier_master: called T#%d\n", global_tid));
1630 __kmp_end_split_barrier(bs_plain_barrier, global_tid);
1646 KC_TRACE(10, (
"__kmpc_barrier_master_nowait: called T#%d\n", global_tid));
1648 if (!TCR_4(__kmp_init_parallel))
1649 __kmp_parallel_initialize();
1651 if (__kmp_env_consistency_check) {
1653 KMP_WARNING(ConstructIdentInvalid);
1655 __kmp_check_barrier(global_tid, ct_barrier, loc);
1659 omp_frame_t *ompt_frame;
1660 if (ompt_enabled.enabled) {
1661 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1662 if (ompt_frame->enter_frame == NULL)
1663 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
1664 OMPT_STORE_RETURN_ADDRESS(global_tid);
1668 __kmp_threads[global_tid]->th.th_ident = loc;
1670 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
1671 #if OMPT_SUPPORT && OMPT_OPTIONAL 1672 if (ompt_enabled.enabled) {
1673 ompt_frame->enter_frame = NULL;
1679 if (__kmp_env_consistency_check) {
1683 if (global_tid < 0) {
1684 KMP_WARNING(ThreadIdentInvalid);
1690 __kmp_pop_sync(global_tid, ct_master, loc);
1710 kmp_int32 rc = __kmp_enter_single(global_tid, loc, TRUE);
1715 KMP_PUSH_PARTITIONED_TIMER(OMP_single);
1718 #if OMPT_SUPPORT && OMPT_OPTIONAL 1719 kmp_info_t *this_thr = __kmp_threads[global_tid];
1720 kmp_team_t *team = this_thr->th.th_team;
1721 int tid = __kmp_tid_from_gtid(global_tid);
1723 if (ompt_enabled.enabled) {
1725 if (ompt_enabled.ompt_callback_work) {
1726 ompt_callbacks.ompt_callback(ompt_callback_work)(
1727 ompt_work_single_executor, ompt_scope_begin,
1728 &(team->t.ompt_team_info.parallel_data),
1729 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1730 1, OMPT_GET_RETURN_ADDRESS(0));
1733 if (ompt_enabled.ompt_callback_work) {
1734 ompt_callbacks.ompt_callback(ompt_callback_work)(
1735 ompt_work_single_other, ompt_scope_begin,
1736 &(team->t.ompt_team_info.parallel_data),
1737 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1738 1, OMPT_GET_RETURN_ADDRESS(0));
1739 ompt_callbacks.ompt_callback(ompt_callback_work)(
1740 ompt_work_single_other, ompt_scope_end,
1741 &(team->t.ompt_team_info.parallel_data),
1742 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1743 1, OMPT_GET_RETURN_ADDRESS(0));
1762 __kmp_exit_single(global_tid);
1763 KMP_POP_PARTITIONED_TIMER();
1765 #if OMPT_SUPPORT && OMPT_OPTIONAL 1766 kmp_info_t *this_thr = __kmp_threads[global_tid];
1767 kmp_team_t *team = this_thr->th.th_team;
1768 int tid = __kmp_tid_from_gtid(global_tid);
1770 if (ompt_enabled.ompt_callback_work) {
1771 ompt_callbacks.ompt_callback(ompt_callback_work)(
1772 ompt_work_single_executor, ompt_scope_end,
1773 &(team->t.ompt_team_info.parallel_data),
1774 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
1775 OMPT_GET_RETURN_ADDRESS(0));
1788 KMP_POP_PARTITIONED_TIMER();
1789 KE_TRACE(10, (
"__kmpc_for_static_fini called T#%d\n", global_tid));
1791 #if OMPT_SUPPORT && OMPT_OPTIONAL 1792 if (ompt_enabled.ompt_callback_work) {
1793 ompt_work_t ompt_work_type = ompt_work_loop;
1794 ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
1795 ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
1799 ompt_work_type = ompt_work_loop;
1801 ompt_work_type = ompt_work_sections;
1803 ompt_work_type = ompt_work_distribute;
1808 KMP_DEBUG_ASSERT(ompt_work_type);
1810 ompt_callbacks.ompt_callback(ompt_callback_work)(
1811 ompt_work_type, ompt_scope_end, &(team_info->parallel_data),
1812 &(task_info->task_data), 0, OMPT_GET_RETURN_ADDRESS(0));
1815 if (__kmp_env_consistency_check)
1816 __kmp_pop_workshare(global_tid, ct_pdo, loc);
1822 void ompc_set_num_threads(
int arg) {
1824 __kmp_set_num_threads(arg, __kmp_entry_gtid());
1827 void ompc_set_dynamic(
int flag) {
1831 thread = __kmp_entry_thread();
1833 __kmp_save_internal_controls(thread);
1835 set__dynamic(thread, flag ? TRUE : FALSE);
1838 void ompc_set_nested(
int flag) {
1842 thread = __kmp_entry_thread();
1844 __kmp_save_internal_controls(thread);
1846 set__nested(thread, flag ? TRUE : FALSE);
1849 void ompc_set_max_active_levels(
int max_active_levels) {
1854 __kmp_set_max_active_levels(__kmp_entry_gtid(), max_active_levels);
1857 void ompc_set_schedule(omp_sched_t kind,
int modifier) {
1859 __kmp_set_schedule(__kmp_entry_gtid(), (kmp_sched_t)kind, modifier);
1862 int ompc_get_ancestor_thread_num(
int level) {
1863 return __kmp_get_ancestor_thread_num(__kmp_entry_gtid(), level);
1866 int ompc_get_team_size(
int level) {
1867 return __kmp_get_team_size(__kmp_entry_gtid(), level);
1870 void kmpc_set_stacksize(
int arg) {
1872 __kmp_aux_set_stacksize(arg);
1875 void kmpc_set_stacksize_s(
size_t arg) {
1877 __kmp_aux_set_stacksize(arg);
1880 void kmpc_set_blocktime(
int arg) {
1884 gtid = __kmp_entry_gtid();
1885 tid = __kmp_tid_from_gtid(gtid);
1886 thread = __kmp_thread_from_gtid(gtid);
1888 __kmp_aux_set_blocktime(arg, thread, tid);
1891 void kmpc_set_library(
int arg) {
1893 __kmp_user_set_library((
enum library_type)arg);
1896 void kmpc_set_defaults(
char const *str) {
1898 __kmp_aux_set_defaults(str, KMP_STRLEN(str));
1901 void kmpc_set_disp_num_buffers(
int arg) {
1904 if (__kmp_init_serial == 0 && arg > 0)
1905 __kmp_dispatch_num_buffers = arg;
1908 int kmpc_set_affinity_mask_proc(
int proc,
void **mask) {
1909 #if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED 1912 if (!TCR_4(__kmp_init_middle)) {
1913 __kmp_middle_initialize();
1915 return __kmp_aux_set_affinity_mask_proc(proc, mask);
1919 int kmpc_unset_affinity_mask_proc(
int proc,
void **mask) {
1920 #if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED 1923 if (!TCR_4(__kmp_init_middle)) {
1924 __kmp_middle_initialize();
1926 return __kmp_aux_unset_affinity_mask_proc(proc, mask);
1930 int kmpc_get_affinity_mask_proc(
int proc,
void **mask) {
1931 #if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED 1934 if (!TCR_4(__kmp_init_middle)) {
1935 __kmp_middle_initialize();
1937 return __kmp_aux_get_affinity_mask_proc(proc, mask);
1987 void *cpy_data,
void (*cpy_func)(
void *,
void *),
1991 KC_TRACE(10, (
"__kmpc_copyprivate: called T#%d\n", gtid));
1995 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
1997 if (__kmp_env_consistency_check) {
1999 KMP_WARNING(ConstructIdentInvalid);
2006 *data_ptr = cpy_data;
2009 omp_frame_t *ompt_frame;
2010 if (ompt_enabled.enabled) {
2011 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2012 if (ompt_frame->enter_frame == NULL)
2013 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
2014 OMPT_STORE_RETURN_ADDRESS(gtid);
2019 __kmp_threads[gtid]->th.th_ident = loc;
2021 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2024 (*cpy_func)(cpy_data, *data_ptr);
2030 if (ompt_enabled.enabled) {
2031 OMPT_STORE_RETURN_ADDRESS(gtid);
2035 __kmp_threads[gtid]->th.th_ident = loc;
2038 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2039 #if OMPT_SUPPORT && OMPT_OPTIONAL 2040 if (ompt_enabled.enabled) {
2041 ompt_frame->enter_frame = NULL;
2048 #define INIT_LOCK __kmp_init_user_lock_with_checks 2049 #define INIT_NESTED_LOCK __kmp_init_nested_user_lock_with_checks 2050 #define ACQUIRE_LOCK __kmp_acquire_user_lock_with_checks 2051 #define ACQUIRE_LOCK_TIMED __kmp_acquire_user_lock_with_checks_timed 2052 #define ACQUIRE_NESTED_LOCK __kmp_acquire_nested_user_lock_with_checks 2053 #define ACQUIRE_NESTED_LOCK_TIMED \ 2054 __kmp_acquire_nested_user_lock_with_checks_timed 2055 #define RELEASE_LOCK __kmp_release_user_lock_with_checks 2056 #define RELEASE_NESTED_LOCK __kmp_release_nested_user_lock_with_checks 2057 #define TEST_LOCK __kmp_test_user_lock_with_checks 2058 #define TEST_NESTED_LOCK __kmp_test_nested_user_lock_with_checks 2059 #define DESTROY_LOCK __kmp_destroy_user_lock_with_checks 2060 #define DESTROY_NESTED_LOCK __kmp_destroy_nested_user_lock_with_checks 2065 #if KMP_USE_DYNAMIC_LOCK 2068 static __forceinline
void __kmp_init_lock_with_hint(
ident_t *loc,
void **lock,
2069 kmp_dyna_lockseq_t seq) {
2070 if (KMP_IS_D_LOCK(seq)) {
2071 KMP_INIT_D_LOCK(lock, seq);
2073 __kmp_itt_lock_creating((kmp_user_lock_p)lock, NULL);
2076 KMP_INIT_I_LOCK(lock, seq);
2078 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2079 __kmp_itt_lock_creating(ilk->lock, loc);
2085 static __forceinline
void 2086 __kmp_init_nest_lock_with_hint(
ident_t *loc,
void **lock,
2087 kmp_dyna_lockseq_t seq) {
2090 if (seq == lockseq_hle || seq == lockseq_rtm || seq == lockseq_adaptive)
2091 seq = __kmp_user_lock_seq;
2095 seq = lockseq_nested_tas;
2099 seq = lockseq_nested_futex;
2102 case lockseq_ticket:
2103 seq = lockseq_nested_ticket;
2105 case lockseq_queuing:
2106 seq = lockseq_nested_queuing;
2109 seq = lockseq_nested_drdpa;
2112 seq = lockseq_nested_queuing;
2114 KMP_INIT_I_LOCK(lock, seq);
2116 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2117 __kmp_itt_lock_creating(ilk->lock, loc);
2122 void __kmpc_init_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
void **user_lock,
2124 KMP_DEBUG_ASSERT(__kmp_init_serial);
2125 if (__kmp_env_consistency_check && user_lock == NULL) {
2126 KMP_FATAL(LockIsUninitialized,
"omp_init_lock_with_hint");
2129 __kmp_init_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2131 #if OMPT_SUPPORT && OMPT_OPTIONAL 2133 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2135 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2136 if (ompt_enabled.ompt_callback_lock_init) {
2137 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2138 ompt_mutex_lock, (omp_lock_hint_t)hint,
2139 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2146 void __kmpc_init_nest_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
2147 void **user_lock, uintptr_t hint) {
2148 KMP_DEBUG_ASSERT(__kmp_init_serial);
2149 if (__kmp_env_consistency_check && user_lock == NULL) {
2150 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock_with_hint");
2153 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2155 #if OMPT_SUPPORT && OMPT_OPTIONAL 2157 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2159 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2160 if (ompt_enabled.ompt_callback_lock_init) {
2161 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2162 ompt_mutex_nest_lock, (omp_lock_hint_t)hint,
2163 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2169 #endif // KMP_USE_DYNAMIC_LOCK 2172 void __kmpc_init_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2173 #if KMP_USE_DYNAMIC_LOCK 2175 KMP_DEBUG_ASSERT(__kmp_init_serial);
2176 if (__kmp_env_consistency_check && user_lock == NULL) {
2177 KMP_FATAL(LockIsUninitialized,
"omp_init_lock");
2179 __kmp_init_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2181 #if OMPT_SUPPORT && OMPT_OPTIONAL 2183 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2185 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2186 if (ompt_enabled.ompt_callback_lock_init) {
2187 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2188 ompt_mutex_lock, omp_lock_hint_none,
2189 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2194 #else // KMP_USE_DYNAMIC_LOCK 2196 static char const *
const func =
"omp_init_lock";
2197 kmp_user_lock_p lck;
2198 KMP_DEBUG_ASSERT(__kmp_init_serial);
2200 if (__kmp_env_consistency_check) {
2201 if (user_lock == NULL) {
2202 KMP_FATAL(LockIsUninitialized, func);
2206 KMP_CHECK_USER_LOCK_INIT();
2208 if ((__kmp_user_lock_kind == lk_tas) &&
2209 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2210 lck = (kmp_user_lock_p)user_lock;
2213 else if ((__kmp_user_lock_kind == lk_futex) &&
2214 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2215 lck = (kmp_user_lock_p)user_lock;
2219 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2222 __kmp_set_user_lock_location(lck, loc);
2224 #if OMPT_SUPPORT && OMPT_OPTIONAL 2226 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2228 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2229 if (ompt_enabled.ompt_callback_lock_init) {
2230 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2231 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2232 (omp_wait_id_t)user_lock, codeptr);
2237 __kmp_itt_lock_creating(lck);
2240 #endif // KMP_USE_DYNAMIC_LOCK 2244 void __kmpc_init_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2245 #if KMP_USE_DYNAMIC_LOCK 2247 KMP_DEBUG_ASSERT(__kmp_init_serial);
2248 if (__kmp_env_consistency_check && user_lock == NULL) {
2249 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock");
2251 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2253 #if OMPT_SUPPORT && OMPT_OPTIONAL 2255 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2257 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2258 if (ompt_enabled.ompt_callback_lock_init) {
2259 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2260 ompt_mutex_nest_lock, omp_lock_hint_none,
2261 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2266 #else // KMP_USE_DYNAMIC_LOCK 2268 static char const *
const func =
"omp_init_nest_lock";
2269 kmp_user_lock_p lck;
2270 KMP_DEBUG_ASSERT(__kmp_init_serial);
2272 if (__kmp_env_consistency_check) {
2273 if (user_lock == NULL) {
2274 KMP_FATAL(LockIsUninitialized, func);
2278 KMP_CHECK_USER_LOCK_INIT();
2280 if ((__kmp_user_lock_kind == lk_tas) &&
2281 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2282 OMP_NEST_LOCK_T_SIZE)) {
2283 lck = (kmp_user_lock_p)user_lock;
2286 else if ((__kmp_user_lock_kind == lk_futex) &&
2287 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2288 OMP_NEST_LOCK_T_SIZE)) {
2289 lck = (kmp_user_lock_p)user_lock;
2293 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2296 INIT_NESTED_LOCK(lck);
2297 __kmp_set_user_lock_location(lck, loc);
2299 #if OMPT_SUPPORT && OMPT_OPTIONAL 2301 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2303 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2304 if (ompt_enabled.ompt_callback_lock_init) {
2305 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2306 ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2307 (omp_wait_id_t)user_lock, codeptr);
2312 __kmp_itt_lock_creating(lck);
2315 #endif // KMP_USE_DYNAMIC_LOCK 2318 void __kmpc_destroy_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2319 #if KMP_USE_DYNAMIC_LOCK 2322 kmp_user_lock_p lck;
2323 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2324 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2326 lck = (kmp_user_lock_p)user_lock;
2328 __kmp_itt_lock_destroyed(lck);
2330 #if OMPT_SUPPORT && OMPT_OPTIONAL 2332 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2334 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2335 if (ompt_enabled.ompt_callback_lock_destroy) {
2336 kmp_user_lock_p lck;
2337 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2338 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2340 lck = (kmp_user_lock_p)user_lock;
2342 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2343 ompt_mutex_lock, (omp_wait_id_t)user_lock, codeptr);
2346 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2348 kmp_user_lock_p lck;
2350 if ((__kmp_user_lock_kind == lk_tas) &&
2351 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2352 lck = (kmp_user_lock_p)user_lock;
2355 else if ((__kmp_user_lock_kind == lk_futex) &&
2356 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2357 lck = (kmp_user_lock_p)user_lock;
2361 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_lock");
2364 #if OMPT_SUPPORT && OMPT_OPTIONAL 2366 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2368 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2369 if (ompt_enabled.ompt_callback_lock_destroy) {
2370 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2371 ompt_mutex_lock, (omp_wait_id_t)user_lock, codeptr);
2376 __kmp_itt_lock_destroyed(lck);
2380 if ((__kmp_user_lock_kind == lk_tas) &&
2381 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2385 else if ((__kmp_user_lock_kind == lk_futex) &&
2386 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2391 __kmp_user_lock_free(user_lock, gtid, lck);
2393 #endif // KMP_USE_DYNAMIC_LOCK 2397 void __kmpc_destroy_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2398 #if KMP_USE_DYNAMIC_LOCK 2401 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(user_lock);
2402 __kmp_itt_lock_destroyed(ilk->lock);
2404 #if OMPT_SUPPORT && OMPT_OPTIONAL 2406 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2408 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2409 if (ompt_enabled.ompt_callback_lock_destroy) {
2410 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2411 ompt_mutex_nest_lock, (omp_wait_id_t)user_lock, codeptr);
2414 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2416 #else // KMP_USE_DYNAMIC_LOCK 2418 kmp_user_lock_p lck;
2420 if ((__kmp_user_lock_kind == lk_tas) &&
2421 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2422 OMP_NEST_LOCK_T_SIZE)) {
2423 lck = (kmp_user_lock_p)user_lock;
2426 else if ((__kmp_user_lock_kind == lk_futex) &&
2427 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2428 OMP_NEST_LOCK_T_SIZE)) {
2429 lck = (kmp_user_lock_p)user_lock;
2433 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_nest_lock");
2436 #if OMPT_SUPPORT && OMPT_OPTIONAL 2438 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2440 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2441 if (ompt_enabled.ompt_callback_lock_destroy) {
2442 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2443 ompt_mutex_nest_lock, (omp_wait_id_t)user_lock, codeptr);
2448 __kmp_itt_lock_destroyed(lck);
2451 DESTROY_NESTED_LOCK(lck);
2453 if ((__kmp_user_lock_kind == lk_tas) &&
2454 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2455 OMP_NEST_LOCK_T_SIZE)) {
2459 else if ((__kmp_user_lock_kind == lk_futex) &&
2460 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2461 OMP_NEST_LOCK_T_SIZE)) {
2466 __kmp_user_lock_free(user_lock, gtid, lck);
2468 #endif // KMP_USE_DYNAMIC_LOCK 2471 void __kmpc_set_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2473 #if KMP_USE_DYNAMIC_LOCK 2474 int tag = KMP_EXTRACT_D_TAG(user_lock);
2476 __kmp_itt_lock_acquiring(
2480 #if OMPT_SUPPORT && OMPT_OPTIONAL 2482 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2484 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2485 if (ompt_enabled.ompt_callback_mutex_acquire) {
2486 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2487 ompt_mutex_lock, omp_lock_hint_none,
2488 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2492 #if KMP_USE_INLINED_TAS 2493 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2494 KMP_ACQUIRE_TAS_LOCK(user_lock, gtid);
2496 #elif KMP_USE_INLINED_FUTEX 2497 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2498 KMP_ACQUIRE_FUTEX_LOCK(user_lock, gtid);
2502 __kmp_direct_set[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2505 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2507 #if OMPT_SUPPORT && OMPT_OPTIONAL 2508 if (ompt_enabled.ompt_callback_mutex_acquired) {
2509 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2510 ompt_mutex_lock, (omp_wait_id_t)user_lock, codeptr);
2514 #else // KMP_USE_DYNAMIC_LOCK 2516 kmp_user_lock_p lck;
2518 if ((__kmp_user_lock_kind == lk_tas) &&
2519 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2520 lck = (kmp_user_lock_p)user_lock;
2523 else if ((__kmp_user_lock_kind == lk_futex) &&
2524 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2525 lck = (kmp_user_lock_p)user_lock;
2529 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_lock");
2533 __kmp_itt_lock_acquiring(lck);
2535 #if OMPT_SUPPORT && OMPT_OPTIONAL 2537 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2539 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2540 if (ompt_enabled.ompt_callback_mutex_acquire) {
2541 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2542 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2543 (omp_wait_id_t)lck, codeptr);
2547 ACQUIRE_LOCK(lck, gtid);
2550 __kmp_itt_lock_acquired(lck);
2553 #if OMPT_SUPPORT && OMPT_OPTIONAL 2554 if (ompt_enabled.ompt_callback_mutex_acquired) {
2555 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2556 ompt_mutex_lock, (omp_wait_id_t)lck, codeptr);
2560 #endif // KMP_USE_DYNAMIC_LOCK 2563 void __kmpc_set_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2564 #if KMP_USE_DYNAMIC_LOCK 2567 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
2569 #if OMPT_SUPPORT && OMPT_OPTIONAL 2571 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2573 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2574 if (ompt_enabled.enabled) {
2575 if (ompt_enabled.ompt_callback_mutex_acquire) {
2576 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2577 ompt_mutex_nest_lock, omp_lock_hint_none,
2578 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2583 int acquire_status =
2584 KMP_D_LOCK_FUNC(user_lock,
set)((kmp_dyna_lock_t *)user_lock, gtid);
2585 (void) acquire_status;
2587 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2590 #if OMPT_SUPPORT && OMPT_OPTIONAL 2591 if (ompt_enabled.enabled) {
2592 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2593 if (ompt_enabled.ompt_callback_mutex_acquired) {
2595 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2596 ompt_mutex_nest_lock, (omp_wait_id_t)user_lock, codeptr);
2599 if (ompt_enabled.ompt_callback_nest_lock) {
2601 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2602 ompt_scope_begin, (omp_wait_id_t)user_lock, codeptr);
2608 #else // KMP_USE_DYNAMIC_LOCK 2610 kmp_user_lock_p lck;
2612 if ((__kmp_user_lock_kind == lk_tas) &&
2613 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2614 OMP_NEST_LOCK_T_SIZE)) {
2615 lck = (kmp_user_lock_p)user_lock;
2618 else if ((__kmp_user_lock_kind == lk_futex) &&
2619 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2620 OMP_NEST_LOCK_T_SIZE)) {
2621 lck = (kmp_user_lock_p)user_lock;
2625 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_nest_lock");
2629 __kmp_itt_lock_acquiring(lck);
2631 #if OMPT_SUPPORT && OMPT_OPTIONAL 2633 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2635 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2636 if (ompt_enabled.enabled) {
2637 if (ompt_enabled.ompt_callback_mutex_acquire) {
2638 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2639 ompt_mutex_nest_lock, omp_lock_hint_none,
2640 __ompt_get_mutex_impl_type(), (omp_wait_id_t)lck, codeptr);
2645 ACQUIRE_NESTED_LOCK(lck, gtid, &acquire_status);
2648 __kmp_itt_lock_acquired(lck);
2651 #if OMPT_SUPPORT && OMPT_OPTIONAL 2652 if (ompt_enabled.enabled) {
2653 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2654 if (ompt_enabled.ompt_callback_mutex_acquired) {
2656 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2657 ompt_mutex_nest_lock, (omp_wait_id_t)lck, codeptr);
2660 if (ompt_enabled.ompt_callback_nest_lock) {
2662 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2663 ompt_scope_begin, (omp_wait_id_t)lck, codeptr);
2669 #endif // KMP_USE_DYNAMIC_LOCK 2672 void __kmpc_unset_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2673 #if KMP_USE_DYNAMIC_LOCK 2675 int tag = KMP_EXTRACT_D_TAG(user_lock);
2677 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2679 #if KMP_USE_INLINED_TAS 2680 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2681 KMP_RELEASE_TAS_LOCK(user_lock, gtid);
2683 #elif KMP_USE_INLINED_FUTEX 2684 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2685 KMP_RELEASE_FUTEX_LOCK(user_lock, gtid);
2689 __kmp_direct_unset[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2692 #if OMPT_SUPPORT && OMPT_OPTIONAL 2694 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2696 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2697 if (ompt_enabled.ompt_callback_mutex_released) {
2698 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2699 ompt_mutex_lock, (omp_wait_id_t)user_lock, codeptr);
2703 #else // KMP_USE_DYNAMIC_LOCK 2705 kmp_user_lock_p lck;
2710 if ((__kmp_user_lock_kind == lk_tas) &&
2711 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2712 #if KMP_OS_LINUX && \ 2713 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) 2716 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2718 TCW_4(((kmp_user_lock_p)user_lock)->tas.lk.poll, 0);
2721 #if OMPT_SUPPORT && OMPT_OPTIONAL 2723 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2725 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2726 if (ompt_enabled.ompt_callback_mutex_released) {
2727 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2728 ompt_mutex_lock, (omp_wait_id_t)lck, codeptr);
2734 lck = (kmp_user_lock_p)user_lock;
2738 else if ((__kmp_user_lock_kind == lk_futex) &&
2739 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2740 lck = (kmp_user_lock_p)user_lock;
2744 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_lock");
2748 __kmp_itt_lock_releasing(lck);
2751 RELEASE_LOCK(lck, gtid);
2753 #if OMPT_SUPPORT && OMPT_OPTIONAL 2755 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2757 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2758 if (ompt_enabled.ompt_callback_mutex_released) {
2759 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2760 ompt_mutex_lock, (omp_wait_id_t)lck, codeptr);
2764 #endif // KMP_USE_DYNAMIC_LOCK 2768 void __kmpc_unset_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2769 #if KMP_USE_DYNAMIC_LOCK 2772 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2774 int release_status =
2775 KMP_D_LOCK_FUNC(user_lock, unset)((kmp_dyna_lock_t *)user_lock, gtid);
2776 (void) release_status;
2778 #if OMPT_SUPPORT && OMPT_OPTIONAL 2780 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2782 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2783 if (ompt_enabled.enabled) {
2784 if (release_status == KMP_LOCK_RELEASED) {
2785 if (ompt_enabled.ompt_callback_mutex_released) {
2787 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2788 ompt_mutex_nest_lock, (omp_wait_id_t)user_lock, codeptr);
2790 }
else if (ompt_enabled.ompt_callback_nest_lock) {
2792 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2793 ompt_scope_end, (omp_wait_id_t)user_lock, codeptr);
2798 #else // KMP_USE_DYNAMIC_LOCK 2800 kmp_user_lock_p lck;
2804 if ((__kmp_user_lock_kind == lk_tas) &&
2805 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2806 OMP_NEST_LOCK_T_SIZE)) {
2807 #if KMP_OS_LINUX && \ 2808 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) 2810 kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
2812 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2815 #if OMPT_SUPPORT && OMPT_OPTIONAL 2816 int release_status = KMP_LOCK_STILL_HELD;
2819 if (--(tl->lk.depth_locked) == 0) {
2820 TCW_4(tl->lk.poll, 0);
2821 #if OMPT_SUPPORT && OMPT_OPTIONAL 2822 release_status = KMP_LOCK_RELEASED;
2827 #if OMPT_SUPPORT && OMPT_OPTIONAL 2829 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2831 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2832 if (ompt_enabled.enabled) {
2833 if (release_status == KMP_LOCK_RELEASED) {
2834 if (ompt_enabled.ompt_callback_mutex_released) {
2836 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2837 ompt_mutex_nest_lock, (omp_wait_id_t)lck, codeptr);
2839 }
else if (ompt_enabled.ompt_callback_nest_lock) {
2841 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2842 ompt_mutex_scope_end, (omp_wait_id_t)lck, codeptr);
2849 lck = (kmp_user_lock_p)user_lock;
2853 else if ((__kmp_user_lock_kind == lk_futex) &&
2854 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2855 OMP_NEST_LOCK_T_SIZE)) {
2856 lck = (kmp_user_lock_p)user_lock;
2860 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_nest_lock");
2864 __kmp_itt_lock_releasing(lck);
2868 release_status = RELEASE_NESTED_LOCK(lck, gtid);
2869 #if OMPT_SUPPORT && OMPT_OPTIONAL 2871 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2873 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2874 if (ompt_enabled.enabled) {
2875 if (release_status == KMP_LOCK_RELEASED) {
2876 if (ompt_enabled.ompt_callback_mutex_released) {
2878 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2879 ompt_mutex_nest_lock, (omp_wait_id_t)lck, codeptr);
2881 }
else if (ompt_enabled.ompt_callback_nest_lock) {
2883 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2884 ompt_mutex_scope_end, (omp_wait_id_t)lck, codeptr);
2889 #endif // KMP_USE_DYNAMIC_LOCK 2893 int __kmpc_test_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2896 #if KMP_USE_DYNAMIC_LOCK 2898 int tag = KMP_EXTRACT_D_TAG(user_lock);
2900 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
2902 #if OMPT_SUPPORT && OMPT_OPTIONAL 2904 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2906 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2907 if (ompt_enabled.ompt_callback_mutex_acquire) {
2908 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2909 ompt_mutex_lock, omp_lock_hint_none,
2910 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
2914 #if KMP_USE_INLINED_TAS 2915 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2916 KMP_TEST_TAS_LOCK(user_lock, gtid, rc);
2918 #elif KMP_USE_INLINED_FUTEX 2919 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2920 KMP_TEST_FUTEX_LOCK(user_lock, gtid, rc);
2924 rc = __kmp_direct_test[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2928 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2930 #if OMPT_SUPPORT && OMPT_OPTIONAL 2931 if (ompt_enabled.ompt_callback_mutex_acquired) {
2932 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2933 ompt_mutex_lock, (omp_wait_id_t)user_lock, codeptr);
2939 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
2944 #else // KMP_USE_DYNAMIC_LOCK 2946 kmp_user_lock_p lck;
2949 if ((__kmp_user_lock_kind == lk_tas) &&
2950 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2951 lck = (kmp_user_lock_p)user_lock;
2954 else if ((__kmp_user_lock_kind == lk_futex) &&
2955 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2956 lck = (kmp_user_lock_p)user_lock;
2960 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_lock");
2964 __kmp_itt_lock_acquiring(lck);
2966 #if OMPT_SUPPORT && OMPT_OPTIONAL 2968 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2970 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2971 if (ompt_enabled.ompt_callback_mutex_acquire) {
2972 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2973 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2974 (omp_wait_id_t)lck, codeptr);
2978 rc = TEST_LOCK(lck, gtid);
2981 __kmp_itt_lock_acquired(lck);
2983 __kmp_itt_lock_cancelled(lck);
2986 #if OMPT_SUPPORT && OMPT_OPTIONAL 2987 if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
2988 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2989 ompt_mutex_lock, (omp_wait_id_t)lck, codeptr);
2993 return (rc ? FTN_TRUE : FTN_FALSE);
2997 #endif // KMP_USE_DYNAMIC_LOCK 3001 int __kmpc_test_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3002 #if KMP_USE_DYNAMIC_LOCK 3005 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3007 #if OMPT_SUPPORT && OMPT_OPTIONAL 3009 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3011 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3012 if (ompt_enabled.ompt_callback_mutex_acquire) {
3013 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3014 ompt_mutex_nest_lock, omp_lock_hint_none,
3015 __ompt_get_mutex_impl_type(user_lock), (omp_wait_id_t)user_lock,
3019 rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
3022 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3024 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3027 #if OMPT_SUPPORT && OMPT_OPTIONAL 3028 if (ompt_enabled.enabled && rc) {
3030 if (ompt_enabled.ompt_callback_mutex_acquired) {
3032 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3033 ompt_mutex_nest_lock, (omp_wait_id_t)user_lock, codeptr);
3036 if (ompt_enabled.ompt_callback_nest_lock) {
3038 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3039 ompt_scope_begin, (omp_wait_id_t)user_lock, codeptr);
3046 #else // KMP_USE_DYNAMIC_LOCK 3048 kmp_user_lock_p lck;
3051 if ((__kmp_user_lock_kind == lk_tas) &&
3052 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3053 OMP_NEST_LOCK_T_SIZE)) {
3054 lck = (kmp_user_lock_p)user_lock;
3057 else if ((__kmp_user_lock_kind == lk_futex) &&
3058 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3059 OMP_NEST_LOCK_T_SIZE)) {
3060 lck = (kmp_user_lock_p)user_lock;
3064 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_nest_lock");
3068 __kmp_itt_lock_acquiring(lck);
3071 #if OMPT_SUPPORT && OMPT_OPTIONAL 3073 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3075 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3076 if (ompt_enabled.enabled) &&
3077 ompt_enabled.ompt_callback_mutex_acquire) {
3078 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3079 ompt_mutex_nest_lock, omp_lock_hint_none,
3080 __ompt_get_mutex_impl_type(), (omp_wait_id_t)lck, codeptr);
3084 rc = TEST_NESTED_LOCK(lck, gtid);
3087 __kmp_itt_lock_acquired(lck);
3089 __kmp_itt_lock_cancelled(lck);
3092 #if OMPT_SUPPORT && OMPT_OPTIONAL 3093 if (ompt_enabled.enabled && rc) {
3095 if (ompt_enabled.ompt_callback_mutex_acquired) {
3097 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3098 ompt_mutex_nest_lock, (omp_wait_id_t)lck, codeptr);
3101 if (ompt_enabled.ompt_callback_nest_lock) {
3103 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3104 ompt_mutex_scope_begin, (omp_wait_id_t)lck, codeptr);
3113 #endif // KMP_USE_DYNAMIC_LOCK 3123 #define __KMP_SET_REDUCTION_METHOD(gtid, rmethod) \ 3124 ((__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) = (rmethod)) 3126 #define __KMP_GET_REDUCTION_METHOD(gtid) \ 3127 (__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) 3133 static __forceinline
void 3134 __kmp_enter_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3135 kmp_critical_name *crit) {
3141 kmp_user_lock_p lck;
3143 #if KMP_USE_DYNAMIC_LOCK 3145 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
3148 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3149 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
3150 KMP_GET_D_TAG(__kmp_user_lock_seq));
3152 __kmp_init_indirect_csptr(crit, loc, global_tid,
3153 KMP_GET_I_TAG(__kmp_user_lock_seq));
3159 if (KMP_EXTRACT_D_TAG(lk) != 0) {
3160 lck = (kmp_user_lock_p)lk;
3161 KMP_DEBUG_ASSERT(lck != NULL);
3162 if (__kmp_env_consistency_check) {
3163 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3165 KMP_D_LOCK_FUNC(lk,
set)(lk, global_tid);
3167 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
3169 KMP_DEBUG_ASSERT(lck != NULL);
3170 if (__kmp_env_consistency_check) {
3171 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3173 KMP_I_LOCK_FUNC(ilk,
set)(lck, global_tid);
3176 #else // KMP_USE_DYNAMIC_LOCK 3181 if (__kmp_base_user_lock_size <= INTEL_CRITICAL_SIZE) {
3182 lck = (kmp_user_lock_p)crit;
3184 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
3186 KMP_DEBUG_ASSERT(lck != NULL);
3188 if (__kmp_env_consistency_check)
3189 __kmp_push_sync(global_tid, ct_critical, loc, lck);
3191 __kmp_acquire_user_lock_with_checks(lck, global_tid);
3193 #endif // KMP_USE_DYNAMIC_LOCK 3197 static __forceinline
void 3198 __kmp_end_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3199 kmp_critical_name *crit) {
3201 kmp_user_lock_p lck;
3203 #if KMP_USE_DYNAMIC_LOCK 3205 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3206 lck = (kmp_user_lock_p)crit;
3207 if (__kmp_env_consistency_check)
3208 __kmp_pop_sync(global_tid, ct_critical, loc);
3209 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
3211 kmp_indirect_lock_t *ilk =
3212 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
3213 if (__kmp_env_consistency_check)
3214 __kmp_pop_sync(global_tid, ct_critical, loc);
3215 KMP_I_LOCK_FUNC(ilk, unset)(ilk->lock, global_tid);
3218 #else // KMP_USE_DYNAMIC_LOCK 3223 if (__kmp_base_user_lock_size > 32) {
3224 lck = *((kmp_user_lock_p *)crit);
3225 KMP_ASSERT(lck != NULL);
3227 lck = (kmp_user_lock_p)crit;
3230 if (__kmp_env_consistency_check)
3231 __kmp_pop_sync(global_tid, ct_critical, loc);
3233 __kmp_release_user_lock_with_checks(lck, global_tid);
3235 #endif // KMP_USE_DYNAMIC_LOCK 3239 static __forceinline
int 3240 __kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
3245 if (th->th.th_teams_microtask) {
3246 *team_p = team = th->th.th_team;
3247 if (team->t.t_level == th->th.th_teams_level) {
3249 KMP_DEBUG_ASSERT(!th->th.th_info.ds.ds_tid);
3251 th->th.th_info.ds.ds_tid = team->t.t_master_tid;
3252 th->th.th_team = team->t.t_parent;
3253 th->th.th_team_nproc = th->th.th_team->t.t_nproc;
3254 th->th.th_task_team = th->th.th_team->t.t_task_team[0];
3255 *task_state = th->th.th_task_state;
3256 th->th.th_task_state = 0;
3264 static __forceinline
void 3265 __kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team,
int task_state) {
3267 th->th.th_info.ds.ds_tid = 0;
3268 th->th.th_team = team;
3269 th->th.th_team_nproc = team->t.t_nproc;
3270 th->th.th_task_team = team->t.t_task_team[task_state];
3271 th->th.th_task_state = task_state;
3293 size_t reduce_size,
void *reduce_data,
3294 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3295 kmp_critical_name *lck) {
3299 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3303 int teams_swapped = 0, task_state;
3305 KA_TRACE(10, (
"__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
3313 if (!TCR_4(__kmp_init_parallel))
3314 __kmp_parallel_initialize();
3317 #if KMP_USE_DYNAMIC_LOCK 3318 if (__kmp_env_consistency_check)
3319 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3321 if (__kmp_env_consistency_check)
3322 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3326 th = __kmp_thread_from_gtid(global_tid);
3327 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3328 #endif // OMP_40_ENABLED 3346 packed_reduction_method = __kmp_determine_reduction_method(
3347 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3348 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3350 if (packed_reduction_method == critical_reduce_block) {
3352 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3355 }
else if (packed_reduction_method == empty_reduce_block) {
3361 }
else if (packed_reduction_method == atomic_reduce_block) {
3371 if (__kmp_env_consistency_check)
3372 __kmp_pop_sync(global_tid, ct_reduce, loc);
3374 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3375 tree_reduce_block)) {
3395 omp_frame_t *ompt_frame;
3396 if (ompt_enabled.enabled) {
3397 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3398 if (ompt_frame->enter_frame == NULL)
3399 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
3400 OMPT_STORE_RETURN_ADDRESS(global_tid);
3404 __kmp_threads[global_tid]->th.th_ident = loc;
3407 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3408 global_tid, FALSE, reduce_size, reduce_data, reduce_func);
3409 retval = (retval != 0) ? (0) : (1);
3410 #if OMPT_SUPPORT && OMPT_OPTIONAL 3411 if (ompt_enabled.enabled) {
3412 ompt_frame->enter_frame = NULL;
3418 if (__kmp_env_consistency_check) {
3420 __kmp_pop_sync(global_tid, ct_reduce, loc);
3430 if (teams_swapped) {
3431 __kmp_restore_swapped_teams(th, team, task_state);
3436 (
"__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
3437 global_tid, packed_reduction_method, retval));
3451 kmp_critical_name *lck) {
3453 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3455 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() enter: called T#%d\n", global_tid));
3457 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3459 if (packed_reduction_method == critical_reduce_block) {
3461 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3463 }
else if (packed_reduction_method == empty_reduce_block) {
3468 }
else if (packed_reduction_method == atomic_reduce_block) {
3475 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3476 tree_reduce_block)) {
3486 if (__kmp_env_consistency_check)
3487 __kmp_pop_sync(global_tid, ct_reduce, loc);
3489 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() exit: called T#%d: method %08x\n",
3490 global_tid, packed_reduction_method));
3513 size_t reduce_size,
void *reduce_data,
3514 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3515 kmp_critical_name *lck) {
3518 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3522 int teams_swapped = 0, task_state;
3525 KA_TRACE(10, (
"__kmpc_reduce() enter: called T#%d\n", global_tid));
3533 if (!TCR_4(__kmp_init_parallel))
3534 __kmp_parallel_initialize();
3537 #if KMP_USE_DYNAMIC_LOCK 3538 if (__kmp_env_consistency_check)
3539 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3541 if (__kmp_env_consistency_check)
3542 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3546 th = __kmp_thread_from_gtid(global_tid);
3547 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3548 #endif // OMP_40_ENABLED 3550 packed_reduction_method = __kmp_determine_reduction_method(
3551 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3552 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3554 if (packed_reduction_method == critical_reduce_block) {
3556 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3559 }
else if (packed_reduction_method == empty_reduce_block) {
3565 }
else if (packed_reduction_method == atomic_reduce_block) {
3569 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3570 tree_reduce_block)) {
3576 omp_frame_t *ompt_frame;
3577 if (ompt_enabled.enabled) {
3578 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3579 if (ompt_frame->enter_frame == NULL)
3580 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
3581 OMPT_STORE_RETURN_ADDRESS(global_tid);
3585 __kmp_threads[global_tid]->th.th_ident =
3589 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3590 global_tid, TRUE, reduce_size, reduce_data, reduce_func);
3591 retval = (retval != 0) ? (0) : (1);
3592 #if OMPT_SUPPORT && OMPT_OPTIONAL 3593 if (ompt_enabled.enabled) {
3594 ompt_frame->enter_frame = NULL;
3600 if (__kmp_env_consistency_check) {
3602 __kmp_pop_sync(global_tid, ct_reduce, loc);
3612 if (teams_swapped) {
3613 __kmp_restore_swapped_teams(th, team, task_state);
3618 (
"__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
3619 global_tid, packed_reduction_method, retval));
3635 kmp_critical_name *lck) {
3637 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3641 int teams_swapped = 0, task_state;
3644 KA_TRACE(10, (
"__kmpc_end_reduce() enter: called T#%d\n", global_tid));
3647 th = __kmp_thread_from_gtid(global_tid);
3648 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3649 #endif // OMP_40_ENABLED 3651 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3656 if (packed_reduction_method == critical_reduce_block) {
3658 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3662 omp_frame_t *ompt_frame;
3663 if (ompt_enabled.enabled) {
3664 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3665 if (ompt_frame->enter_frame == NULL)
3666 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
3667 OMPT_STORE_RETURN_ADDRESS(global_tid);
3671 __kmp_threads[global_tid]->th.th_ident = loc;
3673 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3674 #if OMPT_SUPPORT && OMPT_OPTIONAL 3675 if (ompt_enabled.enabled) {
3676 ompt_frame->enter_frame = NULL;
3680 }
else if (packed_reduction_method == empty_reduce_block) {
3686 omp_frame_t *ompt_frame;
3687 if (ompt_enabled.enabled) {
3688 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3689 if (ompt_frame->enter_frame == NULL)
3690 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
3691 OMPT_STORE_RETURN_ADDRESS(global_tid);
3695 __kmp_threads[global_tid]->th.th_ident = loc;
3697 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3698 #if OMPT_SUPPORT && OMPT_OPTIONAL 3699 if (ompt_enabled.enabled) {
3700 ompt_frame->enter_frame = NULL;
3704 }
else if (packed_reduction_method == atomic_reduce_block) {
3707 omp_frame_t *ompt_frame;
3708 if (ompt_enabled.enabled) {
3709 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3710 if (ompt_frame->enter_frame == NULL)
3711 ompt_frame->enter_frame = OMPT_GET_FRAME_ADDRESS(1);
3712 OMPT_STORE_RETURN_ADDRESS(global_tid);
3717 __kmp_threads[global_tid]->th.th_ident = loc;
3719 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3720 #if OMPT_SUPPORT && OMPT_OPTIONAL 3721 if (ompt_enabled.enabled) {
3722 ompt_frame->enter_frame = NULL;
3726 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3727 tree_reduce_block)) {
3730 __kmp_end_split_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3739 if (teams_swapped) {
3740 __kmp_restore_swapped_teams(th, team, task_state);
3744 if (__kmp_env_consistency_check)
3745 __kmp_pop_sync(global_tid, ct_reduce, loc);
3747 KA_TRACE(10, (
"__kmpc_end_reduce() exit: called T#%d: method %08x\n",
3748 global_tid, packed_reduction_method));
3753 #undef __KMP_GET_REDUCTION_METHOD 3754 #undef __KMP_SET_REDUCTION_METHOD 3758 kmp_uint64 __kmpc_get_taskid() {
3763 gtid = __kmp_get_gtid();
3767 thread = __kmp_thread_from_gtid(gtid);
3768 return thread->th.th_current_task->td_task_id;
3772 kmp_uint64 __kmpc_get_parent_taskid() {
3776 kmp_taskdata_t *parent_task;
3778 gtid = __kmp_get_gtid();
3782 thread = __kmp_thread_from_gtid(gtid);
3783 parent_task = thread->th.th_current_task->td_parent;
3784 return (parent_task == NULL ? 0 : parent_task->td_task_id);
3800 void __kmpc_doacross_init(
ident_t *loc,
int gtid,
int num_dims,
3801 const struct kmp_dim *dims) {
3803 kmp_int64 last, trace_count;
3804 kmp_info_t *th = __kmp_threads[gtid];
3805 kmp_team_t *team = th->th.th_team;
3807 kmp_disp_t *pr_buf = th->th.th_dispatch;
3808 dispatch_shared_info_t *sh_buf;
3812 (
"__kmpc_doacross_init() enter: called T#%d, num dims %d, active %d\n",
3813 gtid, num_dims, !team->t.t_serialized));
3814 KMP_DEBUG_ASSERT(dims != NULL);
3815 KMP_DEBUG_ASSERT(num_dims > 0);
3817 if (team->t.t_serialized) {
3818 KA_TRACE(20, (
"__kmpc_doacross_init() exit: serialized team\n"));
3821 KMP_DEBUG_ASSERT(team->t.t_nproc > 1);
3822 idx = pr_buf->th_doacross_buf_idx++;
3824 sh_buf = &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
3827 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info == NULL);
3828 pr_buf->th_doacross_info = (kmp_int64 *)__kmp_thread_malloc(
3829 th,
sizeof(kmp_int64) * (4 * num_dims + 1));
3830 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
3831 pr_buf->th_doacross_info[0] =
3832 (kmp_int64)num_dims;
3835 pr_buf->th_doacross_info[1] = (kmp_int64)&sh_buf->doacross_num_done;
3836 pr_buf->th_doacross_info[2] = dims[0].lo;
3837 pr_buf->th_doacross_info[3] = dims[0].up;
3838 pr_buf->th_doacross_info[4] = dims[0].st;
3840 for (j = 1; j < num_dims; ++j) {
3843 if (dims[j].st == 1) {
3845 range_length = dims[j].up - dims[j].lo + 1;
3847 if (dims[j].st > 0) {
3848 KMP_DEBUG_ASSERT(dims[j].up > dims[j].lo);
3849 range_length = (kmp_uint64)(dims[j].up - dims[j].lo) / dims[j].st + 1;
3851 KMP_DEBUG_ASSERT(dims[j].lo > dims[j].up);
3853 (kmp_uint64)(dims[j].lo - dims[j].up) / (-dims[j].st) + 1;
3856 pr_buf->th_doacross_info[last++] = range_length;
3857 pr_buf->th_doacross_info[last++] = dims[j].lo;
3858 pr_buf->th_doacross_info[last++] = dims[j].up;
3859 pr_buf->th_doacross_info[last++] = dims[j].st;
3864 if (dims[0].st == 1) {
3865 trace_count = dims[0].up - dims[0].lo + 1;
3866 }
else if (dims[0].st > 0) {
3867 KMP_DEBUG_ASSERT(dims[0].up > dims[0].lo);
3868 trace_count = (kmp_uint64)(dims[0].up - dims[0].lo) / dims[0].st + 1;
3870 KMP_DEBUG_ASSERT(dims[0].lo > dims[0].up);
3871 trace_count = (kmp_uint64)(dims[0].lo - dims[0].up) / (-dims[0].st) + 1;
3873 for (j = 1; j < num_dims; ++j) {
3874 trace_count *= pr_buf->th_doacross_info[4 * j + 1];
3876 KMP_DEBUG_ASSERT(trace_count > 0);
3880 if (idx != sh_buf->doacross_buf_idx) {
3882 __kmp_wait_yield_4((
volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
3889 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET32(
3890 (
volatile kmp_int32 *)&sh_buf->doacross_flags, NULL, 1);
3892 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET64(
3893 (
volatile kmp_int64 *)&sh_buf->doacross_flags, NULL, 1LL);
3895 if (flags == NULL) {
3897 size_t size = trace_count / 8 + 8;
3898 flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
3900 sh_buf->doacross_flags = flags;
3901 }
else if (flags == (kmp_uint32 *)1) {
3904 while (*(
volatile kmp_int32 *)&sh_buf->doacross_flags == 1)
3906 while (*(
volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
3913 KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1);
3914 pr_buf->th_doacross_flags =
3915 sh_buf->doacross_flags;
3917 KA_TRACE(20, (
"__kmpc_doacross_init() exit: T#%d\n", gtid));
3920 void __kmpc_doacross_wait(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
3921 kmp_int32 shft, num_dims, i;
3923 kmp_int64 iter_number;
3924 kmp_info_t *th = __kmp_threads[gtid];
3925 kmp_team_t *team = th->th.th_team;
3927 kmp_int64 lo, up, st;
3929 KA_TRACE(20, (
"__kmpc_doacross_wait() enter: called T#%d\n", gtid));
3930 if (team->t.t_serialized) {
3931 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: serialized team\n"));
3936 pr_buf = th->th.th_dispatch;
3937 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
3938 num_dims = pr_buf->th_doacross_info[0];
3939 lo = pr_buf->th_doacross_info[2];
3940 up = pr_buf->th_doacross_info[3];
3941 st = pr_buf->th_doacross_info[4];
3943 if (vec[0] < lo || vec[0] > up) {
3944 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of " 3945 "bounds [%lld,%lld]\n",
3946 gtid, vec[0], lo, up));
3949 iter_number = vec[0] - lo;
3950 }
else if (st > 0) {
3951 if (vec[0] < lo || vec[0] > up) {
3952 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of " 3953 "bounds [%lld,%lld]\n",
3954 gtid, vec[0], lo, up));
3957 iter_number = (kmp_uint64)(vec[0] - lo) / st;
3959 if (vec[0] > lo || vec[0] < up) {
3960 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of " 3961 "bounds [%lld,%lld]\n",
3962 gtid, vec[0], lo, up));
3965 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
3967 for (i = 1; i < num_dims; ++i) {
3969 kmp_int32 j = i * 4;
3970 ln = pr_buf->th_doacross_info[j + 1];
3971 lo = pr_buf->th_doacross_info[j + 2];
3972 up = pr_buf->th_doacross_info[j + 3];
3973 st = pr_buf->th_doacross_info[j + 4];
3975 if (vec[i] < lo || vec[i] > up) {
3976 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of " 3977 "bounds [%lld,%lld]\n",
3978 gtid, vec[i], lo, up));
3982 }
else if (st > 0) {
3983 if (vec[i] < lo || vec[i] > up) {
3984 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of " 3985 "bounds [%lld,%lld]\n",
3986 gtid, vec[i], lo, up));
3989 iter = (kmp_uint64)(vec[i] - lo) / st;
3991 if (vec[i] > lo || vec[i] < up) {
3992 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of " 3993 "bounds [%lld,%lld]\n",
3994 gtid, vec[i], lo, up));
3997 iter = (kmp_uint64)(lo - vec[i]) / (-st);
3999 iter_number = iter + ln * iter_number;
4001 shft = iter_number % 32;
4004 while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
4009 (
"__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
4010 gtid, (iter_number << 5) + shft));
4013 void __kmpc_doacross_post(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4014 kmp_int32 shft, num_dims, i;
4016 kmp_int64 iter_number;
4017 kmp_info_t *th = __kmp_threads[gtid];
4018 kmp_team_t *team = th->th.th_team;
4022 KA_TRACE(20, (
"__kmpc_doacross_post() enter: called T#%d\n", gtid));
4023 if (team->t.t_serialized) {
4024 KA_TRACE(20, (
"__kmpc_doacross_post() exit: serialized team\n"));
4030 pr_buf = th->th.th_dispatch;
4031 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4032 num_dims = pr_buf->th_doacross_info[0];
4033 lo = pr_buf->th_doacross_info[2];
4034 st = pr_buf->th_doacross_info[4];
4036 iter_number = vec[0] - lo;
4037 }
else if (st > 0) {
4038 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4040 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4042 for (i = 1; i < num_dims; ++i) {
4044 kmp_int32 j = i * 4;
4045 ln = pr_buf->th_doacross_info[j + 1];
4046 lo = pr_buf->th_doacross_info[j + 2];
4047 st = pr_buf->th_doacross_info[j + 4];
4050 }
else if (st > 0) {
4051 iter = (kmp_uint64)(vec[i] - lo) / st;
4053 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4055 iter_number = iter + ln * iter_number;
4057 shft = iter_number % 32;
4061 if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
4062 KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
4063 KA_TRACE(20, (
"__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
4064 (iter_number << 5) + shft));
4067 void __kmpc_doacross_fini(
ident_t *loc,
int gtid) {
4069 kmp_info_t *th = __kmp_threads[gtid];
4070 kmp_team_t *team = th->th.th_team;
4071 kmp_disp_t *pr_buf = th->th.th_dispatch;
4073 KA_TRACE(20, (
"__kmpc_doacross_fini() enter: called T#%d\n", gtid));
4074 if (team->t.t_serialized) {
4075 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: serialized team %p\n", team));
4078 num_done = KMP_TEST_THEN_INC32((kmp_int32 *)pr_buf->th_doacross_info[1]) + 1;
4079 if (num_done == th->th.th_team_nproc) {
4081 int idx = pr_buf->th_doacross_buf_idx - 1;
4082 dispatch_shared_info_t *sh_buf =
4083 &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4084 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info[1] ==
4085 (kmp_int64)&sh_buf->doacross_num_done);
4086 KMP_DEBUG_ASSERT(num_done == sh_buf->doacross_num_done);
4087 KMP_DEBUG_ASSERT(idx == sh_buf->doacross_buf_idx);
4088 __kmp_thread_free(th, CCAST(kmp_uint32 *, sh_buf->doacross_flags));
4089 sh_buf->doacross_flags = NULL;
4090 sh_buf->doacross_num_done = 0;
4091 sh_buf->doacross_buf_idx +=
4092 __kmp_dispatch_num_buffers;
4095 pr_buf->th_doacross_flags = NULL;
4096 __kmp_thread_free(th, (
void *)pr_buf->th_doacross_info);
4097 pr_buf->th_doacross_info = NULL;
4098 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: T#%d\n", gtid));
4103 int __kmpc_get_target_offload(
void) {
4104 if (!__kmp_init_serial) {
4105 __kmp_serial_initialize();
4107 return __kmp_target_offload;
4109 #endif // OMP_50_ENABLED kmp_int32 __kmpc_master(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_barrier_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_single(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void(*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck)
kmp_int32 __kmpc_global_thread_num(ident_t *loc)
void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid)
void __kmpc_flush(ident_t *loc)
kmp_int32 __kmpc_single(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end(ident_t *loc)
void __kmpc_end_ordered(ident_t *loc, kmp_int32 gtid)
void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid)
void __kmpc_begin(ident_t *loc, kmp_int32 flags)
kmp_int32 __kmpc_bound_thread_num(ident_t *loc)
kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void(*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck)
void __kmpc_copyprivate(ident_t *loc, kmp_int32 gtid, size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *), kmp_int32 didit)
void __kmpc_ordered(ident_t *loc, kmp_int32 gtid)
#define KMP_COUNT_BLOCK(name)
Increments specified counter (name).
void __kmpc_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)
void __kmpc_end_barrier_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_threads)
void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)
kmp_int32 __kmpc_in_parallel(ident_t *loc)
kmp_int32 __kmpc_ok_to_fork(ident_t *loc)
kmp_int32 __kmpc_global_num_threads(ident_t *loc)
kmp_int32 __kmpc_bound_num_threads(ident_t *loc)
void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)
void __kmpc_push_num_teams(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams, kmp_int32 num_threads)
void(* kmpc_micro)(kmp_int32 *global_tid, kmp_int32 *bound_tid,...)
kmp_int32 __kmpc_barrier_master_nowait(ident_t *loc, kmp_int32 global_tid)
void __kmpc_serialized_parallel(ident_t *loc, kmp_int32 global_tid)
stats_state_e
the states which a thread can be in
void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)