21#include "ompt-specific.h"
23#define MAX_MESSAGE 512
39 if ((env = getenv(
"KMP_INITIAL_THREAD_BIND")) != NULL &&
40 __kmp_str_match_true(env)) {
41 __kmp_middle_initialize();
42 __kmp_assign_root_init_mask();
43 KC_TRACE(10, (
"__kmpc_begin: middle initialization called\n"));
44 }
else if (__kmp_ignore_mppbeg() == FALSE) {
46 __kmp_internal_begin();
47 KC_TRACE(10, (
"__kmpc_begin: called\n"));
65 if (__kmp_ignore_mppend() == FALSE) {
66 KC_TRACE(10, (
"__kmpc_end: called\n"));
67 KA_TRACE(30, (
"__kmpc_end\n"));
69 __kmp_internal_end_thread(-1);
71#if KMP_OS_WINDOWS && OMPT_SUPPORT
76 if (ompt_enabled.enabled)
77 __kmp_internal_end_library(__kmp_gtid_get_specific());
100 kmp_int32 gtid = __kmp_entry_gtid();
102 KC_TRACE(10, (
"__kmpc_global_thread_num: T#%d\n", gtid));
123 (
"__kmpc_global_num_threads: num_threads = %d\n", __kmp_all_nth));
125 return TCR_4(__kmp_all_nth);
135 KC_TRACE(10, (
"__kmpc_bound_thread_num: called\n"));
136 return __kmp_tid_from_gtid(__kmp_entry_gtid());
145 KC_TRACE(10, (
"__kmpc_bound_num_threads: called\n"));
147 return __kmp_entry_thread()->th.th_team->t.t_nproc;
167 if (__kmp_par_range == 0) {
174 semi2 = strchr(semi2,
';');
178 semi2 = strchr(semi2 + 1,
';');
182 if (__kmp_par_range_filename[0]) {
183 const char *name = semi2 - 1;
184 while ((name > loc->
psource) && (*name !=
'/') && (*name !=
';')) {
187 if ((*name ==
'/') || (*name ==
';')) {
190 if (strncmp(__kmp_par_range_filename, name, semi2 - name)) {
191 return __kmp_par_range < 0;
194 semi3 = strchr(semi2 + 1,
';');
195 if (__kmp_par_range_routine[0]) {
196 if ((semi3 != NULL) && (semi3 > semi2) &&
197 (strncmp(__kmp_par_range_routine, semi2 + 1, semi3 - semi2 - 1))) {
198 return __kmp_par_range < 0;
201 if (KMP_SSCANF(semi3 + 1,
"%d", &line_no) == 1) {
202 if ((line_no >= __kmp_par_range_lb) && (line_no <= __kmp_par_range_ub)) {
203 return __kmp_par_range > 0;
205 return __kmp_par_range < 0;
219 return __kmp_entry_thread()->th.th_root->r.r_active;
232 kmp_int32 num_threads) {
233 KA_TRACE(20, (
"__kmpc_push_num_threads: enter T#%d num_threads=%d\n",
234 global_tid, num_threads));
235 __kmp_assert_valid_gtid(global_tid);
236 __kmp_push_num_threads(loc, global_tid, num_threads);
239void __kmpc_pop_num_threads(
ident_t *loc, kmp_int32 global_tid) {
240 KA_TRACE(20, (
"__kmpc_pop_num_threads: enter\n"));
244void __kmpc_push_proc_bind(
ident_t *loc, kmp_int32 global_tid,
245 kmp_int32 proc_bind) {
246 KA_TRACE(20, (
"__kmpc_push_proc_bind: enter T#%d proc_bind=%d\n", global_tid,
248 __kmp_assert_valid_gtid(global_tid);
249 __kmp_push_proc_bind(loc, global_tid, (kmp_proc_bind_t)proc_bind);
263 int gtid = __kmp_entry_gtid();
265#if (KMP_STATS_ENABLED)
269 if (previous_state == stats_state_e::SERIAL_REGION) {
270 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_parallel_overhead);
272 KMP_PUSH_PARTITIONED_TIMER(OMP_parallel_overhead);
285 va_start(ap, microtask);
288 ompt_frame_t *ompt_frame;
289 if (ompt_enabled.enabled) {
290 kmp_info_t *master_th = __kmp_threads[gtid];
291 ompt_frame = &master_th->th.th_current_task->ompt_task_info.frame;
292 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
294 OMPT_STORE_RETURN_ADDRESS(gtid);
300 __kmp_fork_call(loc, gtid, fork_context_intel, argc,
301 VOLATILE_CAST(microtask_t) microtask,
302 VOLATILE_CAST(launch_t) __kmp_invoke_task_func,
307 __kmp_join_call(loc, gtid
317 if (ompt_enabled.enabled) {
318 ompt_frame->enter_frame = ompt_data_none;
324 if (previous_state == stats_state_e::SERIAL_REGION) {
325 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
326 KMP_SET_THREAD_STATE(previous_state);
328 KMP_POP_PARTITIONED_TIMER();
345 kmp_int32 num_teams, kmp_int32 num_threads) {
347 (
"__kmpc_push_num_teams: enter T#%d num_teams=%d num_threads=%d\n",
348 global_tid, num_teams, num_threads));
349 __kmp_assert_valid_gtid(global_tid);
350 __kmp_push_num_teams(loc, global_tid, num_teams, num_threads);
370 kmp_int32 num_teams_lb, kmp_int32 num_teams_ub,
371 kmp_int32 num_threads) {
372 KA_TRACE(20, (
"__kmpc_push_num_teams_51: enter T#%d num_teams_lb=%d"
373 " num_teams_ub=%d num_threads=%d\n",
374 global_tid, num_teams_lb, num_teams_ub, num_threads));
375 __kmp_assert_valid_gtid(global_tid);
376 __kmp_push_num_teams_51(loc, global_tid, num_teams_lb, num_teams_ub,
392 int gtid = __kmp_entry_gtid();
393 kmp_info_t *this_thr = __kmp_threads[gtid];
395 va_start(ap, microtask);
400 if (previous_state == stats_state_e::SERIAL_REGION) {
401 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_teams_overhead);
403 KMP_PUSH_PARTITIONED_TIMER(OMP_teams_overhead);
408 this_thr->th.th_teams_microtask = microtask;
409 this_thr->th.th_teams_level =
410 this_thr->th.th_team->t.t_level;
413 kmp_team_t *parent_team = this_thr->th.th_team;
414 int tid = __kmp_tid_from_gtid(gtid);
415 if (ompt_enabled.enabled) {
416 parent_team->t.t_implicit_task_taskdata[tid]
417 .ompt_task_info.frame.enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
419 OMPT_STORE_RETURN_ADDRESS(gtid);
424 if (this_thr->th.th_teams_size.nteams == 0) {
425 __kmp_push_num_teams(loc, gtid, 0, 0);
427 KMP_DEBUG_ASSERT(this_thr->th.th_set_nproc >= 1);
428 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nteams >= 1);
429 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nth >= 1);
432 loc, gtid, fork_context_intel, argc,
433 VOLATILE_CAST(microtask_t) __kmp_teams_master,
434 VOLATILE_CAST(launch_t) __kmp_invoke_teams_master, kmp_va_addr_of(ap));
435 __kmp_join_call(loc, gtid
443 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
444 kmp_cg_root_t *tmp = this_thr->th.th_cg_roots;
445 this_thr->th.th_cg_roots = tmp->up;
446 KA_TRACE(100, (
"__kmpc_fork_teams: Thread %p popping node %p and moving up"
447 " to node %p. cg_nthreads was %d\n",
448 this_thr, tmp, this_thr->th.th_cg_roots, tmp->cg_nthreads));
449 KMP_DEBUG_ASSERT(tmp->cg_nthreads);
450 int i = tmp->cg_nthreads--;
455 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
456 this_thr->th.th_current_task->td_icvs.thread_limit =
457 this_thr->th.th_cg_roots->cg_thread_limit;
459 this_thr->th.th_teams_microtask = NULL;
460 this_thr->th.th_teams_level = 0;
461 *(kmp_int64 *)(&this_thr->th.th_teams_size) = 0L;
464 if (previous_state == stats_state_e::SERIAL_REGION) {
465 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
466 KMP_SET_THREAD_STATE(previous_state);
468 KMP_POP_PARTITIONED_TIMER();
477int __kmpc_invoke_task_func(
int gtid) {
return __kmp_invoke_task_func(gtid); }
495 __kmp_assert_valid_gtid(global_tid);
497 OMPT_STORE_RETURN_ADDRESS(global_tid);
499 __kmp_serialized_parallel(loc, global_tid);
510 kmp_internal_control_t *top;
511 kmp_info_t *this_thr;
512 kmp_team_t *serial_team;
515 (
"__kmpc_end_serialized_parallel: called by T#%d\n", global_tid));
523 __kmp_assert_valid_gtid(global_tid);
524 if (!TCR_4(__kmp_init_parallel))
525 __kmp_parallel_initialize();
527 __kmp_resume_if_soft_paused();
529 this_thr = __kmp_threads[global_tid];
530 serial_team = this_thr->th.th_serial_team;
532 kmp_task_team_t *task_team = this_thr->th.th_task_team;
534 if (task_team != NULL && (task_team->tt.tt_found_proxy_tasks ||
535 task_team->tt.tt_hidden_helper_task_encountered))
536 __kmp_task_team_wait(this_thr, serial_team USE_ITT_BUILD_ARG(NULL));
539 KMP_DEBUG_ASSERT(serial_team);
540 KMP_ASSERT(serial_team->t.t_serialized);
541 KMP_DEBUG_ASSERT(this_thr->th.th_team == serial_team);
542 KMP_DEBUG_ASSERT(serial_team != this_thr->th.th_root->r.r_root_team);
543 KMP_DEBUG_ASSERT(serial_team->t.t_threads);
544 KMP_DEBUG_ASSERT(serial_team->t.t_threads[0] == this_thr);
547 if (ompt_enabled.enabled &&
548 this_thr->th.ompt_thread_info.state != ompt_state_overhead) {
549 OMPT_CUR_TASK_INFO(this_thr)->frame.exit_frame = ompt_data_none;
550 if (ompt_enabled.ompt_callback_implicit_task) {
551 ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
552 ompt_scope_end, NULL, OMPT_CUR_TASK_DATA(this_thr), 1,
553 OMPT_CUR_TASK_INFO(this_thr)->thread_num, ompt_task_implicit);
557 ompt_data_t *parent_task_data;
558 __ompt_get_task_info_internal(1, NULL, &parent_task_data, NULL, NULL, NULL);
560 if (ompt_enabled.ompt_callback_parallel_end) {
561 ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
562 &(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
563 ompt_parallel_invoker_program | ompt_parallel_team,
564 OMPT_LOAD_RETURN_ADDRESS(global_tid));
566 __ompt_lw_taskteam_unlink(this_thr);
567 this_thr->th.ompt_thread_info.state = ompt_state_overhead;
573 top = serial_team->t.t_control_stack_top;
574 if (top && top->serial_nesting_level == serial_team->t.t_serialized) {
575 copy_icvs(&serial_team->t.t_threads[0]->th.th_current_task->td_icvs, top);
576 serial_team->t.t_control_stack_top = top->next;
581 KMP_DEBUG_ASSERT(serial_team->t.t_dispatch->th_disp_buffer);
583 dispatch_private_info_t *disp_buffer =
584 serial_team->t.t_dispatch->th_disp_buffer;
585 serial_team->t.t_dispatch->th_disp_buffer =
586 serial_team->t.t_dispatch->th_disp_buffer->next;
587 __kmp_free(disp_buffer);
589 this_thr->th.th_def_allocator = serial_team->t.t_def_allocator;
591 --serial_team->t.t_serialized;
592 if (serial_team->t.t_serialized == 0) {
596#if KMP_ARCH_X86 || KMP_ARCH_X86_64
597 if (__kmp_inherit_fp_control && serial_team->t.t_fp_control_saved) {
598 __kmp_clear_x87_fpu_status_word();
599 __kmp_load_x87_fpu_control_word(&serial_team->t.t_x87_fpu_control_word);
600 __kmp_load_mxcsr(&serial_team->t.t_mxcsr);
604 __kmp_pop_current_task_from_thread(this_thr);
606 if (ompd_state & OMPD_ENABLE_BP)
607 ompd_bp_parallel_end();
610 this_thr->th.th_team = serial_team->t.t_parent;
611 this_thr->th.th_info.ds.ds_tid = serial_team->t.t_master_tid;
614 this_thr->th.th_team_nproc = serial_team->t.t_parent->t.t_nproc;
615 this_thr->th.th_team_master =
616 serial_team->t.t_parent->t.t_threads[0];
617 this_thr->th.th_team_serialized = this_thr->th.th_team->t.t_serialized;
620 this_thr->th.th_dispatch =
621 &this_thr->th.th_team->t.t_dispatch[serial_team->t.t_master_tid];
623 KMP_ASSERT(this_thr->th.th_current_task->td_flags.executing == 0);
624 this_thr->th.th_current_task->td_flags.executing = 1;
626 if (__kmp_tasking_mode != tskm_immediate_exec) {
628 this_thr->th.th_task_team =
629 this_thr->th.th_team->t.t_task_team[this_thr->th.th_task_state];
631 (
"__kmpc_end_serialized_parallel: T#%d restoring task_team %p / "
633 global_tid, this_thr->th.th_task_team, this_thr->th.th_team));
635#if KMP_AFFINITY_SUPPORTED
636 if (this_thr->th.th_team->t.t_level == 0 && __kmp_affinity.flags.reset) {
637 __kmp_reset_root_init_mask(global_tid);
641 if (__kmp_tasking_mode != tskm_immediate_exec) {
642 KA_TRACE(20, (
"__kmpc_end_serialized_parallel: T#%d decreasing nesting "
643 "depth of serial team %p to %d\n",
644 global_tid, serial_team, serial_team->t.t_serialized));
648 serial_team->t.t_level--;
649 if (__kmp_env_consistency_check)
650 __kmp_pop_parallel(global_tid, NULL);
652 if (ompt_enabled.enabled)
653 this_thr->th.ompt_thread_info.state =
654 ((this_thr->th.th_team_serialized) ? ompt_state_work_serial
655 : ompt_state_work_parallel);
668 KC_TRACE(10, (
"__kmpc_flush: called\n"));
673#if OMPT_SUPPORT && OMPT_OPTIONAL
674 if (ompt_enabled.ompt_callback_flush) {
675 ompt_callbacks.ompt_callback(ompt_callback_flush)(
676 __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
691 KC_TRACE(10, (
"__kmpc_barrier: called T#%d\n", global_tid));
692 __kmp_assert_valid_gtid(global_tid);
694 if (!TCR_4(__kmp_init_parallel))
695 __kmp_parallel_initialize();
697 __kmp_resume_if_soft_paused();
699 if (__kmp_env_consistency_check) {
701 KMP_WARNING(ConstructIdentInvalid);
703 __kmp_check_barrier(global_tid, ct_barrier, loc);
707 ompt_frame_t *ompt_frame;
708 if (ompt_enabled.enabled) {
709 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
710 if (ompt_frame->enter_frame.ptr == NULL)
711 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
713 OMPT_STORE_RETURN_ADDRESS(global_tid);
715 __kmp_threads[global_tid]->th.th_ident = loc;
723 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
724#if OMPT_SUPPORT && OMPT_OPTIONAL
725 if (ompt_enabled.enabled) {
726 ompt_frame->enter_frame = ompt_data_none;
741 KC_TRACE(10, (
"__kmpc_master: called T#%d\n", global_tid));
742 __kmp_assert_valid_gtid(global_tid);
744 if (!TCR_4(__kmp_init_parallel))
745 __kmp_parallel_initialize();
747 __kmp_resume_if_soft_paused();
749 if (KMP_MASTER_GTID(global_tid)) {
751 KMP_PUSH_PARTITIONED_TIMER(OMP_master);
755#if OMPT_SUPPORT && OMPT_OPTIONAL
757 if (ompt_enabled.ompt_callback_masked) {
758 kmp_info_t *this_thr = __kmp_threads[global_tid];
759 kmp_team_t *team = this_thr->th.th_team;
761 int tid = __kmp_tid_from_gtid(global_tid);
762 ompt_callbacks.ompt_callback(ompt_callback_masked)(
763 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
764 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
765 OMPT_GET_RETURN_ADDRESS(0));
770 if (__kmp_env_consistency_check) {
771#if KMP_USE_DYNAMIC_LOCK
773 __kmp_push_sync(global_tid, ct_master, loc, NULL, 0);
775 __kmp_check_sync(global_tid, ct_master, loc, NULL, 0);
778 __kmp_push_sync(global_tid, ct_master, loc, NULL);
780 __kmp_check_sync(global_tid, ct_master, loc, NULL);
796 KC_TRACE(10, (
"__kmpc_end_master: called T#%d\n", global_tid));
797 __kmp_assert_valid_gtid(global_tid);
798 KMP_DEBUG_ASSERT(KMP_MASTER_GTID(global_tid));
799 KMP_POP_PARTITIONED_TIMER();
801#if OMPT_SUPPORT && OMPT_OPTIONAL
802 kmp_info_t *this_thr = __kmp_threads[global_tid];
803 kmp_team_t *team = this_thr->th.th_team;
804 if (ompt_enabled.ompt_callback_masked) {
805 int tid = __kmp_tid_from_gtid(global_tid);
806 ompt_callbacks.ompt_callback(ompt_callback_masked)(
807 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
808 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
809 OMPT_GET_RETURN_ADDRESS(0));
813 if (__kmp_env_consistency_check) {
814 if (KMP_MASTER_GTID(global_tid))
815 __kmp_pop_sync(global_tid, ct_master, loc);
830 KC_TRACE(10, (
"__kmpc_masked: called T#%d\n", global_tid));
831 __kmp_assert_valid_gtid(global_tid);
833 if (!TCR_4(__kmp_init_parallel))
834 __kmp_parallel_initialize();
836 __kmp_resume_if_soft_paused();
838 tid = __kmp_tid_from_gtid(global_tid);
841 KMP_PUSH_PARTITIONED_TIMER(OMP_masked);
845#if OMPT_SUPPORT && OMPT_OPTIONAL
847 if (ompt_enabled.ompt_callback_masked) {
848 kmp_info_t *this_thr = __kmp_threads[global_tid];
849 kmp_team_t *team = this_thr->th.th_team;
850 ompt_callbacks.ompt_callback(ompt_callback_masked)(
851 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
852 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
853 OMPT_GET_RETURN_ADDRESS(0));
858 if (__kmp_env_consistency_check) {
859#if KMP_USE_DYNAMIC_LOCK
861 __kmp_push_sync(global_tid, ct_masked, loc, NULL, 0);
863 __kmp_check_sync(global_tid, ct_masked, loc, NULL, 0);
866 __kmp_push_sync(global_tid, ct_masked, loc, NULL);
868 __kmp_check_sync(global_tid, ct_masked, loc, NULL);
884 KC_TRACE(10, (
"__kmpc_end_masked: called T#%d\n", global_tid));
885 __kmp_assert_valid_gtid(global_tid);
886 KMP_POP_PARTITIONED_TIMER();
888#if OMPT_SUPPORT && OMPT_OPTIONAL
889 kmp_info_t *this_thr = __kmp_threads[global_tid];
890 kmp_team_t *team = this_thr->th.th_team;
891 if (ompt_enabled.ompt_callback_masked) {
892 int tid = __kmp_tid_from_gtid(global_tid);
893 ompt_callbacks.ompt_callback(ompt_callback_masked)(
894 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
895 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
896 OMPT_GET_RETURN_ADDRESS(0));
900 if (__kmp_env_consistency_check) {
901 __kmp_pop_sync(global_tid, ct_masked, loc);
915 KMP_DEBUG_ASSERT(__kmp_init_serial);
917 KC_TRACE(10, (
"__kmpc_ordered: called T#%d\n", gtid));
918 __kmp_assert_valid_gtid(gtid);
920 if (!TCR_4(__kmp_init_parallel))
921 __kmp_parallel_initialize();
923 __kmp_resume_if_soft_paused();
926 __kmp_itt_ordered_prep(gtid);
930 th = __kmp_threads[gtid];
932#if OMPT_SUPPORT && OMPT_OPTIONAL
936 OMPT_STORE_RETURN_ADDRESS(gtid);
937 if (ompt_enabled.enabled) {
938 team = __kmp_team_from_gtid(gtid);
939 lck = (ompt_wait_id_t)(uintptr_t)&team->t.t_ordered.dt.t_value;
941 th->th.ompt_thread_info.wait_id = lck;
942 th->th.ompt_thread_info.state = ompt_state_wait_ordered;
945 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
946 if (ompt_enabled.ompt_callback_mutex_acquire) {
947 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
948 ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin, lck,
954 if (th->th.th_dispatch->th_deo_fcn != 0)
955 (*th->th.th_dispatch->th_deo_fcn)(>id, &cid, loc);
957 __kmp_parallel_deo(>id, &cid, loc);
959#if OMPT_SUPPORT && OMPT_OPTIONAL
960 if (ompt_enabled.enabled) {
962 th->th.ompt_thread_info.state = ompt_state_work_parallel;
963 th->th.ompt_thread_info.wait_id = 0;
966 if (ompt_enabled.ompt_callback_mutex_acquired) {
967 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
968 ompt_mutex_ordered, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
974 __kmp_itt_ordered_start(gtid);
989 KC_TRACE(10, (
"__kmpc_end_ordered: called T#%d\n", gtid));
990 __kmp_assert_valid_gtid(gtid);
993 __kmp_itt_ordered_end(gtid);
997 th = __kmp_threads[gtid];
999 if (th->th.th_dispatch->th_dxo_fcn != 0)
1000 (*th->th.th_dispatch->th_dxo_fcn)(>id, &cid, loc);
1002 __kmp_parallel_dxo(>id, &cid, loc);
1004#if OMPT_SUPPORT && OMPT_OPTIONAL
1005 OMPT_STORE_RETURN_ADDRESS(gtid);
1006 if (ompt_enabled.ompt_callback_mutex_released) {
1007 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1009 (ompt_wait_id_t)(uintptr_t)&__kmp_team_from_gtid(gtid)
1010 ->t.t_ordered.dt.t_value,
1011 OMPT_LOAD_RETURN_ADDRESS(gtid));
1016#if KMP_USE_DYNAMIC_LOCK
1018static __forceinline
void
1019__kmp_init_indirect_csptr(kmp_critical_name *crit,
ident_t const *loc,
1020 kmp_int32 gtid, kmp_indirect_locktag_t tag) {
1024 kmp_indirect_lock_t **lck;
1025 lck = (kmp_indirect_lock_t **)crit;
1026 kmp_indirect_lock_t *ilk = __kmp_allocate_indirect_lock(&idx, gtid, tag);
1027 KMP_I_LOCK_FUNC(ilk, init)(ilk->lock);
1028 KMP_SET_I_LOCK_LOCATION(ilk, loc);
1029 KMP_SET_I_LOCK_FLAGS(ilk, kmp_lf_critical_section);
1031 (
"__kmp_init_indirect_csptr: initialized indirect lock #%d\n", tag));
1033 __kmp_itt_critical_creating(ilk->lock, loc);
1035 int status = KMP_COMPARE_AND_STORE_PTR(lck,
nullptr, ilk);
1038 __kmp_itt_critical_destroyed(ilk->lock);
1044 KMP_DEBUG_ASSERT(*lck != NULL);
1048#define KMP_ACQUIRE_TAS_LOCK(lock, gtid) \
1050 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1051 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1052 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1053 if (KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1054 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \
1056 KMP_FSYNC_PREPARE(l); \
1057 KMP_INIT_YIELD(spins); \
1058 kmp_backoff_t backoff = __kmp_spin_backoff_params; \
1060 if (TCR_4(__kmp_nth) > \
1061 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
1064 KMP_YIELD_SPIN(spins); \
1066 __kmp_spin_backoff(&backoff); \
1068 KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1069 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)); \
1071 KMP_FSYNC_ACQUIRED(l); \
1075#define KMP_TEST_TAS_LOCK(lock, gtid, rc) \
1077 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1078 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1079 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1080 rc = KMP_ATOMIC_LD_RLX(&l->lk.poll) == tas_free && \
1081 __kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy); \
1085#define KMP_RELEASE_TAS_LOCK(lock, gtid) \
1086 { KMP_ATOMIC_ST_REL(&((kmp_tas_lock_t *)lock)->lk.poll, KMP_LOCK_FREE(tas)); }
1090#include <sys/syscall.h>
1100#define KMP_ACQUIRE_FUTEX_LOCK(lock, gtid) \
1102 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1103 kmp_int32 gtid_code = (gtid + 1) << 1; \
1105 KMP_FSYNC_PREPARE(ftx); \
1106 kmp_int32 poll_val; \
1107 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( \
1108 &(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1109 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { \
1110 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; \
1112 if (!KMP_COMPARE_AND_STORE_RET32(&(ftx->lk.poll), poll_val, \
1114 KMP_LOCK_BUSY(1, futex))) { \
1117 poll_val |= KMP_LOCK_BUSY(1, futex); \
1120 if ((rc = syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAIT, poll_val, \
1121 NULL, NULL, 0)) != 0) { \
1126 KMP_FSYNC_ACQUIRED(ftx); \
1130#define KMP_TEST_FUTEX_LOCK(lock, gtid, rc) \
1132 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1133 if (KMP_COMPARE_AND_STORE_ACQ32(&(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1134 KMP_LOCK_BUSY(gtid + 1 << 1, futex))) { \
1135 KMP_FSYNC_ACQUIRED(ftx); \
1143#define KMP_RELEASE_FUTEX_LOCK(lock, gtid) \
1145 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1147 KMP_FSYNC_RELEASING(ftx); \
1148 kmp_int32 poll_val = \
1149 KMP_XCHG_FIXED32(&(ftx->lk.poll), KMP_LOCK_FREE(futex)); \
1150 if (KMP_LOCK_STRIP(poll_val) & 1) { \
1151 syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAKE, \
1152 KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \
1155 KMP_YIELD_OVERSUB(); \
1162static kmp_user_lock_p __kmp_get_critical_section_ptr(kmp_critical_name *crit,
1165 kmp_user_lock_p *lck_pp = (kmp_user_lock_p *)crit;
1168 kmp_user_lock_p lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1175 lck = __kmp_user_lock_allocate(&idx, gtid, kmp_lf_critical_section);
1176 __kmp_init_user_lock_with_checks(lck);
1177 __kmp_set_user_lock_location(lck, loc);
1179 __kmp_itt_critical_creating(lck);
1190 int status = KMP_COMPARE_AND_STORE_PTR(lck_pp, 0, lck);
1195 __kmp_itt_critical_destroyed(lck);
1199 __kmp_destroy_user_lock_with_checks(lck);
1200 __kmp_user_lock_free(&idx, gtid, lck);
1201 lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1202 KMP_DEBUG_ASSERT(lck != NULL);
1221 kmp_critical_name *crit) {
1222#if KMP_USE_DYNAMIC_LOCK
1223#if OMPT_SUPPORT && OMPT_OPTIONAL
1224 OMPT_STORE_RETURN_ADDRESS(global_tid);
1229#if OMPT_SUPPORT && OMPT_OPTIONAL
1230 ompt_state_t prev_state = ompt_state_undefined;
1231 ompt_thread_info_t ti;
1233 kmp_user_lock_p lck;
1235 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1236 __kmp_assert_valid_gtid(global_tid);
1240 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1241 KMP_CHECK_USER_LOCK_INIT();
1243 if ((__kmp_user_lock_kind == lk_tas) &&
1244 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1245 lck = (kmp_user_lock_p)crit;
1248 else if ((__kmp_user_lock_kind == lk_futex) &&
1249 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1250 lck = (kmp_user_lock_p)crit;
1254 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
1257 if (__kmp_env_consistency_check)
1258 __kmp_push_sync(global_tid, ct_critical, loc, lck);
1266 __kmp_itt_critical_acquiring(lck);
1268#if OMPT_SUPPORT && OMPT_OPTIONAL
1269 OMPT_STORE_RETURN_ADDRESS(gtid);
1270 void *codeptr_ra = NULL;
1271 if (ompt_enabled.enabled) {
1272 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1274 prev_state = ti.state;
1275 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1276 ti.state = ompt_state_wait_critical;
1279 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1280 if (ompt_enabled.ompt_callback_mutex_acquire) {
1281 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1282 ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
1283 (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1289 __kmp_acquire_user_lock_with_checks(lck, global_tid);
1292 __kmp_itt_critical_acquired(lck);
1294#if OMPT_SUPPORT && OMPT_OPTIONAL
1295 if (ompt_enabled.enabled) {
1297 ti.state = prev_state;
1301 if (ompt_enabled.ompt_callback_mutex_acquired) {
1302 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1303 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1307 KMP_POP_PARTITIONED_TIMER();
1309 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1310 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1314#if KMP_USE_DYNAMIC_LOCK
1317static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
1319#define KMP_TSX_LOCK(seq) lockseq_##seq
1321#define KMP_TSX_LOCK(seq) __kmp_user_lock_seq
1324#if KMP_ARCH_X86 || KMP_ARCH_X86_64
1325#define KMP_CPUINFO_RTM (__kmp_cpuinfo.flags.rtm)
1327#define KMP_CPUINFO_RTM 0
1331 if (hint & kmp_lock_hint_hle)
1332 return KMP_TSX_LOCK(hle);
1333 if (hint & kmp_lock_hint_rtm)
1334 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_queuing) : __kmp_user_lock_seq;
1335 if (hint & kmp_lock_hint_adaptive)
1336 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
1339 if ((hint & omp_lock_hint_contended) && (hint & omp_lock_hint_uncontended))
1340 return __kmp_user_lock_seq;
1341 if ((hint & omp_lock_hint_speculative) &&
1342 (hint & omp_lock_hint_nonspeculative))
1343 return __kmp_user_lock_seq;
1346 if (hint & omp_lock_hint_contended)
1347 return lockseq_queuing;
1350 if ((hint & omp_lock_hint_uncontended) && !(hint & omp_lock_hint_speculative))
1354 if (hint & omp_lock_hint_speculative)
1355 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_spin) : __kmp_user_lock_seq;
1357 return __kmp_user_lock_seq;
1360#if OMPT_SUPPORT && OMPT_OPTIONAL
1361#if KMP_USE_DYNAMIC_LOCK
1362static kmp_mutex_impl_t
1363__ompt_get_mutex_impl_type(
void *user_lock, kmp_indirect_lock_t *ilock = 0) {
1365 switch (KMP_EXTRACT_D_TAG(user_lock)) {
1370 return kmp_mutex_impl_queuing;
1373 return kmp_mutex_impl_spin;
1376 case locktag_rtm_spin:
1377 return kmp_mutex_impl_speculative;
1380 return kmp_mutex_impl_none;
1382 ilock = KMP_LOOKUP_I_LOCK(user_lock);
1385 switch (ilock->type) {
1387 case locktag_adaptive:
1388 case locktag_rtm_queuing:
1389 return kmp_mutex_impl_speculative;
1391 case locktag_nested_tas:
1392 return kmp_mutex_impl_spin;
1394 case locktag_nested_futex:
1396 case locktag_ticket:
1397 case locktag_queuing:
1399 case locktag_nested_ticket:
1400 case locktag_nested_queuing:
1401 case locktag_nested_drdpa:
1402 return kmp_mutex_impl_queuing;
1404 return kmp_mutex_impl_none;
1409static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
1410 switch (__kmp_user_lock_kind) {
1412 return kmp_mutex_impl_spin;
1419 return kmp_mutex_impl_queuing;
1422 case lk_rtm_queuing:
1425 return kmp_mutex_impl_speculative;
1428 return kmp_mutex_impl_none;
1448 kmp_critical_name *crit, uint32_t hint) {
1450 kmp_user_lock_p lck;
1451#if OMPT_SUPPORT && OMPT_OPTIONAL
1452 ompt_state_t prev_state = ompt_state_undefined;
1453 ompt_thread_info_t ti;
1455 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(global_tid);
1457 codeptr = OMPT_GET_RETURN_ADDRESS(0);
1460 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1461 __kmp_assert_valid_gtid(global_tid);
1463 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
1465 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1466 kmp_dyna_lockseq_t lockseq = __kmp_map_hint_to_lock(hint);
1468 if (KMP_IS_D_LOCK(lockseq)) {
1469 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
1470 KMP_GET_D_TAG(lockseq));
1472 __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lockseq));
1478 if (KMP_EXTRACT_D_TAG(lk) != 0) {
1479 lck = (kmp_user_lock_p)lk;
1480 if (__kmp_env_consistency_check) {
1481 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1482 __kmp_map_hint_to_lock(hint));
1485 __kmp_itt_critical_acquiring(lck);
1487#if OMPT_SUPPORT && OMPT_OPTIONAL
1488 if (ompt_enabled.enabled) {
1489 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1491 prev_state = ti.state;
1492 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1493 ti.state = ompt_state_wait_critical;
1496 if (ompt_enabled.ompt_callback_mutex_acquire) {
1497 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1498 ompt_mutex_critical, (
unsigned int)hint,
1499 __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)(uintptr_t)lck,
1504#if KMP_USE_INLINED_TAS
1505 if (lockseq == lockseq_tas && !__kmp_env_consistency_check) {
1506 KMP_ACQUIRE_TAS_LOCK(lck, global_tid);
1508#elif KMP_USE_INLINED_FUTEX
1509 if (lockseq == lockseq_futex && !__kmp_env_consistency_check) {
1510 KMP_ACQUIRE_FUTEX_LOCK(lck, global_tid);
1514 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
1517 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
1519 if (__kmp_env_consistency_check) {
1520 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1521 __kmp_map_hint_to_lock(hint));
1524 __kmp_itt_critical_acquiring(lck);
1526#if OMPT_SUPPORT && OMPT_OPTIONAL
1527 if (ompt_enabled.enabled) {
1528 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1530 prev_state = ti.state;
1531 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1532 ti.state = ompt_state_wait_critical;
1535 if (ompt_enabled.ompt_callback_mutex_acquire) {
1536 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1537 ompt_mutex_critical, (
unsigned int)hint,
1538 __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)(uintptr_t)lck,
1543 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
1545 KMP_POP_PARTITIONED_TIMER();
1548 __kmp_itt_critical_acquired(lck);
1550#if OMPT_SUPPORT && OMPT_OPTIONAL
1551 if (ompt_enabled.enabled) {
1553 ti.state = prev_state;
1557 if (ompt_enabled.ompt_callback_mutex_acquired) {
1558 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1559 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
1564 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1565 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1580 kmp_critical_name *crit) {
1581 kmp_user_lock_p lck;
1583 KC_TRACE(10, (
"__kmpc_end_critical: called T#%d\n", global_tid));
1585#if KMP_USE_DYNAMIC_LOCK
1586 int locktag = KMP_EXTRACT_D_TAG(crit);
1588 lck = (kmp_user_lock_p)crit;
1589 KMP_ASSERT(lck != NULL);
1590 if (__kmp_env_consistency_check) {
1591 __kmp_pop_sync(global_tid, ct_critical, loc);
1594 __kmp_itt_critical_releasing(lck);
1596#if KMP_USE_INLINED_TAS
1597 if (locktag == locktag_tas && !__kmp_env_consistency_check) {
1598 KMP_RELEASE_TAS_LOCK(lck, global_tid);
1600#elif KMP_USE_INLINED_FUTEX
1601 if (locktag == locktag_futex && !__kmp_env_consistency_check) {
1602 KMP_RELEASE_FUTEX_LOCK(lck, global_tid);
1606 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
1609 kmp_indirect_lock_t *ilk =
1610 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
1611 KMP_ASSERT(ilk != NULL);
1613 if (__kmp_env_consistency_check) {
1614 __kmp_pop_sync(global_tid, ct_critical, loc);
1617 __kmp_itt_critical_releasing(lck);
1619 KMP_I_LOCK_FUNC(ilk, unset)(lck, global_tid);
1624 if ((__kmp_user_lock_kind == lk_tas) &&
1625 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1626 lck = (kmp_user_lock_p)crit;
1629 else if ((__kmp_user_lock_kind == lk_futex) &&
1630 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1631 lck = (kmp_user_lock_p)crit;
1635 lck = (kmp_user_lock_p)TCR_PTR(*((kmp_user_lock_p *)crit));
1638 KMP_ASSERT(lck != NULL);
1640 if (__kmp_env_consistency_check)
1641 __kmp_pop_sync(global_tid, ct_critical, loc);
1644 __kmp_itt_critical_releasing(lck);
1648 __kmp_release_user_lock_with_checks(lck, global_tid);
1652#if OMPT_SUPPORT && OMPT_OPTIONAL
1655 OMPT_STORE_RETURN_ADDRESS(global_tid);
1656 if (ompt_enabled.ompt_callback_mutex_released) {
1657 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1658 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck,
1659 OMPT_LOAD_RETURN_ADDRESS(0));
1663 KMP_POP_PARTITIONED_TIMER();
1664 KA_TRACE(15, (
"__kmpc_end_critical: done T#%d\n", global_tid));
1678 KC_TRACE(10, (
"__kmpc_barrier_master: called T#%d\n", global_tid));
1679 __kmp_assert_valid_gtid(global_tid);
1681 if (!TCR_4(__kmp_init_parallel))
1682 __kmp_parallel_initialize();
1684 __kmp_resume_if_soft_paused();
1686 if (__kmp_env_consistency_check)
1687 __kmp_check_barrier(global_tid, ct_barrier, loc);
1690 ompt_frame_t *ompt_frame;
1691 if (ompt_enabled.enabled) {
1692 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1693 if (ompt_frame->enter_frame.ptr == NULL)
1694 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1696 OMPT_STORE_RETURN_ADDRESS(global_tid);
1699 __kmp_threads[global_tid]->th.th_ident = loc;
1701 status = __kmp_barrier(bs_plain_barrier, global_tid, TRUE, 0, NULL, NULL);
1702#if OMPT_SUPPORT && OMPT_OPTIONAL
1703 if (ompt_enabled.enabled) {
1704 ompt_frame->enter_frame = ompt_data_none;
1708 return (status != 0) ? 0 : 1;
1721 KC_TRACE(10, (
"__kmpc_end_barrier_master: called T#%d\n", global_tid));
1722 __kmp_assert_valid_gtid(global_tid);
1723 __kmp_end_split_barrier(bs_plain_barrier, global_tid);
1738 KC_TRACE(10, (
"__kmpc_barrier_master_nowait: called T#%d\n", global_tid));
1739 __kmp_assert_valid_gtid(global_tid);
1741 if (!TCR_4(__kmp_init_parallel))
1742 __kmp_parallel_initialize();
1744 __kmp_resume_if_soft_paused();
1746 if (__kmp_env_consistency_check) {
1748 KMP_WARNING(ConstructIdentInvalid);
1750 __kmp_check_barrier(global_tid, ct_barrier, loc);
1754 ompt_frame_t *ompt_frame;
1755 if (ompt_enabled.enabled) {
1756 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1757 if (ompt_frame->enter_frame.ptr == NULL)
1758 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1760 OMPT_STORE_RETURN_ADDRESS(global_tid);
1763 __kmp_threads[global_tid]->th.th_ident = loc;
1765 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
1766#if OMPT_SUPPORT && OMPT_OPTIONAL
1767 if (ompt_enabled.enabled) {
1768 ompt_frame->enter_frame = ompt_data_none;
1774 if (__kmp_env_consistency_check) {
1780 __kmp_pop_sync(global_tid, ct_master, loc);
1800 __kmp_assert_valid_gtid(global_tid);
1801 kmp_int32 rc = __kmp_enter_single(global_tid, loc, TRUE);
1806 KMP_PUSH_PARTITIONED_TIMER(OMP_single);
1809#if OMPT_SUPPORT && OMPT_OPTIONAL
1810 kmp_info_t *this_thr = __kmp_threads[global_tid];
1811 kmp_team_t *team = this_thr->th.th_team;
1812 int tid = __kmp_tid_from_gtid(global_tid);
1814 if (ompt_enabled.enabled) {
1816 if (ompt_enabled.ompt_callback_work) {
1817 ompt_callbacks.ompt_callback(ompt_callback_work)(
1818 ompt_work_single_executor, ompt_scope_begin,
1819 &(team->t.ompt_team_info.parallel_data),
1820 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1821 1, OMPT_GET_RETURN_ADDRESS(0));
1824 if (ompt_enabled.ompt_callback_work) {
1825 ompt_callbacks.ompt_callback(ompt_callback_work)(
1826 ompt_work_single_other, ompt_scope_begin,
1827 &(team->t.ompt_team_info.parallel_data),
1828 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1829 1, OMPT_GET_RETURN_ADDRESS(0));
1830 ompt_callbacks.ompt_callback(ompt_callback_work)(
1831 ompt_work_single_other, ompt_scope_end,
1832 &(team->t.ompt_team_info.parallel_data),
1833 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1834 1, OMPT_GET_RETURN_ADDRESS(0));
1853 __kmp_assert_valid_gtid(global_tid);
1854 __kmp_exit_single(global_tid);
1855 KMP_POP_PARTITIONED_TIMER();
1857#if OMPT_SUPPORT && OMPT_OPTIONAL
1858 kmp_info_t *this_thr = __kmp_threads[global_tid];
1859 kmp_team_t *team = this_thr->th.th_team;
1860 int tid = __kmp_tid_from_gtid(global_tid);
1862 if (ompt_enabled.ompt_callback_work) {
1863 ompt_callbacks.ompt_callback(ompt_callback_work)(
1864 ompt_work_single_executor, ompt_scope_end,
1865 &(team->t.ompt_team_info.parallel_data),
1866 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
1867 OMPT_GET_RETURN_ADDRESS(0));
1880 KMP_POP_PARTITIONED_TIMER();
1881 KE_TRACE(10, (
"__kmpc_for_static_fini called T#%d\n", global_tid));
1883#if OMPT_SUPPORT && OMPT_OPTIONAL
1884 if (ompt_enabled.ompt_callback_work) {
1885 ompt_work_t ompt_work_type = ompt_work_loop;
1886 ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
1887 ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
1891 ompt_work_type = ompt_work_loop;
1893 ompt_work_type = ompt_work_sections;
1895 ompt_work_type = ompt_work_distribute;
1900 KMP_DEBUG_ASSERT(ompt_work_type);
1902 ompt_callbacks.ompt_callback(ompt_callback_work)(
1903 ompt_work_type, ompt_scope_end, &(team_info->parallel_data),
1904 &(task_info->task_data), 0, OMPT_GET_RETURN_ADDRESS(0));
1907 if (__kmp_env_consistency_check)
1908 __kmp_pop_workshare(global_tid, ct_pdo, loc);
1914void ompc_set_num_threads(
int arg) {
1916 __kmp_set_num_threads(arg, __kmp_entry_gtid());
1919void ompc_set_dynamic(
int flag) {
1923 thread = __kmp_entry_thread();
1925 __kmp_save_internal_controls(thread);
1927 set__dynamic(thread, flag ?
true :
false);
1930void ompc_set_nested(
int flag) {
1934 thread = __kmp_entry_thread();
1936 __kmp_save_internal_controls(thread);
1938 set__max_active_levels(thread, flag ? __kmp_dflt_max_active_levels : 1);
1941void ompc_set_max_active_levels(
int max_active_levels) {
1946 __kmp_set_max_active_levels(__kmp_entry_gtid(), max_active_levels);
1949void ompc_set_schedule(omp_sched_t kind,
int modifier) {
1951 __kmp_set_schedule(__kmp_entry_gtid(), (kmp_sched_t)kind, modifier);
1954int ompc_get_ancestor_thread_num(
int level) {
1955 return __kmp_get_ancestor_thread_num(__kmp_entry_gtid(), level);
1958int ompc_get_team_size(
int level) {
1959 return __kmp_get_team_size(__kmp_entry_gtid(), level);
1963void KMP_EXPAND_NAME(ompc_set_affinity_format)(
char const *format) {
1964 if (!__kmp_init_serial) {
1965 __kmp_serial_initialize();
1967 __kmp_strncpy_truncate(__kmp_affinity_format, KMP_AFFINITY_FORMAT_SIZE,
1968 format, KMP_STRLEN(format) + 1);
1971size_t KMP_EXPAND_NAME(ompc_get_affinity_format)(
char *buffer,
size_t size) {
1973 if (!__kmp_init_serial) {
1974 __kmp_serial_initialize();
1976 format_size = KMP_STRLEN(__kmp_affinity_format);
1977 if (buffer && size) {
1978 __kmp_strncpy_truncate(buffer, size, __kmp_affinity_format,
1984void KMP_EXPAND_NAME(ompc_display_affinity)(
char const *format) {
1986 if (!TCR_4(__kmp_init_middle)) {
1987 __kmp_middle_initialize();
1989 __kmp_assign_root_init_mask();
1990 gtid = __kmp_get_gtid();
1991#if KMP_AFFINITY_SUPPORTED
1992 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 &&
1993 __kmp_affinity.flags.reset) {
1994 __kmp_reset_root_init_mask(gtid);
1997 __kmp_aux_display_affinity(gtid, format);
2000size_t KMP_EXPAND_NAME(ompc_capture_affinity)(
char *buffer,
size_t buf_size,
2001 char const *format) {
2003 size_t num_required;
2004 kmp_str_buf_t capture_buf;
2005 if (!TCR_4(__kmp_init_middle)) {
2006 __kmp_middle_initialize();
2008 __kmp_assign_root_init_mask();
2009 gtid = __kmp_get_gtid();
2010#if KMP_AFFINITY_SUPPORTED
2011 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 &&
2012 __kmp_affinity.flags.reset) {
2013 __kmp_reset_root_init_mask(gtid);
2016 __kmp_str_buf_init(&capture_buf);
2017 num_required = __kmp_aux_capture_affinity(gtid, format, &capture_buf);
2018 if (buffer && buf_size) {
2019 __kmp_strncpy_truncate(buffer, buf_size, capture_buf.str,
2020 capture_buf.used + 1);
2022 __kmp_str_buf_free(&capture_buf);
2023 return num_required;
2026void kmpc_set_stacksize(
int arg) {
2028 __kmp_aux_set_stacksize(arg);
2031void kmpc_set_stacksize_s(
size_t arg) {
2033 __kmp_aux_set_stacksize(arg);
2036void kmpc_set_blocktime(
int arg) {
2040 gtid = __kmp_entry_gtid();
2041 tid = __kmp_tid_from_gtid(gtid);
2042 thread = __kmp_thread_from_gtid(gtid);
2044 __kmp_aux_set_blocktime(arg, thread, tid);
2047void kmpc_set_library(
int arg) {
2049 __kmp_user_set_library((
enum library_type)arg);
2052void kmpc_set_defaults(
char const *str) {
2054 __kmp_aux_set_defaults(str, KMP_STRLEN(str));
2057void kmpc_set_disp_num_buffers(
int arg) {
2060 if (__kmp_init_serial == FALSE && arg >= KMP_MIN_DISP_NUM_BUFF &&
2061 arg <= KMP_MAX_DISP_NUM_BUFF) {
2062 __kmp_dispatch_num_buffers = arg;
2066int kmpc_set_affinity_mask_proc(
int proc,
void **mask) {
2067#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2070 if (!TCR_4(__kmp_init_middle)) {
2071 __kmp_middle_initialize();
2073 __kmp_assign_root_init_mask();
2074 return __kmp_aux_set_affinity_mask_proc(proc, mask);
2078int kmpc_unset_affinity_mask_proc(
int proc,
void **mask) {
2079#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2082 if (!TCR_4(__kmp_init_middle)) {
2083 __kmp_middle_initialize();
2085 __kmp_assign_root_init_mask();
2086 return __kmp_aux_unset_affinity_mask_proc(proc, mask);
2090int kmpc_get_affinity_mask_proc(
int proc,
void **mask) {
2091#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2094 if (!TCR_4(__kmp_init_middle)) {
2095 __kmp_middle_initialize();
2097 __kmp_assign_root_init_mask();
2098 return __kmp_aux_get_affinity_mask_proc(proc, mask);
2148 void *cpy_data,
void (*cpy_func)(
void *,
void *),
2151 KC_TRACE(10, (
"__kmpc_copyprivate: called T#%d\n", gtid));
2152 __kmp_assert_valid_gtid(gtid);
2156 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2158 if (__kmp_env_consistency_check) {
2160 KMP_WARNING(ConstructIdentInvalid);
2167 *data_ptr = cpy_data;
2170 ompt_frame_t *ompt_frame;
2171 if (ompt_enabled.enabled) {
2172 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2173 if (ompt_frame->enter_frame.ptr == NULL)
2174 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2176 OMPT_STORE_RETURN_ADDRESS(gtid);
2180 __kmp_threads[gtid]->th.th_ident = loc;
2182 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2185 (*cpy_func)(cpy_data, *data_ptr);
2191 OMPT_STORE_RETURN_ADDRESS(gtid);
2194 __kmp_threads[gtid]->th.th_ident = loc;
2197 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2198#if OMPT_SUPPORT && OMPT_OPTIONAL
2199 if (ompt_enabled.enabled) {
2200 ompt_frame->enter_frame = ompt_data_none;
2226 KC_TRACE(10, (
"__kmpc_copyprivate_light: called T#%d\n", gtid));
2230 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2232 if (__kmp_env_consistency_check) {
2234 KMP_WARNING(ConstructIdentInvalid);
2241 *data_ptr = cpy_data;
2244 ompt_frame_t *ompt_frame;
2245 if (ompt_enabled.enabled) {
2246 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2247 if (ompt_frame->enter_frame.ptr == NULL)
2248 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2249 OMPT_STORE_RETURN_ADDRESS(gtid);
2254 __kmp_threads[gtid]->th.th_ident = loc;
2256 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2263#define INIT_LOCK __kmp_init_user_lock_with_checks
2264#define INIT_NESTED_LOCK __kmp_init_nested_user_lock_with_checks
2265#define ACQUIRE_LOCK __kmp_acquire_user_lock_with_checks
2266#define ACQUIRE_LOCK_TIMED __kmp_acquire_user_lock_with_checks_timed
2267#define ACQUIRE_NESTED_LOCK __kmp_acquire_nested_user_lock_with_checks
2268#define ACQUIRE_NESTED_LOCK_TIMED \
2269 __kmp_acquire_nested_user_lock_with_checks_timed
2270#define RELEASE_LOCK __kmp_release_user_lock_with_checks
2271#define RELEASE_NESTED_LOCK __kmp_release_nested_user_lock_with_checks
2272#define TEST_LOCK __kmp_test_user_lock_with_checks
2273#define TEST_NESTED_LOCK __kmp_test_nested_user_lock_with_checks
2274#define DESTROY_LOCK __kmp_destroy_user_lock_with_checks
2275#define DESTROY_NESTED_LOCK __kmp_destroy_nested_user_lock_with_checks
2280#if KMP_USE_DYNAMIC_LOCK
2283static __forceinline
void __kmp_init_lock_with_hint(
ident_t *loc,
void **lock,
2284 kmp_dyna_lockseq_t seq) {
2285 if (KMP_IS_D_LOCK(seq)) {
2286 KMP_INIT_D_LOCK(lock, seq);
2288 __kmp_itt_lock_creating((kmp_user_lock_p)lock, NULL);
2291 KMP_INIT_I_LOCK(lock, seq);
2293 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2294 __kmp_itt_lock_creating(ilk->lock, loc);
2300static __forceinline
void
2301__kmp_init_nest_lock_with_hint(
ident_t *loc,
void **lock,
2302 kmp_dyna_lockseq_t seq) {
2305 if (seq == lockseq_hle || seq == lockseq_rtm_queuing ||
2306 seq == lockseq_rtm_spin || seq == lockseq_adaptive)
2307 seq = __kmp_user_lock_seq;
2311 seq = lockseq_nested_tas;
2315 seq = lockseq_nested_futex;
2318 case lockseq_ticket:
2319 seq = lockseq_nested_ticket;
2321 case lockseq_queuing:
2322 seq = lockseq_nested_queuing;
2325 seq = lockseq_nested_drdpa;
2328 seq = lockseq_nested_queuing;
2330 KMP_INIT_I_LOCK(lock, seq);
2332 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2333 __kmp_itt_lock_creating(ilk->lock, loc);
2338void __kmpc_init_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
void **user_lock,
2340 KMP_DEBUG_ASSERT(__kmp_init_serial);
2341 if (__kmp_env_consistency_check && user_lock == NULL) {
2342 KMP_FATAL(LockIsUninitialized,
"omp_init_lock_with_hint");
2345 __kmp_init_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2347#if OMPT_SUPPORT && OMPT_OPTIONAL
2349 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2351 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2352 if (ompt_enabled.ompt_callback_lock_init) {
2353 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2354 ompt_mutex_lock, (omp_lock_hint_t)hint,
2355 __ompt_get_mutex_impl_type(user_lock),
2356 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2362void __kmpc_init_nest_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
2363 void **user_lock, uintptr_t hint) {
2364 KMP_DEBUG_ASSERT(__kmp_init_serial);
2365 if (__kmp_env_consistency_check && user_lock == NULL) {
2366 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock_with_hint");
2369 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2371#if OMPT_SUPPORT && OMPT_OPTIONAL
2373 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2375 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2376 if (ompt_enabled.ompt_callback_lock_init) {
2377 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2378 ompt_mutex_nest_lock, (omp_lock_hint_t)hint,
2379 __ompt_get_mutex_impl_type(user_lock),
2380 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2388void __kmpc_init_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2389#if KMP_USE_DYNAMIC_LOCK
2391 KMP_DEBUG_ASSERT(__kmp_init_serial);
2392 if (__kmp_env_consistency_check && user_lock == NULL) {
2393 KMP_FATAL(LockIsUninitialized,
"omp_init_lock");
2395 __kmp_init_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2397#if OMPT_SUPPORT && OMPT_OPTIONAL
2399 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2401 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2402 if (ompt_enabled.ompt_callback_lock_init) {
2403 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2404 ompt_mutex_lock, omp_lock_hint_none,
2405 __ompt_get_mutex_impl_type(user_lock),
2406 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2412 static char const *
const func =
"omp_init_lock";
2413 kmp_user_lock_p lck;
2414 KMP_DEBUG_ASSERT(__kmp_init_serial);
2416 if (__kmp_env_consistency_check) {
2417 if (user_lock == NULL) {
2418 KMP_FATAL(LockIsUninitialized, func);
2422 KMP_CHECK_USER_LOCK_INIT();
2424 if ((__kmp_user_lock_kind == lk_tas) &&
2425 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2426 lck = (kmp_user_lock_p)user_lock;
2429 else if ((__kmp_user_lock_kind == lk_futex) &&
2430 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2431 lck = (kmp_user_lock_p)user_lock;
2435 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2438 __kmp_set_user_lock_location(lck, loc);
2440#if OMPT_SUPPORT && OMPT_OPTIONAL
2442 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2444 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2445 if (ompt_enabled.ompt_callback_lock_init) {
2446 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2447 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2448 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2453 __kmp_itt_lock_creating(lck);
2460void __kmpc_init_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2461#if KMP_USE_DYNAMIC_LOCK
2463 KMP_DEBUG_ASSERT(__kmp_init_serial);
2464 if (__kmp_env_consistency_check && user_lock == NULL) {
2465 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock");
2467 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2469#if OMPT_SUPPORT && OMPT_OPTIONAL
2471 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2473 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2474 if (ompt_enabled.ompt_callback_lock_init) {
2475 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2476 ompt_mutex_nest_lock, omp_lock_hint_none,
2477 __ompt_get_mutex_impl_type(user_lock),
2478 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2484 static char const *
const func =
"omp_init_nest_lock";
2485 kmp_user_lock_p lck;
2486 KMP_DEBUG_ASSERT(__kmp_init_serial);
2488 if (__kmp_env_consistency_check) {
2489 if (user_lock == NULL) {
2490 KMP_FATAL(LockIsUninitialized, func);
2494 KMP_CHECK_USER_LOCK_INIT();
2496 if ((__kmp_user_lock_kind == lk_tas) &&
2497 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2498 OMP_NEST_LOCK_T_SIZE)) {
2499 lck = (kmp_user_lock_p)user_lock;
2502 else if ((__kmp_user_lock_kind == lk_futex) &&
2503 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2504 OMP_NEST_LOCK_T_SIZE)) {
2505 lck = (kmp_user_lock_p)user_lock;
2509 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2512 INIT_NESTED_LOCK(lck);
2513 __kmp_set_user_lock_location(lck, loc);
2515#if OMPT_SUPPORT && OMPT_OPTIONAL
2517 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2519 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2520 if (ompt_enabled.ompt_callback_lock_init) {
2521 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2522 ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2523 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2528 __kmp_itt_lock_creating(lck);
2534void __kmpc_destroy_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2535#if KMP_USE_DYNAMIC_LOCK
2538 kmp_user_lock_p lck;
2539 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2540 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2542 lck = (kmp_user_lock_p)user_lock;
2544 __kmp_itt_lock_destroyed(lck);
2546#if OMPT_SUPPORT && OMPT_OPTIONAL
2548 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2550 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2551 if (ompt_enabled.ompt_callback_lock_destroy) {
2552 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2553 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2556 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2558 kmp_user_lock_p lck;
2560 if ((__kmp_user_lock_kind == lk_tas) &&
2561 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2562 lck = (kmp_user_lock_p)user_lock;
2565 else if ((__kmp_user_lock_kind == lk_futex) &&
2566 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2567 lck = (kmp_user_lock_p)user_lock;
2571 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_lock");
2574#if OMPT_SUPPORT && OMPT_OPTIONAL
2576 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2578 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2579 if (ompt_enabled.ompt_callback_lock_destroy) {
2580 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2581 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2586 __kmp_itt_lock_destroyed(lck);
2590 if ((__kmp_user_lock_kind == lk_tas) &&
2591 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2595 else if ((__kmp_user_lock_kind == lk_futex) &&
2596 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2601 __kmp_user_lock_free(user_lock, gtid, lck);
2607void __kmpc_destroy_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2608#if KMP_USE_DYNAMIC_LOCK
2611 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(user_lock);
2612 __kmp_itt_lock_destroyed(ilk->lock);
2614#if OMPT_SUPPORT && OMPT_OPTIONAL
2616 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2618 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2619 if (ompt_enabled.ompt_callback_lock_destroy) {
2620 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2621 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2624 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2628 kmp_user_lock_p lck;
2630 if ((__kmp_user_lock_kind == lk_tas) &&
2631 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2632 OMP_NEST_LOCK_T_SIZE)) {
2633 lck = (kmp_user_lock_p)user_lock;
2636 else if ((__kmp_user_lock_kind == lk_futex) &&
2637 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2638 OMP_NEST_LOCK_T_SIZE)) {
2639 lck = (kmp_user_lock_p)user_lock;
2643 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_nest_lock");
2646#if OMPT_SUPPORT && OMPT_OPTIONAL
2648 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2650 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2651 if (ompt_enabled.ompt_callback_lock_destroy) {
2652 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2653 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2658 __kmp_itt_lock_destroyed(lck);
2661 DESTROY_NESTED_LOCK(lck);
2663 if ((__kmp_user_lock_kind == lk_tas) &&
2664 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2665 OMP_NEST_LOCK_T_SIZE)) {
2669 else if ((__kmp_user_lock_kind == lk_futex) &&
2670 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2671 OMP_NEST_LOCK_T_SIZE)) {
2676 __kmp_user_lock_free(user_lock, gtid, lck);
2681void __kmpc_set_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2683#if KMP_USE_DYNAMIC_LOCK
2684 int tag = KMP_EXTRACT_D_TAG(user_lock);
2686 __kmp_itt_lock_acquiring(
2690#if OMPT_SUPPORT && OMPT_OPTIONAL
2692 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2694 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2695 if (ompt_enabled.ompt_callback_mutex_acquire) {
2696 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2697 ompt_mutex_lock, omp_lock_hint_none,
2698 __ompt_get_mutex_impl_type(user_lock),
2699 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2702#if KMP_USE_INLINED_TAS
2703 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2704 KMP_ACQUIRE_TAS_LOCK(user_lock, gtid);
2706#elif KMP_USE_INLINED_FUTEX
2707 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2708 KMP_ACQUIRE_FUTEX_LOCK(user_lock, gtid);
2712 __kmp_direct_set[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2715 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2717#if OMPT_SUPPORT && OMPT_OPTIONAL
2718 if (ompt_enabled.ompt_callback_mutex_acquired) {
2719 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2720 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2726 kmp_user_lock_p lck;
2728 if ((__kmp_user_lock_kind == lk_tas) &&
2729 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2730 lck = (kmp_user_lock_p)user_lock;
2733 else if ((__kmp_user_lock_kind == lk_futex) &&
2734 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2735 lck = (kmp_user_lock_p)user_lock;
2739 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_lock");
2743 __kmp_itt_lock_acquiring(lck);
2745#if OMPT_SUPPORT && OMPT_OPTIONAL
2747 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2749 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2750 if (ompt_enabled.ompt_callback_mutex_acquire) {
2751 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2752 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2753 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2757 ACQUIRE_LOCK(lck, gtid);
2760 __kmp_itt_lock_acquired(lck);
2763#if OMPT_SUPPORT && OMPT_OPTIONAL
2764 if (ompt_enabled.ompt_callback_mutex_acquired) {
2765 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2766 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2773void __kmpc_set_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2774#if KMP_USE_DYNAMIC_LOCK
2777 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
2779#if OMPT_SUPPORT && OMPT_OPTIONAL
2781 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2783 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2784 if (ompt_enabled.enabled) {
2785 if (ompt_enabled.ompt_callback_mutex_acquire) {
2786 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2787 ompt_mutex_nest_lock, omp_lock_hint_none,
2788 __ompt_get_mutex_impl_type(user_lock),
2789 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2793 int acquire_status =
2794 KMP_D_LOCK_FUNC(user_lock, set)((kmp_dyna_lock_t *)user_lock, gtid);
2795 (void)acquire_status;
2797 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2800#if OMPT_SUPPORT && OMPT_OPTIONAL
2801 if (ompt_enabled.enabled) {
2802 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2803 if (ompt_enabled.ompt_callback_mutex_acquired) {
2805 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2806 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
2810 if (ompt_enabled.ompt_callback_nest_lock) {
2812 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2813 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2821 kmp_user_lock_p lck;
2823 if ((__kmp_user_lock_kind == lk_tas) &&
2824 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2825 OMP_NEST_LOCK_T_SIZE)) {
2826 lck = (kmp_user_lock_p)user_lock;
2829 else if ((__kmp_user_lock_kind == lk_futex) &&
2830 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2831 OMP_NEST_LOCK_T_SIZE)) {
2832 lck = (kmp_user_lock_p)user_lock;
2836 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_nest_lock");
2840 __kmp_itt_lock_acquiring(lck);
2842#if OMPT_SUPPORT && OMPT_OPTIONAL
2844 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2846 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2847 if (ompt_enabled.enabled) {
2848 if (ompt_enabled.ompt_callback_mutex_acquire) {
2849 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2850 ompt_mutex_nest_lock, omp_lock_hint_none,
2851 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
2857 ACQUIRE_NESTED_LOCK(lck, gtid, &acquire_status);
2860 __kmp_itt_lock_acquired(lck);
2863#if OMPT_SUPPORT && OMPT_OPTIONAL
2864 if (ompt_enabled.enabled) {
2865 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2866 if (ompt_enabled.ompt_callback_mutex_acquired) {
2868 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2869 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2872 if (ompt_enabled.ompt_callback_nest_lock) {
2874 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2875 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2884void __kmpc_unset_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2885#if KMP_USE_DYNAMIC_LOCK
2887 int tag = KMP_EXTRACT_D_TAG(user_lock);
2889 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2891#if KMP_USE_INLINED_TAS
2892 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2893 KMP_RELEASE_TAS_LOCK(user_lock, gtid);
2895#elif KMP_USE_INLINED_FUTEX
2896 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2897 KMP_RELEASE_FUTEX_LOCK(user_lock, gtid);
2901 __kmp_direct_unset[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2904#if OMPT_SUPPORT && OMPT_OPTIONAL
2906 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2908 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2909 if (ompt_enabled.ompt_callback_mutex_released) {
2910 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2911 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2917 kmp_user_lock_p lck;
2922 if ((__kmp_user_lock_kind == lk_tas) &&
2923 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2924#if KMP_OS_LINUX && \
2925 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
2928 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2930 TCW_4(((kmp_user_lock_p)user_lock)->tas.lk.poll, 0);
2933#if OMPT_SUPPORT && OMPT_OPTIONAL
2935 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2937 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2938 if (ompt_enabled.ompt_callback_mutex_released) {
2939 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2940 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2946 lck = (kmp_user_lock_p)user_lock;
2950 else if ((__kmp_user_lock_kind == lk_futex) &&
2951 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2952 lck = (kmp_user_lock_p)user_lock;
2956 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_lock");
2960 __kmp_itt_lock_releasing(lck);
2963 RELEASE_LOCK(lck, gtid);
2965#if OMPT_SUPPORT && OMPT_OPTIONAL
2967 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2969 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2970 if (ompt_enabled.ompt_callback_mutex_released) {
2971 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2972 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2980void __kmpc_unset_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2981#if KMP_USE_DYNAMIC_LOCK
2984 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2986 int release_status =
2987 KMP_D_LOCK_FUNC(user_lock, unset)((kmp_dyna_lock_t *)user_lock, gtid);
2988 (void)release_status;
2990#if OMPT_SUPPORT && OMPT_OPTIONAL
2992 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2994 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2995 if (ompt_enabled.enabled) {
2996 if (release_status == KMP_LOCK_RELEASED) {
2997 if (ompt_enabled.ompt_callback_mutex_released) {
2999 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3000 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3003 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3005 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3006 ompt_scope_end, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3013 kmp_user_lock_p lck;
3017 if ((__kmp_user_lock_kind == lk_tas) &&
3018 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3019 OMP_NEST_LOCK_T_SIZE)) {
3020#if KMP_OS_LINUX && \
3021 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
3023 kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
3025 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3028#if OMPT_SUPPORT && OMPT_OPTIONAL
3029 int release_status = KMP_LOCK_STILL_HELD;
3032 if (--(tl->lk.depth_locked) == 0) {
3033 TCW_4(tl->lk.poll, 0);
3034#if OMPT_SUPPORT && OMPT_OPTIONAL
3035 release_status = KMP_LOCK_RELEASED;
3040#if OMPT_SUPPORT && OMPT_OPTIONAL
3042 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3044 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3045 if (ompt_enabled.enabled) {
3046 if (release_status == KMP_LOCK_RELEASED) {
3047 if (ompt_enabled.ompt_callback_mutex_released) {
3049 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3050 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3052 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3054 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3055 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3062 lck = (kmp_user_lock_p)user_lock;
3066 else if ((__kmp_user_lock_kind == lk_futex) &&
3067 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3068 OMP_NEST_LOCK_T_SIZE)) {
3069 lck = (kmp_user_lock_p)user_lock;
3073 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_nest_lock");
3077 __kmp_itt_lock_releasing(lck);
3081 release_status = RELEASE_NESTED_LOCK(lck, gtid);
3082#if OMPT_SUPPORT && OMPT_OPTIONAL
3084 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3086 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3087 if (ompt_enabled.enabled) {
3088 if (release_status == KMP_LOCK_RELEASED) {
3089 if (ompt_enabled.ompt_callback_mutex_released) {
3091 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3092 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3094 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3096 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3097 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3106int __kmpc_test_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3109#if KMP_USE_DYNAMIC_LOCK
3111 int tag = KMP_EXTRACT_D_TAG(user_lock);
3113 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3115#if OMPT_SUPPORT && OMPT_OPTIONAL
3117 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3119 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3120 if (ompt_enabled.ompt_callback_mutex_acquire) {
3121 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3122 ompt_mutex_lock, omp_lock_hint_none,
3123 __ompt_get_mutex_impl_type(user_lock),
3124 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3127#if KMP_USE_INLINED_TAS
3128 if (tag == locktag_tas && !__kmp_env_consistency_check) {
3129 KMP_TEST_TAS_LOCK(user_lock, gtid, rc);
3131#elif KMP_USE_INLINED_FUTEX
3132 if (tag == locktag_futex && !__kmp_env_consistency_check) {
3133 KMP_TEST_FUTEX_LOCK(user_lock, gtid, rc);
3137 rc = __kmp_direct_test[tag]((kmp_dyna_lock_t *)user_lock, gtid);
3141 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3143#if OMPT_SUPPORT && OMPT_OPTIONAL
3144 if (ompt_enabled.ompt_callback_mutex_acquired) {
3145 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3146 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3152 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3159 kmp_user_lock_p lck;
3162 if ((__kmp_user_lock_kind == lk_tas) &&
3163 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
3164 lck = (kmp_user_lock_p)user_lock;
3167 else if ((__kmp_user_lock_kind == lk_futex) &&
3168 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3169 lck = (kmp_user_lock_p)user_lock;
3173 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_lock");
3177 __kmp_itt_lock_acquiring(lck);
3179#if OMPT_SUPPORT && OMPT_OPTIONAL
3181 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3183 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3184 if (ompt_enabled.ompt_callback_mutex_acquire) {
3185 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3186 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
3187 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3191 rc = TEST_LOCK(lck, gtid);
3194 __kmp_itt_lock_acquired(lck);
3196 __kmp_itt_lock_cancelled(lck);
3199#if OMPT_SUPPORT && OMPT_OPTIONAL
3200 if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
3201 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3202 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3206 return (rc ? FTN_TRUE : FTN_FALSE);
3214int __kmpc_test_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3215#if KMP_USE_DYNAMIC_LOCK
3218 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3220#if OMPT_SUPPORT && OMPT_OPTIONAL
3222 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3224 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3225 if (ompt_enabled.ompt_callback_mutex_acquire) {
3226 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3227 ompt_mutex_nest_lock, omp_lock_hint_none,
3228 __ompt_get_mutex_impl_type(user_lock),
3229 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3232 rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
3235 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3237 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3240#if OMPT_SUPPORT && OMPT_OPTIONAL
3241 if (ompt_enabled.enabled && rc) {
3243 if (ompt_enabled.ompt_callback_mutex_acquired) {
3245 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3246 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3250 if (ompt_enabled.ompt_callback_nest_lock) {
3252 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3253 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3262 kmp_user_lock_p lck;
3265 if ((__kmp_user_lock_kind == lk_tas) &&
3266 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3267 OMP_NEST_LOCK_T_SIZE)) {
3268 lck = (kmp_user_lock_p)user_lock;
3271 else if ((__kmp_user_lock_kind == lk_futex) &&
3272 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3273 OMP_NEST_LOCK_T_SIZE)) {
3274 lck = (kmp_user_lock_p)user_lock;
3278 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_nest_lock");
3282 __kmp_itt_lock_acquiring(lck);
3285#if OMPT_SUPPORT && OMPT_OPTIONAL
3287 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3289 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3290 if (ompt_enabled.enabled) &&
3291 ompt_enabled.ompt_callback_mutex_acquire) {
3292 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3293 ompt_mutex_nest_lock, omp_lock_hint_none,
3294 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
3299 rc = TEST_NESTED_LOCK(lck, gtid);
3302 __kmp_itt_lock_acquired(lck);
3304 __kmp_itt_lock_cancelled(lck);
3307#if OMPT_SUPPORT && OMPT_OPTIONAL
3308 if (ompt_enabled.enabled && rc) {
3310 if (ompt_enabled.ompt_callback_mutex_acquired) {
3312 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3313 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3316 if (ompt_enabled.ompt_callback_nest_lock) {
3318 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3319 ompt_mutex_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3338#define __KMP_SET_REDUCTION_METHOD(gtid, rmethod) \
3339 ((__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) = (rmethod))
3341#define __KMP_GET_REDUCTION_METHOD(gtid) \
3342 (__kmp_threads[(gtid)]->th.th_local.packed_reduction_method)
3348static __forceinline
void
3349__kmp_enter_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3350 kmp_critical_name *crit) {
3356 kmp_user_lock_p lck;
3358#if KMP_USE_DYNAMIC_LOCK
3360 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
3363 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3364 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
3365 KMP_GET_D_TAG(__kmp_user_lock_seq));
3367 __kmp_init_indirect_csptr(crit, loc, global_tid,
3368 KMP_GET_I_TAG(__kmp_user_lock_seq));
3374 if (KMP_EXTRACT_D_TAG(lk) != 0) {
3375 lck = (kmp_user_lock_p)lk;
3376 KMP_DEBUG_ASSERT(lck != NULL);
3377 if (__kmp_env_consistency_check) {
3378 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3380 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
3382 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
3384 KMP_DEBUG_ASSERT(lck != NULL);
3385 if (__kmp_env_consistency_check) {
3386 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3388 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
3396 if (__kmp_base_user_lock_size <= INTEL_CRITICAL_SIZE) {
3397 lck = (kmp_user_lock_p)crit;
3399 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
3401 KMP_DEBUG_ASSERT(lck != NULL);
3403 if (__kmp_env_consistency_check)
3404 __kmp_push_sync(global_tid, ct_critical, loc, lck);
3406 __kmp_acquire_user_lock_with_checks(lck, global_tid);
3412static __forceinline
void
3413__kmp_end_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3414 kmp_critical_name *crit) {
3416 kmp_user_lock_p lck;
3418#if KMP_USE_DYNAMIC_LOCK
3420 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3421 lck = (kmp_user_lock_p)crit;
3422 if (__kmp_env_consistency_check)
3423 __kmp_pop_sync(global_tid, ct_critical, loc);
3424 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
3426 kmp_indirect_lock_t *ilk =
3427 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
3428 if (__kmp_env_consistency_check)
3429 __kmp_pop_sync(global_tid, ct_critical, loc);
3430 KMP_I_LOCK_FUNC(ilk, unset)(ilk->lock, global_tid);
3438 if (__kmp_base_user_lock_size > 32) {
3439 lck = *((kmp_user_lock_p *)crit);
3440 KMP_ASSERT(lck != NULL);
3442 lck = (kmp_user_lock_p)crit;
3445 if (__kmp_env_consistency_check)
3446 __kmp_pop_sync(global_tid, ct_critical, loc);
3448 __kmp_release_user_lock_with_checks(lck, global_tid);
3453static __forceinline
int
3454__kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
3459 if (th->th.th_teams_microtask) {
3460 *team_p = team = th->th.th_team;
3461 if (team->t.t_level == th->th.th_teams_level) {
3463 KMP_DEBUG_ASSERT(!th->th.th_info.ds.ds_tid);
3465 th->th.th_info.ds.ds_tid = team->t.t_master_tid;
3466 th->th.th_team = team->t.t_parent;
3467 th->th.th_team_nproc = th->th.th_team->t.t_nproc;
3468 th->th.th_task_team = th->th.th_team->t.t_task_team[0];
3469 *task_state = th->th.th_task_state;
3470 th->th.th_task_state = 0;
3478static __forceinline
void
3479__kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team,
int task_state) {
3481 th->th.th_info.ds.ds_tid = 0;
3482 th->th.th_team = team;
3483 th->th.th_team_nproc = team->t.t_nproc;
3484 th->th.th_task_team = team->t.t_task_team[task_state];
3485 __kmp_type_convert(task_state, &(th->th.th_task_state));
3506 size_t reduce_size,
void *reduce_data,
3507 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3508 kmp_critical_name *lck) {
3512 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3515 int teams_swapped = 0, task_state;
3516 KA_TRACE(10, (
"__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
3517 __kmp_assert_valid_gtid(global_tid);
3525 if (!TCR_4(__kmp_init_parallel))
3526 __kmp_parallel_initialize();
3528 __kmp_resume_if_soft_paused();
3531#if KMP_USE_DYNAMIC_LOCK
3532 if (__kmp_env_consistency_check)
3533 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3535 if (__kmp_env_consistency_check)
3536 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3539 th = __kmp_thread_from_gtid(global_tid);
3540 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3558 packed_reduction_method = __kmp_determine_reduction_method(
3559 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3560 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3562 OMPT_REDUCTION_DECL(th, global_tid);
3563 if (packed_reduction_method == critical_reduce_block) {
3565 OMPT_REDUCTION_BEGIN;
3567 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3570 }
else if (packed_reduction_method == empty_reduce_block) {
3572 OMPT_REDUCTION_BEGIN;
3578 }
else if (packed_reduction_method == atomic_reduce_block) {
3588 if (__kmp_env_consistency_check)
3589 __kmp_pop_sync(global_tid, ct_reduce, loc);
3591 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3592 tree_reduce_block)) {
3612 ompt_frame_t *ompt_frame;
3613 if (ompt_enabled.enabled) {
3614 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3615 if (ompt_frame->enter_frame.ptr == NULL)
3616 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3618 OMPT_STORE_RETURN_ADDRESS(global_tid);
3621 __kmp_threads[global_tid]->th.th_ident = loc;
3624 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3625 global_tid, FALSE, reduce_size, reduce_data, reduce_func);
3626 retval = (retval != 0) ? (0) : (1);
3627#if OMPT_SUPPORT && OMPT_OPTIONAL
3628 if (ompt_enabled.enabled) {
3629 ompt_frame->enter_frame = ompt_data_none;
3635 if (__kmp_env_consistency_check) {
3637 __kmp_pop_sync(global_tid, ct_reduce, loc);
3646 if (teams_swapped) {
3647 __kmp_restore_swapped_teams(th, team, task_state);
3651 (
"__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
3652 global_tid, packed_reduction_method, retval));
3666 kmp_critical_name *lck) {
3668 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3670 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() enter: called T#%d\n", global_tid));
3671 __kmp_assert_valid_gtid(global_tid);
3673 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3675 OMPT_REDUCTION_DECL(__kmp_thread_from_gtid(global_tid), global_tid);
3677 if (packed_reduction_method == critical_reduce_block) {
3679 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3682 }
else if (packed_reduction_method == empty_reduce_block) {
3689 }
else if (packed_reduction_method == atomic_reduce_block) {
3696 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3697 tree_reduce_block)) {
3708 if (__kmp_env_consistency_check)
3709 __kmp_pop_sync(global_tid, ct_reduce, loc);
3711 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() exit: called T#%d: method %08x\n",
3712 global_tid, packed_reduction_method));
3735 size_t reduce_size,
void *reduce_data,
3736 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3737 kmp_critical_name *lck) {
3740 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3743 int teams_swapped = 0, task_state;
3745 KA_TRACE(10, (
"__kmpc_reduce() enter: called T#%d\n", global_tid));
3746 __kmp_assert_valid_gtid(global_tid);
3754 if (!TCR_4(__kmp_init_parallel))
3755 __kmp_parallel_initialize();
3757 __kmp_resume_if_soft_paused();
3760#if KMP_USE_DYNAMIC_LOCK
3761 if (__kmp_env_consistency_check)
3762 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3764 if (__kmp_env_consistency_check)
3765 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3768 th = __kmp_thread_from_gtid(global_tid);
3769 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3771 packed_reduction_method = __kmp_determine_reduction_method(
3772 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3773 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3775 OMPT_REDUCTION_DECL(th, global_tid);
3777 if (packed_reduction_method == critical_reduce_block) {
3779 OMPT_REDUCTION_BEGIN;
3780 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3783 }
else if (packed_reduction_method == empty_reduce_block) {
3785 OMPT_REDUCTION_BEGIN;
3790 }
else if (packed_reduction_method == atomic_reduce_block) {
3794 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3795 tree_reduce_block)) {
3801 ompt_frame_t *ompt_frame;
3802 if (ompt_enabled.enabled) {
3803 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3804 if (ompt_frame->enter_frame.ptr == NULL)
3805 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3807 OMPT_STORE_RETURN_ADDRESS(global_tid);
3810 __kmp_threads[global_tid]->th.th_ident =
3814 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3815 global_tid, TRUE, reduce_size, reduce_data, reduce_func);
3816 retval = (retval != 0) ? (0) : (1);
3817#if OMPT_SUPPORT && OMPT_OPTIONAL
3818 if (ompt_enabled.enabled) {
3819 ompt_frame->enter_frame = ompt_data_none;
3825 if (__kmp_env_consistency_check) {
3827 __kmp_pop_sync(global_tid, ct_reduce, loc);
3836 if (teams_swapped) {
3837 __kmp_restore_swapped_teams(th, team, task_state);
3841 (
"__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
3842 global_tid, packed_reduction_method, retval));
3857 kmp_critical_name *lck) {
3859 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3862 int teams_swapped = 0, task_state;
3864 KA_TRACE(10, (
"__kmpc_end_reduce() enter: called T#%d\n", global_tid));
3865 __kmp_assert_valid_gtid(global_tid);
3867 th = __kmp_thread_from_gtid(global_tid);
3868 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3870 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3874 OMPT_REDUCTION_DECL(th, global_tid);
3876 if (packed_reduction_method == critical_reduce_block) {
3877 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3883 ompt_frame_t *ompt_frame;
3884 if (ompt_enabled.enabled) {
3885 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3886 if (ompt_frame->enter_frame.ptr == NULL)
3887 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3889 OMPT_STORE_RETURN_ADDRESS(global_tid);
3892 __kmp_threads[global_tid]->th.th_ident = loc;
3894 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3895#if OMPT_SUPPORT && OMPT_OPTIONAL
3896 if (ompt_enabled.enabled) {
3897 ompt_frame->enter_frame = ompt_data_none;
3901 }
else if (packed_reduction_method == empty_reduce_block) {
3909 ompt_frame_t *ompt_frame;
3910 if (ompt_enabled.enabled) {
3911 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3912 if (ompt_frame->enter_frame.ptr == NULL)
3913 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3915 OMPT_STORE_RETURN_ADDRESS(global_tid);
3918 __kmp_threads[global_tid]->th.th_ident = loc;
3920 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3921#if OMPT_SUPPORT && OMPT_OPTIONAL
3922 if (ompt_enabled.enabled) {
3923 ompt_frame->enter_frame = ompt_data_none;
3927 }
else if (packed_reduction_method == atomic_reduce_block) {
3930 ompt_frame_t *ompt_frame;
3931 if (ompt_enabled.enabled) {
3932 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3933 if (ompt_frame->enter_frame.ptr == NULL)
3934 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3936 OMPT_STORE_RETURN_ADDRESS(global_tid);
3940 __kmp_threads[global_tid]->th.th_ident = loc;
3942 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3943#if OMPT_SUPPORT && OMPT_OPTIONAL
3944 if (ompt_enabled.enabled) {
3945 ompt_frame->enter_frame = ompt_data_none;
3949 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3950 tree_reduce_block)) {
3953 __kmp_end_split_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3961 if (teams_swapped) {
3962 __kmp_restore_swapped_teams(th, team, task_state);
3965 if (__kmp_env_consistency_check)
3966 __kmp_pop_sync(global_tid, ct_reduce, loc);
3968 KA_TRACE(10, (
"__kmpc_end_reduce() exit: called T#%d: method %08x\n",
3969 global_tid, packed_reduction_method));
3974#undef __KMP_GET_REDUCTION_METHOD
3975#undef __KMP_SET_REDUCTION_METHOD
3979kmp_uint64 __kmpc_get_taskid() {
3984 gtid = __kmp_get_gtid();
3988 thread = __kmp_thread_from_gtid(gtid);
3989 return thread->th.th_current_task->td_task_id;
3993kmp_uint64 __kmpc_get_parent_taskid() {
3997 kmp_taskdata_t *parent_task;
3999 gtid = __kmp_get_gtid();
4003 thread = __kmp_thread_from_gtid(gtid);
4004 parent_task = thread->th.th_current_task->td_parent;
4005 return (parent_task == NULL ? 0 : parent_task->td_task_id);
4021 const struct kmp_dim *dims) {
4022 __kmp_assert_valid_gtid(gtid);
4024 kmp_int64 last, trace_count;
4025 kmp_info_t *th = __kmp_threads[gtid];
4026 kmp_team_t *team = th->th.th_team;
4028 kmp_disp_t *pr_buf = th->th.th_dispatch;
4029 dispatch_shared_info_t *sh_buf;
4033 (
"__kmpc_doacross_init() enter: called T#%d, num dims %d, active %d\n",
4034 gtid, num_dims, !team->t.t_serialized));
4035 KMP_DEBUG_ASSERT(dims != NULL);
4036 KMP_DEBUG_ASSERT(num_dims > 0);
4038 if (team->t.t_serialized) {
4039 KA_TRACE(20, (
"__kmpc_doacross_init() exit: serialized team\n"));
4042 KMP_DEBUG_ASSERT(team->t.t_nproc > 1);
4043 idx = pr_buf->th_doacross_buf_idx++;
4045 sh_buf = &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4048 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info == NULL);
4049 pr_buf->th_doacross_info = (kmp_int64 *)__kmp_thread_malloc(
4050 th,
sizeof(kmp_int64) * (4 * num_dims + 1));
4051 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4052 pr_buf->th_doacross_info[0] =
4053 (kmp_int64)num_dims;
4056 pr_buf->th_doacross_info[1] = (kmp_int64)&sh_buf->doacross_num_done;
4057 pr_buf->th_doacross_info[2] = dims[0].lo;
4058 pr_buf->th_doacross_info[3] = dims[0].up;
4059 pr_buf->th_doacross_info[4] = dims[0].st;
4061 for (j = 1; j < num_dims; ++j) {
4064 if (dims[j].st == 1) {
4066 range_length = dims[j].up - dims[j].lo + 1;
4068 if (dims[j].st > 0) {
4069 KMP_DEBUG_ASSERT(dims[j].up > dims[j].lo);
4070 range_length = (kmp_uint64)(dims[j].up - dims[j].lo) / dims[j].st + 1;
4072 KMP_DEBUG_ASSERT(dims[j].lo > dims[j].up);
4074 (kmp_uint64)(dims[j].lo - dims[j].up) / (-dims[j].st) + 1;
4077 pr_buf->th_doacross_info[last++] = range_length;
4078 pr_buf->th_doacross_info[last++] = dims[j].lo;
4079 pr_buf->th_doacross_info[last++] = dims[j].up;
4080 pr_buf->th_doacross_info[last++] = dims[j].st;
4085 if (dims[0].st == 1) {
4086 trace_count = dims[0].up - dims[0].lo + 1;
4087 }
else if (dims[0].st > 0) {
4088 KMP_DEBUG_ASSERT(dims[0].up > dims[0].lo);
4089 trace_count = (kmp_uint64)(dims[0].up - dims[0].lo) / dims[0].st + 1;
4091 KMP_DEBUG_ASSERT(dims[0].lo > dims[0].up);
4092 trace_count = (kmp_uint64)(dims[0].lo - dims[0].up) / (-dims[0].st) + 1;
4094 for (j = 1; j < num_dims; ++j) {
4095 trace_count *= pr_buf->th_doacross_info[4 * j + 1];
4097 KMP_DEBUG_ASSERT(trace_count > 0);
4101 if (idx != sh_buf->doacross_buf_idx) {
4103 __kmp_wait_4((
volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
4110 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET32(
4111 (
volatile kmp_int32 *)&sh_buf->doacross_flags, NULL, 1);
4113 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET64(
4114 (
volatile kmp_int64 *)&sh_buf->doacross_flags, NULL, 1LL);
4116 if (flags == NULL) {
4119 (size_t)trace_count / 8 + 8;
4120 flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
4122 sh_buf->doacross_flags = flags;
4123 }
else if (flags == (kmp_uint32 *)1) {
4126 while (*(
volatile kmp_int32 *)&sh_buf->doacross_flags == 1)
4128 while (*(
volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
4135 KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1);
4136 pr_buf->th_doacross_flags =
4137 sh_buf->doacross_flags;
4139 KA_TRACE(20, (
"__kmpc_doacross_init() exit: T#%d\n", gtid));
4142void __kmpc_doacross_wait(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4143 __kmp_assert_valid_gtid(gtid);
4147 kmp_int64 iter_number;
4148 kmp_info_t *th = __kmp_threads[gtid];
4149 kmp_team_t *team = th->th.th_team;
4151 kmp_int64 lo, up, st;
4153 KA_TRACE(20, (
"__kmpc_doacross_wait() enter: called T#%d\n", gtid));
4154 if (team->t.t_serialized) {
4155 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: serialized team\n"));
4160 pr_buf = th->th.th_dispatch;
4161 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4162 num_dims = (size_t)pr_buf->th_doacross_info[0];
4163 lo = pr_buf->th_doacross_info[2];
4164 up = pr_buf->th_doacross_info[3];
4165 st = pr_buf->th_doacross_info[4];
4166#
if OMPT_SUPPORT && OMPT_OPTIONAL
4167 ompt_dependence_t deps[num_dims];
4170 if (vec[0] < lo || vec[0] > up) {
4171 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4172 "bounds [%lld,%lld]\n",
4173 gtid, vec[0], lo, up));
4176 iter_number = vec[0] - lo;
4177 }
else if (st > 0) {
4178 if (vec[0] < lo || vec[0] > up) {
4179 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4180 "bounds [%lld,%lld]\n",
4181 gtid, vec[0], lo, up));
4184 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4186 if (vec[0] > lo || vec[0] < up) {
4187 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4188 "bounds [%lld,%lld]\n",
4189 gtid, vec[0], lo, up));
4192 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4194#if OMPT_SUPPORT && OMPT_OPTIONAL
4195 deps[0].variable.value = iter_number;
4196 deps[0].dependence_type = ompt_dependence_type_sink;
4198 for (i = 1; i < num_dims; ++i) {
4201 ln = pr_buf->th_doacross_info[j + 1];
4202 lo = pr_buf->th_doacross_info[j + 2];
4203 up = pr_buf->th_doacross_info[j + 3];
4204 st = pr_buf->th_doacross_info[j + 4];
4206 if (vec[i] < lo || vec[i] > up) {
4207 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4208 "bounds [%lld,%lld]\n",
4209 gtid, vec[i], lo, up));
4213 }
else if (st > 0) {
4214 if (vec[i] < lo || vec[i] > up) {
4215 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4216 "bounds [%lld,%lld]\n",
4217 gtid, vec[i], lo, up));
4220 iter = (kmp_uint64)(vec[i] - lo) / st;
4222 if (vec[i] > lo || vec[i] < up) {
4223 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4224 "bounds [%lld,%lld]\n",
4225 gtid, vec[i], lo, up));
4228 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4230 iter_number = iter + ln * iter_number;
4231#if OMPT_SUPPORT && OMPT_OPTIONAL
4232 deps[i].variable.value = iter;
4233 deps[i].dependence_type = ompt_dependence_type_sink;
4236 shft = iter_number % 32;
4239 while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
4243#if OMPT_SUPPORT && OMPT_OPTIONAL
4244 if (ompt_enabled.ompt_callback_dependences) {
4245 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4246 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4250 (
"__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
4251 gtid, (iter_number << 5) + shft));
4254void __kmpc_doacross_post(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4255 __kmp_assert_valid_gtid(gtid);
4259 kmp_int64 iter_number;
4260 kmp_info_t *th = __kmp_threads[gtid];
4261 kmp_team_t *team = th->th.th_team;
4265 KA_TRACE(20, (
"__kmpc_doacross_post() enter: called T#%d\n", gtid));
4266 if (team->t.t_serialized) {
4267 KA_TRACE(20, (
"__kmpc_doacross_post() exit: serialized team\n"));
4273 pr_buf = th->th.th_dispatch;
4274 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4275 num_dims = (size_t)pr_buf->th_doacross_info[0];
4276 lo = pr_buf->th_doacross_info[2];
4277 st = pr_buf->th_doacross_info[4];
4278#
if OMPT_SUPPORT && OMPT_OPTIONAL
4279 ompt_dependence_t deps[num_dims];
4282 iter_number = vec[0] - lo;
4283 }
else if (st > 0) {
4284 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4286 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4288#if OMPT_SUPPORT && OMPT_OPTIONAL
4289 deps[0].variable.value = iter_number;
4290 deps[0].dependence_type = ompt_dependence_type_source;
4292 for (i = 1; i < num_dims; ++i) {
4295 ln = pr_buf->th_doacross_info[j + 1];
4296 lo = pr_buf->th_doacross_info[j + 2];
4297 st = pr_buf->th_doacross_info[j + 4];
4300 }
else if (st > 0) {
4301 iter = (kmp_uint64)(vec[i] - lo) / st;
4303 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4305 iter_number = iter + ln * iter_number;
4306#if OMPT_SUPPORT && OMPT_OPTIONAL
4307 deps[i].variable.value = iter;
4308 deps[i].dependence_type = ompt_dependence_type_source;
4311#if OMPT_SUPPORT && OMPT_OPTIONAL
4312 if (ompt_enabled.ompt_callback_dependences) {
4313 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4314 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4317 shft = iter_number % 32;
4321 if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
4322 KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
4323 KA_TRACE(20, (
"__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
4324 (iter_number << 5) + shft));
4327void __kmpc_doacross_fini(
ident_t *loc,
int gtid) {
4328 __kmp_assert_valid_gtid(gtid);
4330 kmp_info_t *th = __kmp_threads[gtid];
4331 kmp_team_t *team = th->th.th_team;
4332 kmp_disp_t *pr_buf = th->th.th_dispatch;
4334 KA_TRACE(20, (
"__kmpc_doacross_fini() enter: called T#%d\n", gtid));
4335 if (team->t.t_serialized) {
4336 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: serialized team %p\n", team));
4340 KMP_TEST_THEN_INC32((kmp_uintptr_t)(pr_buf->th_doacross_info[1])) + 1;
4341 if (num_done == th->th.th_team_nproc) {
4343 int idx = pr_buf->th_doacross_buf_idx - 1;
4344 dispatch_shared_info_t *sh_buf =
4345 &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4346 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info[1] ==
4347 (kmp_int64)&sh_buf->doacross_num_done);
4348 KMP_DEBUG_ASSERT(num_done == sh_buf->doacross_num_done);
4349 KMP_DEBUG_ASSERT(idx == sh_buf->doacross_buf_idx);
4350 __kmp_thread_free(th, CCAST(kmp_uint32 *, sh_buf->doacross_flags));
4351 sh_buf->doacross_flags = NULL;
4352 sh_buf->doacross_num_done = 0;
4353 sh_buf->doacross_buf_idx +=
4354 __kmp_dispatch_num_buffers;
4357 pr_buf->th_doacross_flags = NULL;
4358 __kmp_thread_free(th, (
void *)pr_buf->th_doacross_info);
4359 pr_buf->th_doacross_info = NULL;
4360 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: T#%d\n", gtid));
4364void *omp_alloc(
size_t size, omp_allocator_handle_t allocator) {
4365 return __kmp_alloc(__kmp_entry_gtid(), 0, size, allocator);
4368void *omp_aligned_alloc(
size_t align,
size_t size,
4369 omp_allocator_handle_t allocator) {
4370 return __kmp_alloc(__kmp_entry_gtid(), align, size, allocator);
4373void *omp_calloc(
size_t nmemb,
size_t size, omp_allocator_handle_t allocator) {
4374 return __kmp_calloc(__kmp_entry_gtid(), 0, nmemb, size, allocator);
4377void *omp_aligned_calloc(
size_t align,
size_t nmemb,
size_t size,
4378 omp_allocator_handle_t allocator) {
4379 return __kmp_calloc(__kmp_entry_gtid(), align, nmemb, size, allocator);
4382void *omp_realloc(
void *ptr,
size_t size, omp_allocator_handle_t allocator,
4383 omp_allocator_handle_t free_allocator) {
4384 return __kmp_realloc(__kmp_entry_gtid(), ptr, size, allocator,
4388void omp_free(
void *ptr, omp_allocator_handle_t allocator) {
4389 ___kmpc_free(__kmp_entry_gtid(), ptr, allocator);
4393int __kmpc_get_target_offload(
void) {
4394 if (!__kmp_init_serial) {
4395 __kmp_serial_initialize();
4397 return __kmp_target_offload;
4400int __kmpc_pause_resource(kmp_pause_status_t level) {
4401 if (!__kmp_init_serial) {
4404 return __kmp_pause_resource(level);
4407void __kmpc_error(
ident_t *loc,
int severity,
const char *message) {
4408 if (!__kmp_init_serial)
4409 __kmp_serial_initialize();
4411 KMP_ASSERT(severity == severity_warning || severity == severity_fatal);
4414 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_error) {
4415 ompt_callbacks.ompt_callback(ompt_callback_error)(
4416 (ompt_severity_t)severity, message, KMP_STRLEN(message),
4417 OMPT_GET_RETURN_ADDRESS(0));
4423 kmp_str_loc_t str_loc = __kmp_str_loc_init(loc->
psource,
false);
4425 __kmp_str_format(
"%s:%s:%s", str_loc.file, str_loc.line, str_loc.col);
4426 __kmp_str_loc_free(&str_loc);
4428 src_loc = __kmp_str_format(
"unknown");
4431 if (severity == severity_warning)
4432 KMP_WARNING(UserDirectedWarning, src_loc, message);
4434 KMP_FATAL(UserDirectedError, src_loc, message);
4436 __kmp_str_free(&src_loc);
4440void __kmpc_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4442#if OMPT_SUPPORT && OMPT_OPTIONAL
4443 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4444 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4445 int tid = __kmp_tid_from_gtid(gtid);
4446 ompt_callbacks.ompt_callback(ompt_callback_work)(
4447 ompt_work_scope, ompt_scope_begin,
4448 &(team->t.ompt_team_info.parallel_data),
4449 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4450 OMPT_GET_RETURN_ADDRESS(0));
4456void __kmpc_end_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4458#if OMPT_SUPPORT && OMPT_OPTIONAL
4459 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4460 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4461 int tid = __kmp_tid_from_gtid(gtid);
4462 ompt_callbacks.ompt_callback(ompt_callback_work)(
4463 ompt_work_scope, ompt_scope_end,
4464 &(team->t.ompt_team_info.parallel_data),
4465 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4466 OMPT_GET_RETURN_ADDRESS(0));
4471#ifdef KMP_USE_VERSION_SYMBOLS
4480#ifdef omp_set_affinity_format
4481#undef omp_set_affinity_format
4483#ifdef omp_get_affinity_format
4484#undef omp_get_affinity_format
4486#ifdef omp_display_affinity
4487#undef omp_display_affinity
4489#ifdef omp_capture_affinity
4490#undef omp_capture_affinity
4492KMP_VERSION_OMPC_SYMBOL(ompc_set_affinity_format, omp_set_affinity_format, 50,
4494KMP_VERSION_OMPC_SYMBOL(ompc_get_affinity_format, omp_get_affinity_format, 50,
4496KMP_VERSION_OMPC_SYMBOL(ompc_display_affinity, omp_display_affinity, 50,
4498KMP_VERSION_OMPC_SYMBOL(ompc_capture_affinity, omp_capture_affinity, 50,
@ KMP_IDENT_WORK_SECTIONS
@ KMP_IDENT_WORK_DISTRIBUTE
kmp_int32 __kmpc_ok_to_fork(ident_t *loc)
void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)
void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_threads)
void __kmpc_serialized_parallel(ident_t *loc, kmp_int32 global_tid)
void __kmpc_push_num_teams(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams, kmp_int32 num_threads)
void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)
void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid)
void(* kmpc_micro)(kmp_int32 *global_tid, kmp_int32 *bound_tid,...)
void __kmpc_push_num_teams_51(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams_lb, kmp_int32 num_teams_ub, kmp_int32 num_threads)
void __kmpc_begin(ident_t *loc, kmp_int32 flags)
void __kmpc_end(ident_t *loc)
#define KMP_COUNT_BLOCK(name)
Increments specified counter (name).
stats_state_e
the states which a thread can be in
void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
void __kmpc_end_barrier_master(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_barrier_master_nowait(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
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)
void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid)
void __kmpc_flush(ident_t *loc)
kmp_int32 __kmpc_barrier_master(ident_t *loc, kmp_int32 global_tid)
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_copyprivate_light(ident_t *loc, kmp_int32 gtid, void *cpy_data)
kmp_int32 __kmpc_global_num_threads(ident_t *loc)
kmp_int32 __kmpc_global_thread_num(ident_t *loc)
kmp_int32 __kmpc_in_parallel(ident_t *loc)
kmp_int32 __kmpc_bound_thread_num(ident_t *loc)
kmp_int32 __kmpc_bound_num_threads(ident_t *loc)
void __kmpc_end_ordered(ident_t *loc, kmp_int32 gtid)
void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)
void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_masked(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit, uint32_t hint)
kmp_int32 __kmpc_single(ident_t *loc, kmp_int32 global_tid)
void __kmpc_doacross_init(ident_t *loc, int gtid, int num_dims, const struct kmp_dim *dims)
void __kmpc_end_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_single(ident_t *loc, kmp_int32 global_tid)
void __kmpc_ordered(ident_t *loc, kmp_int32 gtid)
kmp_int32 __kmpc_masked(ident_t *loc, kmp_int32 global_tid, kmp_int32 filter)
void __kmpc_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)