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();
344 kmp_int32 cond,
void *args) {
345 int gtid = __kmp_entry_gtid();
355 void *exit_frame_ptr;
359 __kmp_invoke_microtask(VOLATILE_CAST(microtask_t) microtask, gtid,
368 __kmp_invoke_microtask(VOLATILE_CAST(microtask_t) microtask, gtid,
394 kmp_int32 num_teams, kmp_int32 num_threads) {
396 (
"__kmpc_push_num_teams: enter T#%d num_teams=%d num_threads=%d\n",
397 global_tid, num_teams, num_threads));
398 __kmp_assert_valid_gtid(global_tid);
399 __kmp_push_num_teams(loc, global_tid, num_teams, num_threads);
413 kmp_int32 thread_limit) {
414 __kmp_assert_valid_gtid(global_tid);
415 kmp_info_t *thread = __kmp_threads[global_tid];
416 if (thread_limit > 0)
417 thread->th.th_current_task->td_icvs.task_thread_limit = thread_limit;
437 kmp_int32 num_teams_lb, kmp_int32 num_teams_ub,
438 kmp_int32 num_threads) {
439 KA_TRACE(20, (
"__kmpc_push_num_teams_51: enter T#%d num_teams_lb=%d"
440 " num_teams_ub=%d num_threads=%d\n",
441 global_tid, num_teams_lb, num_teams_ub, num_threads));
442 __kmp_assert_valid_gtid(global_tid);
443 __kmp_push_num_teams_51(loc, global_tid, num_teams_lb, num_teams_ub,
459 int gtid = __kmp_entry_gtid();
460 kmp_info_t *this_thr = __kmp_threads[gtid];
462 va_start(ap, microtask);
467 if (previous_state == stats_state_e::SERIAL_REGION) {
468 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_teams_overhead);
470 KMP_PUSH_PARTITIONED_TIMER(OMP_teams_overhead);
475 this_thr->th.th_teams_microtask = microtask;
476 this_thr->th.th_teams_level =
477 this_thr->th.th_team->t.t_level;
480 kmp_team_t *parent_team = this_thr->th.th_team;
481 int tid = __kmp_tid_from_gtid(gtid);
482 if (ompt_enabled.enabled) {
483 parent_team->t.t_implicit_task_taskdata[tid]
484 .ompt_task_info.frame.enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
486 OMPT_STORE_RETURN_ADDRESS(gtid);
491 if (this_thr->th.th_teams_size.nteams == 0) {
492 __kmp_push_num_teams(loc, gtid, 0, 0);
494 KMP_DEBUG_ASSERT(this_thr->th.th_set_nproc >= 1);
495 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nteams >= 1);
496 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nth >= 1);
499 loc, gtid, fork_context_intel, argc,
500 VOLATILE_CAST(microtask_t) __kmp_teams_master,
501 VOLATILE_CAST(launch_t) __kmp_invoke_teams_master, kmp_va_addr_of(ap));
502 __kmp_join_call(loc, gtid
510 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
511 kmp_cg_root_t *tmp = this_thr->th.th_cg_roots;
512 this_thr->th.th_cg_roots = tmp->up;
513 KA_TRACE(100, (
"__kmpc_fork_teams: Thread %p popping node %p and moving up"
514 " to node %p. cg_nthreads was %d\n",
515 this_thr, tmp, this_thr->th.th_cg_roots, tmp->cg_nthreads));
516 KMP_DEBUG_ASSERT(tmp->cg_nthreads);
517 int i = tmp->cg_nthreads--;
522 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
523 this_thr->th.th_current_task->td_icvs.thread_limit =
524 this_thr->th.th_cg_roots->cg_thread_limit;
526 this_thr->th.th_teams_microtask = NULL;
527 this_thr->th.th_teams_level = 0;
528 *(kmp_int64 *)(&this_thr->th.th_teams_size) = 0L;
531 if (previous_state == stats_state_e::SERIAL_REGION) {
532 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
533 KMP_SET_THREAD_STATE(previous_state);
535 KMP_POP_PARTITIONED_TIMER();
544int __kmpc_invoke_task_func(
int gtid) {
return __kmp_invoke_task_func(gtid); }
562 __kmp_assert_valid_gtid(global_tid);
564 OMPT_STORE_RETURN_ADDRESS(global_tid);
566 __kmp_serialized_parallel(loc, global_tid);
577 kmp_internal_control_t *top;
578 kmp_info_t *this_thr;
579 kmp_team_t *serial_team;
582 (
"__kmpc_end_serialized_parallel: called by T#%d\n", global_tid));
590 __kmp_assert_valid_gtid(global_tid);
591 if (!TCR_4(__kmp_init_parallel))
592 __kmp_parallel_initialize();
594 __kmp_resume_if_soft_paused();
596 this_thr = __kmp_threads[global_tid];
597 serial_team = this_thr->th.th_serial_team;
599 kmp_task_team_t *task_team = this_thr->th.th_task_team;
601 if (task_team != NULL && (task_team->tt.tt_found_proxy_tasks ||
602 task_team->tt.tt_hidden_helper_task_encountered))
603 __kmp_task_team_wait(this_thr, serial_team USE_ITT_BUILD_ARG(NULL));
606 KMP_DEBUG_ASSERT(serial_team);
607 KMP_ASSERT(serial_team->t.t_serialized);
608 KMP_DEBUG_ASSERT(this_thr->th.th_team == serial_team);
609 KMP_DEBUG_ASSERT(serial_team != this_thr->th.th_root->r.r_root_team);
610 KMP_DEBUG_ASSERT(serial_team->t.t_threads);
611 KMP_DEBUG_ASSERT(serial_team->t.t_threads[0] == this_thr);
614 if (ompt_enabled.enabled &&
615 this_thr->th.ompt_thread_info.state != ompt_state_overhead) {
616 OMPT_CUR_TASK_INFO(this_thr)->frame.exit_frame = ompt_data_none;
617 if (ompt_enabled.ompt_callback_implicit_task) {
618 ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
619 ompt_scope_end, NULL, OMPT_CUR_TASK_DATA(this_thr), 1,
620 OMPT_CUR_TASK_INFO(this_thr)->thread_num, ompt_task_implicit);
624 ompt_data_t *parent_task_data;
625 __ompt_get_task_info_internal(1, NULL, &parent_task_data, NULL, NULL, NULL);
627 if (ompt_enabled.ompt_callback_parallel_end) {
628 ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
629 &(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
630 ompt_parallel_invoker_program | ompt_parallel_team,
631 OMPT_LOAD_RETURN_ADDRESS(global_tid));
633 __ompt_lw_taskteam_unlink(this_thr);
634 this_thr->th.ompt_thread_info.state = ompt_state_overhead;
640 top = serial_team->t.t_control_stack_top;
641 if (top && top->serial_nesting_level == serial_team->t.t_serialized) {
642 copy_icvs(&serial_team->t.t_threads[0]->th.th_current_task->td_icvs, top);
643 serial_team->t.t_control_stack_top = top->next;
648 KMP_DEBUG_ASSERT(serial_team->t.t_dispatch->th_disp_buffer);
650 dispatch_private_info_t *disp_buffer =
651 serial_team->t.t_dispatch->th_disp_buffer;
652 serial_team->t.t_dispatch->th_disp_buffer =
653 serial_team->t.t_dispatch->th_disp_buffer->next;
654 __kmp_free(disp_buffer);
656 this_thr->th.th_def_allocator = serial_team->t.t_def_allocator;
658 --serial_team->t.t_serialized;
659 if (serial_team->t.t_serialized == 0) {
663#if KMP_ARCH_X86 || KMP_ARCH_X86_64
664 if (__kmp_inherit_fp_control && serial_team->t.t_fp_control_saved) {
665 __kmp_clear_x87_fpu_status_word();
666 __kmp_load_x87_fpu_control_word(&serial_team->t.t_x87_fpu_control_word);
667 __kmp_load_mxcsr(&serial_team->t.t_mxcsr);
671 __kmp_pop_current_task_from_thread(this_thr);
673 if (ompd_state & OMPD_ENABLE_BP)
674 ompd_bp_parallel_end();
677 this_thr->th.th_team = serial_team->t.t_parent;
678 this_thr->th.th_info.ds.ds_tid = serial_team->t.t_master_tid;
681 this_thr->th.th_team_nproc = serial_team->t.t_parent->t.t_nproc;
682 this_thr->th.th_team_master =
683 serial_team->t.t_parent->t.t_threads[0];
684 this_thr->th.th_team_serialized = this_thr->th.th_team->t.t_serialized;
687 this_thr->th.th_dispatch =
688 &this_thr->th.th_team->t.t_dispatch[serial_team->t.t_master_tid];
690 KMP_ASSERT(this_thr->th.th_current_task->td_flags.executing == 0);
691 this_thr->th.th_current_task->td_flags.executing = 1;
693 if (__kmp_tasking_mode != tskm_immediate_exec) {
695 this_thr->th.th_task_team =
696 this_thr->th.th_team->t.t_task_team[this_thr->th.th_task_state];
698 (
"__kmpc_end_serialized_parallel: T#%d restoring task_team %p / "
700 global_tid, this_thr->th.th_task_team, this_thr->th.th_team));
702#if KMP_AFFINITY_SUPPORTED
703 if (this_thr->th.th_team->t.t_level == 0 && __kmp_affinity.flags.reset) {
704 __kmp_reset_root_init_mask(global_tid);
708 if (__kmp_tasking_mode != tskm_immediate_exec) {
709 KA_TRACE(20, (
"__kmpc_end_serialized_parallel: T#%d decreasing nesting "
710 "depth of serial team %p to %d\n",
711 global_tid, serial_team, serial_team->t.t_serialized));
715 serial_team->t.t_level--;
716 if (__kmp_env_consistency_check)
717 __kmp_pop_parallel(global_tid, NULL);
719 if (ompt_enabled.enabled)
720 this_thr->th.ompt_thread_info.state =
721 ((this_thr->th.th_team_serialized) ? ompt_state_work_serial
722 : ompt_state_work_parallel);
735 KC_TRACE(10, (
"__kmpc_flush: called\n"));
740#if OMPT_SUPPORT && OMPT_OPTIONAL
741 if (ompt_enabled.ompt_callback_flush) {
742 ompt_callbacks.ompt_callback(ompt_callback_flush)(
743 __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
758 KC_TRACE(10, (
"__kmpc_barrier: called T#%d\n", global_tid));
759 __kmp_assert_valid_gtid(global_tid);
761 if (!TCR_4(__kmp_init_parallel))
762 __kmp_parallel_initialize();
764 __kmp_resume_if_soft_paused();
766 if (__kmp_env_consistency_check) {
768 KMP_WARNING(ConstructIdentInvalid);
770 __kmp_check_barrier(global_tid, ct_barrier, loc);
774 ompt_frame_t *ompt_frame;
775 if (ompt_enabled.enabled) {
776 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
777 if (ompt_frame->enter_frame.ptr == NULL)
778 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
780 OMPT_STORE_RETURN_ADDRESS(global_tid);
782 __kmp_threads[global_tid]->th.th_ident = loc;
790 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
791#if OMPT_SUPPORT && OMPT_OPTIONAL
792 if (ompt_enabled.enabled) {
793 ompt_frame->enter_frame = ompt_data_none;
808 KC_TRACE(10, (
"__kmpc_master: called T#%d\n", global_tid));
809 __kmp_assert_valid_gtid(global_tid);
811 if (!TCR_4(__kmp_init_parallel))
812 __kmp_parallel_initialize();
814 __kmp_resume_if_soft_paused();
816 if (KMP_MASTER_GTID(global_tid)) {
818 KMP_PUSH_PARTITIONED_TIMER(OMP_master);
822#if OMPT_SUPPORT && OMPT_OPTIONAL
824 if (ompt_enabled.ompt_callback_masked) {
825 kmp_info_t *this_thr = __kmp_threads[global_tid];
826 kmp_team_t *team = this_thr->th.th_team;
828 int tid = __kmp_tid_from_gtid(global_tid);
829 ompt_callbacks.ompt_callback(ompt_callback_masked)(
830 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
831 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
832 OMPT_GET_RETURN_ADDRESS(0));
837 if (__kmp_env_consistency_check) {
838#if KMP_USE_DYNAMIC_LOCK
840 __kmp_push_sync(global_tid, ct_master, loc, NULL, 0);
842 __kmp_check_sync(global_tid, ct_master, loc, NULL, 0);
845 __kmp_push_sync(global_tid, ct_master, loc, NULL);
847 __kmp_check_sync(global_tid, ct_master, loc, NULL);
863 KC_TRACE(10, (
"__kmpc_end_master: called T#%d\n", global_tid));
864 __kmp_assert_valid_gtid(global_tid);
865 KMP_DEBUG_ASSERT(KMP_MASTER_GTID(global_tid));
866 KMP_POP_PARTITIONED_TIMER();
868#if OMPT_SUPPORT && OMPT_OPTIONAL
869 kmp_info_t *this_thr = __kmp_threads[global_tid];
870 kmp_team_t *team = this_thr->th.th_team;
871 if (ompt_enabled.ompt_callback_masked) {
872 int tid = __kmp_tid_from_gtid(global_tid);
873 ompt_callbacks.ompt_callback(ompt_callback_masked)(
874 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
875 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
876 OMPT_GET_RETURN_ADDRESS(0));
880 if (__kmp_env_consistency_check) {
881 if (KMP_MASTER_GTID(global_tid))
882 __kmp_pop_sync(global_tid, ct_master, loc);
897 KC_TRACE(10, (
"__kmpc_masked: called T#%d\n", global_tid));
898 __kmp_assert_valid_gtid(global_tid);
900 if (!TCR_4(__kmp_init_parallel))
901 __kmp_parallel_initialize();
903 __kmp_resume_if_soft_paused();
905 tid = __kmp_tid_from_gtid(global_tid);
908 KMP_PUSH_PARTITIONED_TIMER(OMP_masked);
912#if OMPT_SUPPORT && OMPT_OPTIONAL
914 if (ompt_enabled.ompt_callback_masked) {
915 kmp_info_t *this_thr = __kmp_threads[global_tid];
916 kmp_team_t *team = this_thr->th.th_team;
917 ompt_callbacks.ompt_callback(ompt_callback_masked)(
918 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
919 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
920 OMPT_GET_RETURN_ADDRESS(0));
925 if (__kmp_env_consistency_check) {
926#if KMP_USE_DYNAMIC_LOCK
928 __kmp_push_sync(global_tid, ct_masked, loc, NULL, 0);
930 __kmp_check_sync(global_tid, ct_masked, loc, NULL, 0);
933 __kmp_push_sync(global_tid, ct_masked, loc, NULL);
935 __kmp_check_sync(global_tid, ct_masked, loc, NULL);
951 KC_TRACE(10, (
"__kmpc_end_masked: called T#%d\n", global_tid));
952 __kmp_assert_valid_gtid(global_tid);
953 KMP_POP_PARTITIONED_TIMER();
955#if OMPT_SUPPORT && OMPT_OPTIONAL
956 kmp_info_t *this_thr = __kmp_threads[global_tid];
957 kmp_team_t *team = this_thr->th.th_team;
958 if (ompt_enabled.ompt_callback_masked) {
959 int tid = __kmp_tid_from_gtid(global_tid);
960 ompt_callbacks.ompt_callback(ompt_callback_masked)(
961 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
962 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
963 OMPT_GET_RETURN_ADDRESS(0));
967 if (__kmp_env_consistency_check) {
968 __kmp_pop_sync(global_tid, ct_masked, loc);
982 KMP_DEBUG_ASSERT(__kmp_init_serial);
984 KC_TRACE(10, (
"__kmpc_ordered: called T#%d\n", gtid));
985 __kmp_assert_valid_gtid(gtid);
987 if (!TCR_4(__kmp_init_parallel))
988 __kmp_parallel_initialize();
990 __kmp_resume_if_soft_paused();
993 __kmp_itt_ordered_prep(gtid);
997 th = __kmp_threads[gtid];
999#if OMPT_SUPPORT && OMPT_OPTIONAL
1003 OMPT_STORE_RETURN_ADDRESS(gtid);
1004 if (ompt_enabled.enabled) {
1005 team = __kmp_team_from_gtid(gtid);
1006 lck = (ompt_wait_id_t)(uintptr_t)&team->t.t_ordered.dt.t_value;
1008 th->th.ompt_thread_info.wait_id = lck;
1009 th->th.ompt_thread_info.state = ompt_state_wait_ordered;
1012 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1013 if (ompt_enabled.ompt_callback_mutex_acquire) {
1014 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1015 ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin, lck,
1021 if (th->th.th_dispatch->th_deo_fcn != 0)
1022 (*th->th.th_dispatch->th_deo_fcn)(>id, &cid, loc);
1024 __kmp_parallel_deo(>id, &cid, loc);
1026#if OMPT_SUPPORT && OMPT_OPTIONAL
1027 if (ompt_enabled.enabled) {
1029 th->th.ompt_thread_info.state = ompt_state_work_parallel;
1030 th->th.ompt_thread_info.wait_id = 0;
1033 if (ompt_enabled.ompt_callback_mutex_acquired) {
1034 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1035 ompt_mutex_ordered, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1041 __kmp_itt_ordered_start(gtid);
1056 KC_TRACE(10, (
"__kmpc_end_ordered: called T#%d\n", gtid));
1057 __kmp_assert_valid_gtid(gtid);
1060 __kmp_itt_ordered_end(gtid);
1064 th = __kmp_threads[gtid];
1066 if (th->th.th_dispatch->th_dxo_fcn != 0)
1067 (*th->th.th_dispatch->th_dxo_fcn)(>id, &cid, loc);
1069 __kmp_parallel_dxo(>id, &cid, loc);
1071#if OMPT_SUPPORT && OMPT_OPTIONAL
1072 OMPT_STORE_RETURN_ADDRESS(gtid);
1073 if (ompt_enabled.ompt_callback_mutex_released) {
1074 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1076 (ompt_wait_id_t)(uintptr_t)&__kmp_team_from_gtid(gtid)
1077 ->t.t_ordered.dt.t_value,
1078 OMPT_LOAD_RETURN_ADDRESS(gtid));
1083#if KMP_USE_DYNAMIC_LOCK
1085static __forceinline
void
1086__kmp_init_indirect_csptr(kmp_critical_name *crit,
ident_t const *loc,
1087 kmp_int32 gtid, kmp_indirect_locktag_t tag) {
1091 kmp_indirect_lock_t **lck;
1092 lck = (kmp_indirect_lock_t **)crit;
1093 kmp_indirect_lock_t *ilk = __kmp_allocate_indirect_lock(&idx, gtid, tag);
1094 KMP_I_LOCK_FUNC(ilk, init)(ilk->lock);
1095 KMP_SET_I_LOCK_LOCATION(ilk, loc);
1096 KMP_SET_I_LOCK_FLAGS(ilk, kmp_lf_critical_section);
1098 (
"__kmp_init_indirect_csptr: initialized indirect lock #%d\n", tag));
1100 __kmp_itt_critical_creating(ilk->lock, loc);
1102 int status = KMP_COMPARE_AND_STORE_PTR(lck,
nullptr, ilk);
1105 __kmp_itt_critical_destroyed(ilk->lock);
1111 KMP_DEBUG_ASSERT(*lck != NULL);
1115#define KMP_ACQUIRE_TAS_LOCK(lock, gtid) \
1117 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1118 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1119 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1120 if (KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1121 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \
1123 KMP_FSYNC_PREPARE(l); \
1124 KMP_INIT_YIELD(spins); \
1125 kmp_backoff_t backoff = __kmp_spin_backoff_params; \
1127 if (TCR_4(__kmp_nth) > \
1128 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
1131 KMP_YIELD_SPIN(spins); \
1133 __kmp_spin_backoff(&backoff); \
1135 KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1136 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)); \
1138 KMP_FSYNC_ACQUIRED(l); \
1142#define KMP_TEST_TAS_LOCK(lock, gtid, rc) \
1144 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1145 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1146 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1147 rc = KMP_ATOMIC_LD_RLX(&l->lk.poll) == tas_free && \
1148 __kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy); \
1152#define KMP_RELEASE_TAS_LOCK(lock, gtid) \
1153 { KMP_ATOMIC_ST_REL(&((kmp_tas_lock_t *)lock)->lk.poll, KMP_LOCK_FREE(tas)); }
1157#include <sys/syscall.h>
1167#define KMP_ACQUIRE_FUTEX_LOCK(lock, gtid) \
1169 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1170 kmp_int32 gtid_code = (gtid + 1) << 1; \
1172 KMP_FSYNC_PREPARE(ftx); \
1173 kmp_int32 poll_val; \
1174 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( \
1175 &(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1176 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { \
1177 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; \
1179 if (!KMP_COMPARE_AND_STORE_RET32(&(ftx->lk.poll), poll_val, \
1181 KMP_LOCK_BUSY(1, futex))) { \
1184 poll_val |= KMP_LOCK_BUSY(1, futex); \
1187 if ((rc = syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAIT, poll_val, \
1188 NULL, NULL, 0)) != 0) { \
1193 KMP_FSYNC_ACQUIRED(ftx); \
1197#define KMP_TEST_FUTEX_LOCK(lock, gtid, rc) \
1199 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1200 if (KMP_COMPARE_AND_STORE_ACQ32(&(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1201 KMP_LOCK_BUSY(gtid + 1 << 1, futex))) { \
1202 KMP_FSYNC_ACQUIRED(ftx); \
1210#define KMP_RELEASE_FUTEX_LOCK(lock, gtid) \
1212 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1214 KMP_FSYNC_RELEASING(ftx); \
1215 kmp_int32 poll_val = \
1216 KMP_XCHG_FIXED32(&(ftx->lk.poll), KMP_LOCK_FREE(futex)); \
1217 if (KMP_LOCK_STRIP(poll_val) & 1) { \
1218 syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAKE, \
1219 KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \
1222 KMP_YIELD_OVERSUB(); \
1229static kmp_user_lock_p __kmp_get_critical_section_ptr(kmp_critical_name *crit,
1232 kmp_user_lock_p *lck_pp = (kmp_user_lock_p *)crit;
1235 kmp_user_lock_p lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1242 lck = __kmp_user_lock_allocate(&idx, gtid, kmp_lf_critical_section);
1243 __kmp_init_user_lock_with_checks(lck);
1244 __kmp_set_user_lock_location(lck, loc);
1246 __kmp_itt_critical_creating(lck);
1257 int status = KMP_COMPARE_AND_STORE_PTR(lck_pp, 0, lck);
1262 __kmp_itt_critical_destroyed(lck);
1266 __kmp_destroy_user_lock_with_checks(lck);
1267 __kmp_user_lock_free(&idx, gtid, lck);
1268 lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1269 KMP_DEBUG_ASSERT(lck != NULL);
1288 kmp_critical_name *crit) {
1289#if KMP_USE_DYNAMIC_LOCK
1290#if OMPT_SUPPORT && OMPT_OPTIONAL
1291 OMPT_STORE_RETURN_ADDRESS(global_tid);
1293 __kmpc_critical_with_hint(loc, global_tid, crit, omp_lock_hint_none);
1296#if OMPT_SUPPORT && OMPT_OPTIONAL
1297 ompt_state_t prev_state = ompt_state_undefined;
1298 ompt_thread_info_t ti;
1300 kmp_user_lock_p lck;
1302 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1303 __kmp_assert_valid_gtid(global_tid);
1307 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1308 KMP_CHECK_USER_LOCK_INIT();
1310 if ((__kmp_user_lock_kind == lk_tas) &&
1311 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1312 lck = (kmp_user_lock_p)crit;
1315 else if ((__kmp_user_lock_kind == lk_futex) &&
1316 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1317 lck = (kmp_user_lock_p)crit;
1321 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
1324 if (__kmp_env_consistency_check)
1325 __kmp_push_sync(global_tid, ct_critical, loc, lck);
1333 __kmp_itt_critical_acquiring(lck);
1335#if OMPT_SUPPORT && OMPT_OPTIONAL
1336 OMPT_STORE_RETURN_ADDRESS(gtid);
1337 void *codeptr_ra = NULL;
1338 if (ompt_enabled.enabled) {
1339 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1341 prev_state = ti.state;
1342 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1343 ti.state = ompt_state_wait_critical;
1346 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1347 if (ompt_enabled.ompt_callback_mutex_acquire) {
1348 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1349 ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
1350 (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1356 __kmp_acquire_user_lock_with_checks(lck, global_tid);
1359 __kmp_itt_critical_acquired(lck);
1361#if OMPT_SUPPORT && OMPT_OPTIONAL
1362 if (ompt_enabled.enabled) {
1364 ti.state = prev_state;
1368 if (ompt_enabled.ompt_callback_mutex_acquired) {
1369 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1370 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1374 KMP_POP_PARTITIONED_TIMER();
1376 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1377 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1381#if KMP_USE_DYNAMIC_LOCK
1384static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
1386#define KMP_TSX_LOCK(seq) lockseq_##seq
1388#define KMP_TSX_LOCK(seq) __kmp_user_lock_seq
1391#if KMP_ARCH_X86 || KMP_ARCH_X86_64
1392#define KMP_CPUINFO_RTM (__kmp_cpuinfo.flags.rtm)
1394#define KMP_CPUINFO_RTM 0
1398 if (hint & kmp_lock_hint_hle)
1399 return KMP_TSX_LOCK(hle);
1400 if (hint & kmp_lock_hint_rtm)
1401 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_queuing) : __kmp_user_lock_seq;
1402 if (hint & kmp_lock_hint_adaptive)
1403 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
1406 if ((hint & omp_lock_hint_contended) && (hint & omp_lock_hint_uncontended))
1407 return __kmp_user_lock_seq;
1408 if ((hint & omp_lock_hint_speculative) &&
1409 (hint & omp_lock_hint_nonspeculative))
1410 return __kmp_user_lock_seq;
1413 if (hint & omp_lock_hint_contended)
1414 return lockseq_queuing;
1417 if ((hint & omp_lock_hint_uncontended) && !(hint & omp_lock_hint_speculative))
1421 if (hint & omp_lock_hint_speculative)
1422 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_spin) : __kmp_user_lock_seq;
1424 return __kmp_user_lock_seq;
1427#if OMPT_SUPPORT && OMPT_OPTIONAL
1428#if KMP_USE_DYNAMIC_LOCK
1429static kmp_mutex_impl_t
1430__ompt_get_mutex_impl_type(
void *user_lock, kmp_indirect_lock_t *ilock = 0) {
1432 switch (KMP_EXTRACT_D_TAG(user_lock)) {
1437 return kmp_mutex_impl_queuing;
1440 return kmp_mutex_impl_spin;
1443 case locktag_rtm_spin:
1444 return kmp_mutex_impl_speculative;
1447 return kmp_mutex_impl_none;
1449 ilock = KMP_LOOKUP_I_LOCK(user_lock);
1452 switch (ilock->type) {
1454 case locktag_adaptive:
1455 case locktag_rtm_queuing:
1456 return kmp_mutex_impl_speculative;
1458 case locktag_nested_tas:
1459 return kmp_mutex_impl_spin;
1461 case locktag_nested_futex:
1463 case locktag_ticket:
1464 case locktag_queuing:
1466 case locktag_nested_ticket:
1467 case locktag_nested_queuing:
1468 case locktag_nested_drdpa:
1469 return kmp_mutex_impl_queuing;
1471 return kmp_mutex_impl_none;
1476static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
1477 switch (__kmp_user_lock_kind) {
1479 return kmp_mutex_impl_spin;
1486 return kmp_mutex_impl_queuing;
1489 case lk_rtm_queuing:
1492 return kmp_mutex_impl_speculative;
1495 return kmp_mutex_impl_none;
1514void __kmpc_critical_with_hint(
ident_t *loc, kmp_int32 global_tid,
1515 kmp_critical_name *crit, uint32_t hint) {
1517 kmp_user_lock_p lck;
1518#if OMPT_SUPPORT && OMPT_OPTIONAL
1519 ompt_state_t prev_state = ompt_state_undefined;
1520 ompt_thread_info_t ti;
1522 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(global_tid);
1524 codeptr = OMPT_GET_RETURN_ADDRESS(0);
1527 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1528 __kmp_assert_valid_gtid(global_tid);
1530 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
1532 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1533 kmp_dyna_lockseq_t lockseq = __kmp_map_hint_to_lock(hint);
1535 if (KMP_IS_D_LOCK(lockseq)) {
1536 KMP_COMPARE_AND_STORE_ACQ32(
1537 (
volatile kmp_int32 *)&((kmp_base_tas_lock_t *)crit)->poll, 0,
1538 KMP_GET_D_TAG(lockseq));
1540 __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lockseq));
1546 if (KMP_EXTRACT_D_TAG(lk) != 0) {
1547 lck = (kmp_user_lock_p)lk;
1548 if (__kmp_env_consistency_check) {
1549 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1550 __kmp_map_hint_to_lock(hint));
1553 __kmp_itt_critical_acquiring(lck);
1555#if OMPT_SUPPORT && OMPT_OPTIONAL
1556 if (ompt_enabled.enabled) {
1557 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1559 prev_state = ti.state;
1560 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1561 ti.state = ompt_state_wait_critical;
1564 if (ompt_enabled.ompt_callback_mutex_acquire) {
1565 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1566 ompt_mutex_critical, (
unsigned int)hint,
1567 __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)(uintptr_t)lck,
1572#if KMP_USE_INLINED_TAS
1573 if (lockseq == lockseq_tas && !__kmp_env_consistency_check) {
1574 KMP_ACQUIRE_TAS_LOCK(lck, global_tid);
1576#elif KMP_USE_INLINED_FUTEX
1577 if (lockseq == lockseq_futex && !__kmp_env_consistency_check) {
1578 KMP_ACQUIRE_FUTEX_LOCK(lck, global_tid);
1582 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
1585 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
1587 if (__kmp_env_consistency_check) {
1588 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1589 __kmp_map_hint_to_lock(hint));
1592 __kmp_itt_critical_acquiring(lck);
1594#if OMPT_SUPPORT && OMPT_OPTIONAL
1595 if (ompt_enabled.enabled) {
1596 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1598 prev_state = ti.state;
1599 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1600 ti.state = ompt_state_wait_critical;
1603 if (ompt_enabled.ompt_callback_mutex_acquire) {
1604 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1605 ompt_mutex_critical, (
unsigned int)hint,
1606 __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)(uintptr_t)lck,
1611 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
1613 KMP_POP_PARTITIONED_TIMER();
1616 __kmp_itt_critical_acquired(lck);
1618#if OMPT_SUPPORT && OMPT_OPTIONAL
1619 if (ompt_enabled.enabled) {
1621 ti.state = prev_state;
1625 if (ompt_enabled.ompt_callback_mutex_acquired) {
1626 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1627 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
1632 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1633 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1648 kmp_critical_name *crit) {
1649 kmp_user_lock_p lck;
1651 KC_TRACE(10, (
"__kmpc_end_critical: called T#%d\n", global_tid));
1653#if KMP_USE_DYNAMIC_LOCK
1654 int locktag = KMP_EXTRACT_D_TAG(crit);
1656 lck = (kmp_user_lock_p)crit;
1657 KMP_ASSERT(lck != NULL);
1658 if (__kmp_env_consistency_check) {
1659 __kmp_pop_sync(global_tid, ct_critical, loc);
1662 __kmp_itt_critical_releasing(lck);
1664#if KMP_USE_INLINED_TAS
1665 if (locktag == locktag_tas && !__kmp_env_consistency_check) {
1666 KMP_RELEASE_TAS_LOCK(lck, global_tid);
1668#elif KMP_USE_INLINED_FUTEX
1669 if (locktag == locktag_futex && !__kmp_env_consistency_check) {
1670 KMP_RELEASE_FUTEX_LOCK(lck, global_tid);
1674 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
1677 kmp_indirect_lock_t *ilk =
1678 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
1679 KMP_ASSERT(ilk != NULL);
1681 if (__kmp_env_consistency_check) {
1682 __kmp_pop_sync(global_tid, ct_critical, loc);
1685 __kmp_itt_critical_releasing(lck);
1687 KMP_I_LOCK_FUNC(ilk, unset)(lck, global_tid);
1692 if ((__kmp_user_lock_kind == lk_tas) &&
1693 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1694 lck = (kmp_user_lock_p)crit;
1697 else if ((__kmp_user_lock_kind == lk_futex) &&
1698 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1699 lck = (kmp_user_lock_p)crit;
1703 lck = (kmp_user_lock_p)TCR_PTR(*((kmp_user_lock_p *)crit));
1706 KMP_ASSERT(lck != NULL);
1708 if (__kmp_env_consistency_check)
1709 __kmp_pop_sync(global_tid, ct_critical, loc);
1712 __kmp_itt_critical_releasing(lck);
1716 __kmp_release_user_lock_with_checks(lck, global_tid);
1720#if OMPT_SUPPORT && OMPT_OPTIONAL
1723 OMPT_STORE_RETURN_ADDRESS(global_tid);
1724 if (ompt_enabled.ompt_callback_mutex_released) {
1725 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1726 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck,
1727 OMPT_LOAD_RETURN_ADDRESS(0));
1731 KMP_POP_PARTITIONED_TIMER();
1732 KA_TRACE(15, (
"__kmpc_end_critical: done T#%d\n", global_tid));
1746 KC_TRACE(10, (
"__kmpc_barrier_master: called T#%d\n", global_tid));
1747 __kmp_assert_valid_gtid(global_tid);
1749 if (!TCR_4(__kmp_init_parallel))
1750 __kmp_parallel_initialize();
1752 __kmp_resume_if_soft_paused();
1754 if (__kmp_env_consistency_check)
1755 __kmp_check_barrier(global_tid, ct_barrier, loc);
1758 ompt_frame_t *ompt_frame;
1759 if (ompt_enabled.enabled) {
1760 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1761 if (ompt_frame->enter_frame.ptr == NULL)
1762 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1764 OMPT_STORE_RETURN_ADDRESS(global_tid);
1767 __kmp_threads[global_tid]->th.th_ident = loc;
1769 status = __kmp_barrier(bs_plain_barrier, global_tid, TRUE, 0, NULL, NULL);
1770#if OMPT_SUPPORT && OMPT_OPTIONAL
1771 if (ompt_enabled.enabled) {
1772 ompt_frame->enter_frame = ompt_data_none;
1776 return (status != 0) ? 0 : 1;
1789 KC_TRACE(10, (
"__kmpc_end_barrier_master: called T#%d\n", global_tid));
1790 __kmp_assert_valid_gtid(global_tid);
1791 __kmp_end_split_barrier(bs_plain_barrier, global_tid);
1806 KC_TRACE(10, (
"__kmpc_barrier_master_nowait: called T#%d\n", global_tid));
1807 __kmp_assert_valid_gtid(global_tid);
1809 if (!TCR_4(__kmp_init_parallel))
1810 __kmp_parallel_initialize();
1812 __kmp_resume_if_soft_paused();
1814 if (__kmp_env_consistency_check) {
1816 KMP_WARNING(ConstructIdentInvalid);
1818 __kmp_check_barrier(global_tid, ct_barrier, loc);
1822 ompt_frame_t *ompt_frame;
1823 if (ompt_enabled.enabled) {
1824 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1825 if (ompt_frame->enter_frame.ptr == NULL)
1826 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1828 OMPT_STORE_RETURN_ADDRESS(global_tid);
1831 __kmp_threads[global_tid]->th.th_ident = loc;
1833 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
1834#if OMPT_SUPPORT && OMPT_OPTIONAL
1835 if (ompt_enabled.enabled) {
1836 ompt_frame->enter_frame = ompt_data_none;
1842 if (__kmp_env_consistency_check) {
1848 __kmp_pop_sync(global_tid, ct_master, loc);
1868 __kmp_assert_valid_gtid(global_tid);
1869 kmp_int32 rc = __kmp_enter_single(global_tid, loc, TRUE);
1874 KMP_PUSH_PARTITIONED_TIMER(OMP_single);
1877#if OMPT_SUPPORT && OMPT_OPTIONAL
1878 kmp_info_t *this_thr = __kmp_threads[global_tid];
1879 kmp_team_t *team = this_thr->th.th_team;
1880 int tid = __kmp_tid_from_gtid(global_tid);
1882 if (ompt_enabled.enabled) {
1884 if (ompt_enabled.ompt_callback_work) {
1885 ompt_callbacks.ompt_callback(ompt_callback_work)(
1886 ompt_work_single_executor, ompt_scope_begin,
1887 &(team->t.ompt_team_info.parallel_data),
1888 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1889 1, OMPT_GET_RETURN_ADDRESS(0));
1892 if (ompt_enabled.ompt_callback_work) {
1893 ompt_callbacks.ompt_callback(ompt_callback_work)(
1894 ompt_work_single_other, ompt_scope_begin,
1895 &(team->t.ompt_team_info.parallel_data),
1896 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1897 1, OMPT_GET_RETURN_ADDRESS(0));
1898 ompt_callbacks.ompt_callback(ompt_callback_work)(
1899 ompt_work_single_other, ompt_scope_end,
1900 &(team->t.ompt_team_info.parallel_data),
1901 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1902 1, OMPT_GET_RETURN_ADDRESS(0));
1921 __kmp_assert_valid_gtid(global_tid);
1922 __kmp_exit_single(global_tid);
1923 KMP_POP_PARTITIONED_TIMER();
1925#if OMPT_SUPPORT && OMPT_OPTIONAL
1926 kmp_info_t *this_thr = __kmp_threads[global_tid];
1927 kmp_team_t *team = this_thr->th.th_team;
1928 int tid = __kmp_tid_from_gtid(global_tid);
1930 if (ompt_enabled.ompt_callback_work) {
1931 ompt_callbacks.ompt_callback(ompt_callback_work)(
1932 ompt_work_single_executor, ompt_scope_end,
1933 &(team->t.ompt_team_info.parallel_data),
1934 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
1935 OMPT_GET_RETURN_ADDRESS(0));
1948 KMP_POP_PARTITIONED_TIMER();
1949 KE_TRACE(10, (
"__kmpc_for_static_fini called T#%d\n", global_tid));
1951#if OMPT_SUPPORT && OMPT_OPTIONAL
1952 if (ompt_enabled.ompt_callback_work) {
1953 ompt_work_t ompt_work_type = ompt_work_loop;
1954 ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
1955 ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
1959 ompt_work_type = ompt_work_loop;
1961 ompt_work_type = ompt_work_sections;
1963 ompt_work_type = ompt_work_distribute;
1968 KMP_DEBUG_ASSERT(ompt_work_type);
1970 ompt_callbacks.ompt_callback(ompt_callback_work)(
1971 ompt_work_type, ompt_scope_end, &(team_info->parallel_data),
1972 &(task_info->task_data), 0, OMPT_GET_RETURN_ADDRESS(0));
1975 if (__kmp_env_consistency_check)
1976 __kmp_pop_workshare(global_tid, ct_pdo, loc);
1982void ompc_set_num_threads(
int arg) {
1984 __kmp_set_num_threads(arg, __kmp_entry_gtid());
1987void ompc_set_dynamic(
int flag) {
1991 thread = __kmp_entry_thread();
1993 __kmp_save_internal_controls(thread);
1995 set__dynamic(thread, flag ?
true : false);
1998void ompc_set_nested(
int flag) {
2002 thread = __kmp_entry_thread();
2004 __kmp_save_internal_controls(thread);
2006 set__max_active_levels(thread, flag ? __kmp_dflt_max_active_levels : 1);
2009void ompc_set_max_active_levels(
int max_active_levels) {
2014 __kmp_set_max_active_levels(__kmp_entry_gtid(), max_active_levels);
2017void ompc_set_schedule(omp_sched_t kind,
int modifier) {
2019 __kmp_set_schedule(__kmp_entry_gtid(), (kmp_sched_t)kind, modifier);
2022int ompc_get_ancestor_thread_num(
int level) {
2023 return __kmp_get_ancestor_thread_num(__kmp_entry_gtid(), level);
2026int ompc_get_team_size(
int level) {
2027 return __kmp_get_team_size(__kmp_entry_gtid(), level);
2031void KMP_EXPAND_NAME(ompc_set_affinity_format)(
char const *format) {
2032 if (!__kmp_init_serial) {
2033 __kmp_serial_initialize();
2035 __kmp_strncpy_truncate(__kmp_affinity_format, KMP_AFFINITY_FORMAT_SIZE,
2036 format, KMP_STRLEN(format) + 1);
2039size_t KMP_EXPAND_NAME(ompc_get_affinity_format)(
char *buffer,
size_t size) {
2041 if (!__kmp_init_serial) {
2042 __kmp_serial_initialize();
2044 format_size = KMP_STRLEN(__kmp_affinity_format);
2045 if (buffer && size) {
2046 __kmp_strncpy_truncate(buffer, size, __kmp_affinity_format,
2052void KMP_EXPAND_NAME(ompc_display_affinity)(
char const *format) {
2054 if (!TCR_4(__kmp_init_middle)) {
2055 __kmp_middle_initialize();
2057 __kmp_assign_root_init_mask();
2058 gtid = __kmp_get_gtid();
2059#if KMP_AFFINITY_SUPPORTED
2060 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 &&
2061 __kmp_affinity.flags.reset) {
2062 __kmp_reset_root_init_mask(gtid);
2065 __kmp_aux_display_affinity(gtid, format);
2068size_t KMP_EXPAND_NAME(ompc_capture_affinity)(
char *buffer,
size_t buf_size,
2069 char const *format) {
2071 size_t num_required;
2072 kmp_str_buf_t capture_buf;
2073 if (!TCR_4(__kmp_init_middle)) {
2074 __kmp_middle_initialize();
2076 __kmp_assign_root_init_mask();
2077 gtid = __kmp_get_gtid();
2078#if KMP_AFFINITY_SUPPORTED
2079 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 &&
2080 __kmp_affinity.flags.reset) {
2081 __kmp_reset_root_init_mask(gtid);
2084 __kmp_str_buf_init(&capture_buf);
2085 num_required = __kmp_aux_capture_affinity(gtid, format, &capture_buf);
2086 if (buffer && buf_size) {
2087 __kmp_strncpy_truncate(buffer, buf_size, capture_buf.str,
2088 capture_buf.used + 1);
2090 __kmp_str_buf_free(&capture_buf);
2091 return num_required;
2094void kmpc_set_stacksize(
int arg) {
2096 __kmp_aux_set_stacksize(arg);
2099void kmpc_set_stacksize_s(
size_t arg) {
2101 __kmp_aux_set_stacksize(arg);
2104void kmpc_set_blocktime(
int arg) {
2105 int gtid, tid, bt = arg;
2108 gtid = __kmp_entry_gtid();
2109 tid = __kmp_tid_from_gtid(gtid);
2110 thread = __kmp_thread_from_gtid(gtid);
2112 __kmp_aux_convert_blocktime(&bt);
2113 __kmp_aux_set_blocktime(bt, thread, tid);
2116void kmpc_set_library(
int arg) {
2118 __kmp_user_set_library((
enum library_type)arg);
2121void kmpc_set_defaults(
char const *str) {
2123 __kmp_aux_set_defaults(str, KMP_STRLEN(str));
2126void kmpc_set_disp_num_buffers(
int arg) {
2129 if (__kmp_init_serial == FALSE && arg >= KMP_MIN_DISP_NUM_BUFF &&
2130 arg <= KMP_MAX_DISP_NUM_BUFF) {
2131 __kmp_dispatch_num_buffers = arg;
2135int kmpc_set_affinity_mask_proc(
int proc,
void **mask) {
2136#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2139 if (!TCR_4(__kmp_init_middle)) {
2140 __kmp_middle_initialize();
2142 __kmp_assign_root_init_mask();
2143 return __kmp_aux_set_affinity_mask_proc(proc, mask);
2147int kmpc_unset_affinity_mask_proc(
int proc,
void **mask) {
2148#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2151 if (!TCR_4(__kmp_init_middle)) {
2152 __kmp_middle_initialize();
2154 __kmp_assign_root_init_mask();
2155 return __kmp_aux_unset_affinity_mask_proc(proc, mask);
2159int kmpc_get_affinity_mask_proc(
int proc,
void **mask) {
2160#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2163 if (!TCR_4(__kmp_init_middle)) {
2164 __kmp_middle_initialize();
2166 __kmp_assign_root_init_mask();
2167 return __kmp_aux_get_affinity_mask_proc(proc, mask);
2217 void *cpy_data,
void (*cpy_func)(
void *,
void *),
2220 KC_TRACE(10, (
"__kmpc_copyprivate: called T#%d\n", gtid));
2221 __kmp_assert_valid_gtid(gtid);
2225 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2227 if (__kmp_env_consistency_check) {
2229 KMP_WARNING(ConstructIdentInvalid);
2236 *data_ptr = cpy_data;
2239 ompt_frame_t *ompt_frame;
2240 if (ompt_enabled.enabled) {
2241 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2242 if (ompt_frame->enter_frame.ptr == NULL)
2243 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2245 OMPT_STORE_RETURN_ADDRESS(gtid);
2249 __kmp_threads[gtid]->th.th_ident = loc;
2251 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2254 (*cpy_func)(cpy_data, *data_ptr);
2260 OMPT_STORE_RETURN_ADDRESS(gtid);
2263 __kmp_threads[gtid]->th.th_ident = loc;
2266 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2267#if OMPT_SUPPORT && OMPT_OPTIONAL
2268 if (ompt_enabled.enabled) {
2269 ompt_frame->enter_frame = ompt_data_none;
2295 KC_TRACE(10, (
"__kmpc_copyprivate_light: called T#%d\n", gtid));
2299 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2301 if (__kmp_env_consistency_check) {
2303 KMP_WARNING(ConstructIdentInvalid);
2310 *data_ptr = cpy_data;
2313 ompt_frame_t *ompt_frame;
2314 if (ompt_enabled.enabled) {
2315 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2316 if (ompt_frame->enter_frame.ptr == NULL)
2317 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2318 OMPT_STORE_RETURN_ADDRESS(gtid);
2323 __kmp_threads[gtid]->th.th_ident = loc;
2325 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2332#define INIT_LOCK __kmp_init_user_lock_with_checks
2333#define INIT_NESTED_LOCK __kmp_init_nested_user_lock_with_checks
2334#define ACQUIRE_LOCK __kmp_acquire_user_lock_with_checks
2335#define ACQUIRE_LOCK_TIMED __kmp_acquire_user_lock_with_checks_timed
2336#define ACQUIRE_NESTED_LOCK __kmp_acquire_nested_user_lock_with_checks
2337#define ACQUIRE_NESTED_LOCK_TIMED \
2338 __kmp_acquire_nested_user_lock_with_checks_timed
2339#define RELEASE_LOCK __kmp_release_user_lock_with_checks
2340#define RELEASE_NESTED_LOCK __kmp_release_nested_user_lock_with_checks
2341#define TEST_LOCK __kmp_test_user_lock_with_checks
2342#define TEST_NESTED_LOCK __kmp_test_nested_user_lock_with_checks
2343#define DESTROY_LOCK __kmp_destroy_user_lock_with_checks
2344#define DESTROY_NESTED_LOCK __kmp_destroy_nested_user_lock_with_checks
2349#if KMP_USE_DYNAMIC_LOCK
2352static __forceinline
void __kmp_init_lock_with_hint(
ident_t *loc,
void **lock,
2353 kmp_dyna_lockseq_t seq) {
2354 if (KMP_IS_D_LOCK(seq)) {
2355 KMP_INIT_D_LOCK(lock, seq);
2357 __kmp_itt_lock_creating((kmp_user_lock_p)lock, NULL);
2360 KMP_INIT_I_LOCK(lock, seq);
2362 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2363 __kmp_itt_lock_creating(ilk->lock, loc);
2369static __forceinline
void
2370__kmp_init_nest_lock_with_hint(
ident_t *loc,
void **lock,
2371 kmp_dyna_lockseq_t seq) {
2374 if (seq == lockseq_hle || seq == lockseq_rtm_queuing ||
2375 seq == lockseq_rtm_spin || seq == lockseq_adaptive)
2376 seq = __kmp_user_lock_seq;
2380 seq = lockseq_nested_tas;
2384 seq = lockseq_nested_futex;
2387 case lockseq_ticket:
2388 seq = lockseq_nested_ticket;
2390 case lockseq_queuing:
2391 seq = lockseq_nested_queuing;
2394 seq = lockseq_nested_drdpa;
2397 seq = lockseq_nested_queuing;
2399 KMP_INIT_I_LOCK(lock, seq);
2401 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2402 __kmp_itt_lock_creating(ilk->lock, loc);
2407void __kmpc_init_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
void **user_lock,
2409 KMP_DEBUG_ASSERT(__kmp_init_serial);
2410 if (__kmp_env_consistency_check && user_lock == NULL) {
2411 KMP_FATAL(LockIsUninitialized,
"omp_init_lock_with_hint");
2414 __kmp_init_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2416#if OMPT_SUPPORT && OMPT_OPTIONAL
2418 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2420 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2421 if (ompt_enabled.ompt_callback_lock_init) {
2422 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2423 ompt_mutex_lock, (omp_lock_hint_t)hint,
2424 __ompt_get_mutex_impl_type(user_lock),
2425 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2431void __kmpc_init_nest_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
2432 void **user_lock, uintptr_t hint) {
2433 KMP_DEBUG_ASSERT(__kmp_init_serial);
2434 if (__kmp_env_consistency_check && user_lock == NULL) {
2435 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock_with_hint");
2438 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
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_nest_lock, (omp_lock_hint_t)hint,
2448 __ompt_get_mutex_impl_type(user_lock),
2449 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2457void __kmpc_init_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2458#if KMP_USE_DYNAMIC_LOCK
2460 KMP_DEBUG_ASSERT(__kmp_init_serial);
2461 if (__kmp_env_consistency_check && user_lock == NULL) {
2462 KMP_FATAL(LockIsUninitialized,
"omp_init_lock");
2464 __kmp_init_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2466#if OMPT_SUPPORT && OMPT_OPTIONAL
2468 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2470 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2471 if (ompt_enabled.ompt_callback_lock_init) {
2472 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2473 ompt_mutex_lock, omp_lock_hint_none,
2474 __ompt_get_mutex_impl_type(user_lock),
2475 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2481 static char const *
const func =
"omp_init_lock";
2482 kmp_user_lock_p lck;
2483 KMP_DEBUG_ASSERT(__kmp_init_serial);
2485 if (__kmp_env_consistency_check) {
2486 if (user_lock == NULL) {
2487 KMP_FATAL(LockIsUninitialized, func);
2491 KMP_CHECK_USER_LOCK_INIT();
2493 if ((__kmp_user_lock_kind == lk_tas) &&
2494 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2495 lck = (kmp_user_lock_p)user_lock;
2498 else if ((__kmp_user_lock_kind == lk_futex) &&
2499 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2500 lck = (kmp_user_lock_p)user_lock;
2504 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2507 __kmp_set_user_lock_location(lck, loc);
2509#if OMPT_SUPPORT && OMPT_OPTIONAL
2511 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2513 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2514 if (ompt_enabled.ompt_callback_lock_init) {
2515 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2516 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2517 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2522 __kmp_itt_lock_creating(lck);
2529void __kmpc_init_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2530#if KMP_USE_DYNAMIC_LOCK
2532 KMP_DEBUG_ASSERT(__kmp_init_serial);
2533 if (__kmp_env_consistency_check && user_lock == NULL) {
2534 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock");
2536 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2538#if OMPT_SUPPORT && OMPT_OPTIONAL
2540 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2542 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2543 if (ompt_enabled.ompt_callback_lock_init) {
2544 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2545 ompt_mutex_nest_lock, omp_lock_hint_none,
2546 __ompt_get_mutex_impl_type(user_lock),
2547 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2553 static char const *
const func =
"omp_init_nest_lock";
2554 kmp_user_lock_p lck;
2555 KMP_DEBUG_ASSERT(__kmp_init_serial);
2557 if (__kmp_env_consistency_check) {
2558 if (user_lock == NULL) {
2559 KMP_FATAL(LockIsUninitialized, func);
2563 KMP_CHECK_USER_LOCK_INIT();
2565 if ((__kmp_user_lock_kind == lk_tas) &&
2566 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2567 OMP_NEST_LOCK_T_SIZE)) {
2568 lck = (kmp_user_lock_p)user_lock;
2571 else if ((__kmp_user_lock_kind == lk_futex) &&
2572 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2573 OMP_NEST_LOCK_T_SIZE)) {
2574 lck = (kmp_user_lock_p)user_lock;
2578 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2581 INIT_NESTED_LOCK(lck);
2582 __kmp_set_user_lock_location(lck, loc);
2584#if OMPT_SUPPORT && OMPT_OPTIONAL
2586 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2588 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2589 if (ompt_enabled.ompt_callback_lock_init) {
2590 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2591 ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2592 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2597 __kmp_itt_lock_creating(lck);
2603void __kmpc_destroy_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2604#if KMP_USE_DYNAMIC_LOCK
2607 kmp_user_lock_p lck;
2608 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2609 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2611 lck = (kmp_user_lock_p)user_lock;
2613 __kmp_itt_lock_destroyed(lck);
2615#if OMPT_SUPPORT && OMPT_OPTIONAL
2617 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2619 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2620 if (ompt_enabled.ompt_callback_lock_destroy) {
2621 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2622 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2625 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2627 kmp_user_lock_p lck;
2629 if ((__kmp_user_lock_kind == lk_tas) &&
2630 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2631 lck = (kmp_user_lock_p)user_lock;
2634 else if ((__kmp_user_lock_kind == lk_futex) &&
2635 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2636 lck = (kmp_user_lock_p)user_lock;
2640 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_lock");
2643#if OMPT_SUPPORT && OMPT_OPTIONAL
2645 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2647 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2648 if (ompt_enabled.ompt_callback_lock_destroy) {
2649 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2650 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2655 __kmp_itt_lock_destroyed(lck);
2659 if ((__kmp_user_lock_kind == lk_tas) &&
2660 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2664 else if ((__kmp_user_lock_kind == lk_futex) &&
2665 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2670 __kmp_user_lock_free(user_lock, gtid, lck);
2676void __kmpc_destroy_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2677#if KMP_USE_DYNAMIC_LOCK
2680 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(user_lock);
2681 __kmp_itt_lock_destroyed(ilk->lock);
2683#if OMPT_SUPPORT && OMPT_OPTIONAL
2685 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2687 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2688 if (ompt_enabled.ompt_callback_lock_destroy) {
2689 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2690 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2693 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2697 kmp_user_lock_p lck;
2699 if ((__kmp_user_lock_kind == lk_tas) &&
2700 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2701 OMP_NEST_LOCK_T_SIZE)) {
2702 lck = (kmp_user_lock_p)user_lock;
2705 else if ((__kmp_user_lock_kind == lk_futex) &&
2706 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2707 OMP_NEST_LOCK_T_SIZE)) {
2708 lck = (kmp_user_lock_p)user_lock;
2712 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_nest_lock");
2715#if OMPT_SUPPORT && OMPT_OPTIONAL
2717 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2719 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2720 if (ompt_enabled.ompt_callback_lock_destroy) {
2721 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2722 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2727 __kmp_itt_lock_destroyed(lck);
2730 DESTROY_NESTED_LOCK(lck);
2732 if ((__kmp_user_lock_kind == lk_tas) &&
2733 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2734 OMP_NEST_LOCK_T_SIZE)) {
2738 else if ((__kmp_user_lock_kind == lk_futex) &&
2739 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2740 OMP_NEST_LOCK_T_SIZE)) {
2745 __kmp_user_lock_free(user_lock, gtid, lck);
2750void __kmpc_set_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2752#if KMP_USE_DYNAMIC_LOCK
2753 int tag = KMP_EXTRACT_D_TAG(user_lock);
2755 __kmp_itt_lock_acquiring(
2759#if OMPT_SUPPORT && OMPT_OPTIONAL
2761 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2763 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2764 if (ompt_enabled.ompt_callback_mutex_acquire) {
2765 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2766 ompt_mutex_lock, omp_lock_hint_none,
2767 __ompt_get_mutex_impl_type(user_lock),
2768 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2771#if KMP_USE_INLINED_TAS
2772 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2773 KMP_ACQUIRE_TAS_LOCK(user_lock, gtid);
2775#elif KMP_USE_INLINED_FUTEX
2776 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2777 KMP_ACQUIRE_FUTEX_LOCK(user_lock, gtid);
2781 __kmp_direct_set[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2784 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2786#if OMPT_SUPPORT && OMPT_OPTIONAL
2787 if (ompt_enabled.ompt_callback_mutex_acquired) {
2788 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2789 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2795 kmp_user_lock_p lck;
2797 if ((__kmp_user_lock_kind == lk_tas) &&
2798 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2799 lck = (kmp_user_lock_p)user_lock;
2802 else if ((__kmp_user_lock_kind == lk_futex) &&
2803 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2804 lck = (kmp_user_lock_p)user_lock;
2808 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_lock");
2812 __kmp_itt_lock_acquiring(lck);
2814#if OMPT_SUPPORT && OMPT_OPTIONAL
2816 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2818 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2819 if (ompt_enabled.ompt_callback_mutex_acquire) {
2820 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2821 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2822 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2826 ACQUIRE_LOCK(lck, gtid);
2829 __kmp_itt_lock_acquired(lck);
2832#if OMPT_SUPPORT && OMPT_OPTIONAL
2833 if (ompt_enabled.ompt_callback_mutex_acquired) {
2834 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2835 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2842void __kmpc_set_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2843#if KMP_USE_DYNAMIC_LOCK
2846 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
2848#if OMPT_SUPPORT && OMPT_OPTIONAL
2850 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2852 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2853 if (ompt_enabled.enabled) {
2854 if (ompt_enabled.ompt_callback_mutex_acquire) {
2855 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2856 ompt_mutex_nest_lock, omp_lock_hint_none,
2857 __ompt_get_mutex_impl_type(user_lock),
2858 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2862 int acquire_status =
2863 KMP_D_LOCK_FUNC(user_lock, set)((kmp_dyna_lock_t *)user_lock, gtid);
2864 (void)acquire_status;
2866 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2869#if OMPT_SUPPORT && OMPT_OPTIONAL
2870 if (ompt_enabled.enabled) {
2871 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2872 if (ompt_enabled.ompt_callback_mutex_acquired) {
2874 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2875 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
2879 if (ompt_enabled.ompt_callback_nest_lock) {
2881 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2882 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2890 kmp_user_lock_p lck;
2892 if ((__kmp_user_lock_kind == lk_tas) &&
2893 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2894 OMP_NEST_LOCK_T_SIZE)) {
2895 lck = (kmp_user_lock_p)user_lock;
2898 else if ((__kmp_user_lock_kind == lk_futex) &&
2899 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2900 OMP_NEST_LOCK_T_SIZE)) {
2901 lck = (kmp_user_lock_p)user_lock;
2905 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_nest_lock");
2909 __kmp_itt_lock_acquiring(lck);
2911#if OMPT_SUPPORT && OMPT_OPTIONAL
2913 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2915 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2916 if (ompt_enabled.enabled) {
2917 if (ompt_enabled.ompt_callback_mutex_acquire) {
2918 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2919 ompt_mutex_nest_lock, omp_lock_hint_none,
2920 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
2926 ACQUIRE_NESTED_LOCK(lck, gtid, &acquire_status);
2929 __kmp_itt_lock_acquired(lck);
2932#if OMPT_SUPPORT && OMPT_OPTIONAL
2933 if (ompt_enabled.enabled) {
2934 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2935 if (ompt_enabled.ompt_callback_mutex_acquired) {
2937 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2938 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2941 if (ompt_enabled.ompt_callback_nest_lock) {
2943 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2944 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2953void __kmpc_unset_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2954#if KMP_USE_DYNAMIC_LOCK
2956 int tag = KMP_EXTRACT_D_TAG(user_lock);
2958 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2960#if KMP_USE_INLINED_TAS
2961 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2962 KMP_RELEASE_TAS_LOCK(user_lock, gtid);
2964#elif KMP_USE_INLINED_FUTEX
2965 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2966 KMP_RELEASE_FUTEX_LOCK(user_lock, gtid);
2970 __kmp_direct_unset[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2973#if OMPT_SUPPORT && OMPT_OPTIONAL
2975 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2977 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2978 if (ompt_enabled.ompt_callback_mutex_released) {
2979 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2980 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2986 kmp_user_lock_p lck;
2991 if ((__kmp_user_lock_kind == lk_tas) &&
2992 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2993#if KMP_OS_LINUX && \
2994 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
2997 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2999 TCW_4(((kmp_user_lock_p)user_lock)->tas.lk.poll, 0);
3002#if OMPT_SUPPORT && OMPT_OPTIONAL
3004 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3006 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3007 if (ompt_enabled.ompt_callback_mutex_released) {
3008 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3009 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3015 lck = (kmp_user_lock_p)user_lock;
3019 else if ((__kmp_user_lock_kind == lk_futex) &&
3020 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3021 lck = (kmp_user_lock_p)user_lock;
3025 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_lock");
3029 __kmp_itt_lock_releasing(lck);
3032 RELEASE_LOCK(lck, gtid);
3034#if OMPT_SUPPORT && OMPT_OPTIONAL
3036 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3038 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3039 if (ompt_enabled.ompt_callback_mutex_released) {
3040 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3041 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3049void __kmpc_unset_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3050#if KMP_USE_DYNAMIC_LOCK
3053 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3055 int release_status =
3056 KMP_D_LOCK_FUNC(user_lock, unset)((kmp_dyna_lock_t *)user_lock, gtid);
3057 (void)release_status;
3059#if OMPT_SUPPORT && OMPT_OPTIONAL
3061 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3063 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3064 if (ompt_enabled.enabled) {
3065 if (release_status == KMP_LOCK_RELEASED) {
3066 if (ompt_enabled.ompt_callback_mutex_released) {
3068 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3069 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3072 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3074 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3075 ompt_scope_end, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3082 kmp_user_lock_p lck;
3086 if ((__kmp_user_lock_kind == lk_tas) &&
3087 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3088 OMP_NEST_LOCK_T_SIZE)) {
3089#if KMP_OS_LINUX && \
3090 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
3092 kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
3094 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3097#if OMPT_SUPPORT && OMPT_OPTIONAL
3098 int release_status = KMP_LOCK_STILL_HELD;
3101 if (--(tl->lk.depth_locked) == 0) {
3102 TCW_4(tl->lk.poll, 0);
3103#if OMPT_SUPPORT && OMPT_OPTIONAL
3104 release_status = KMP_LOCK_RELEASED;
3109#if OMPT_SUPPORT && OMPT_OPTIONAL
3111 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3113 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3114 if (ompt_enabled.enabled) {
3115 if (release_status == KMP_LOCK_RELEASED) {
3116 if (ompt_enabled.ompt_callback_mutex_released) {
3118 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3119 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3121 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3123 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3124 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3131 lck = (kmp_user_lock_p)user_lock;
3135 else if ((__kmp_user_lock_kind == lk_futex) &&
3136 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3137 OMP_NEST_LOCK_T_SIZE)) {
3138 lck = (kmp_user_lock_p)user_lock;
3142 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_nest_lock");
3146 __kmp_itt_lock_releasing(lck);
3150 release_status = RELEASE_NESTED_LOCK(lck, gtid);
3151#if OMPT_SUPPORT && OMPT_OPTIONAL
3153 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3155 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3156 if (ompt_enabled.enabled) {
3157 if (release_status == KMP_LOCK_RELEASED) {
3158 if (ompt_enabled.ompt_callback_mutex_released) {
3160 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3161 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3163 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3165 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3166 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3175int __kmpc_test_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3178#if KMP_USE_DYNAMIC_LOCK
3180 int tag = KMP_EXTRACT_D_TAG(user_lock);
3182 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3184#if OMPT_SUPPORT && OMPT_OPTIONAL
3186 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3188 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3189 if (ompt_enabled.ompt_callback_mutex_acquire) {
3190 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3191 ompt_mutex_test_lock, omp_lock_hint_none,
3192 __ompt_get_mutex_impl_type(user_lock),
3193 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3196#if KMP_USE_INLINED_TAS
3197 if (tag == locktag_tas && !__kmp_env_consistency_check) {
3198 KMP_TEST_TAS_LOCK(user_lock, gtid, rc);
3200#elif KMP_USE_INLINED_FUTEX
3201 if (tag == locktag_futex && !__kmp_env_consistency_check) {
3202 KMP_TEST_FUTEX_LOCK(user_lock, gtid, rc);
3206 rc = __kmp_direct_test[tag]((kmp_dyna_lock_t *)user_lock, gtid);
3210 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3212#if OMPT_SUPPORT && OMPT_OPTIONAL
3213 if (ompt_enabled.ompt_callback_mutex_acquired) {
3214 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3215 ompt_mutex_test_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3221 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3228 kmp_user_lock_p lck;
3231 if ((__kmp_user_lock_kind == lk_tas) &&
3232 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
3233 lck = (kmp_user_lock_p)user_lock;
3236 else if ((__kmp_user_lock_kind == lk_futex) &&
3237 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3238 lck = (kmp_user_lock_p)user_lock;
3242 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_lock");
3246 __kmp_itt_lock_acquiring(lck);
3248#if OMPT_SUPPORT && OMPT_OPTIONAL
3250 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3252 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3253 if (ompt_enabled.ompt_callback_mutex_acquire) {
3254 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3255 ompt_mutex_test_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
3256 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3260 rc = TEST_LOCK(lck, gtid);
3263 __kmp_itt_lock_acquired(lck);
3265 __kmp_itt_lock_cancelled(lck);
3268#if OMPT_SUPPORT && OMPT_OPTIONAL
3269 if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
3270 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3271 ompt_mutex_test_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3275 return (rc ? FTN_TRUE : FTN_FALSE);
3283int __kmpc_test_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3284#if KMP_USE_DYNAMIC_LOCK
3287 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3289#if OMPT_SUPPORT && OMPT_OPTIONAL
3291 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3293 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3294 if (ompt_enabled.ompt_callback_mutex_acquire) {
3295 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3296 ompt_mutex_test_nest_lock, omp_lock_hint_none,
3297 __ompt_get_mutex_impl_type(user_lock),
3298 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3301 rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
3304 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3306 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3309#if OMPT_SUPPORT && OMPT_OPTIONAL
3310 if (ompt_enabled.enabled && rc) {
3312 if (ompt_enabled.ompt_callback_mutex_acquired) {
3314 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3315 ompt_mutex_test_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3319 if (ompt_enabled.ompt_callback_nest_lock) {
3321 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3322 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3331 kmp_user_lock_p lck;
3334 if ((__kmp_user_lock_kind == lk_tas) &&
3335 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3336 OMP_NEST_LOCK_T_SIZE)) {
3337 lck = (kmp_user_lock_p)user_lock;
3340 else if ((__kmp_user_lock_kind == lk_futex) &&
3341 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3342 OMP_NEST_LOCK_T_SIZE)) {
3343 lck = (kmp_user_lock_p)user_lock;
3347 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_nest_lock");
3351 __kmp_itt_lock_acquiring(lck);
3354#if OMPT_SUPPORT && OMPT_OPTIONAL
3356 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3358 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3359 if (ompt_enabled.enabled) &&
3360 ompt_enabled.ompt_callback_mutex_acquire) {
3361 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3362 ompt_mutex_test_nest_lock, omp_lock_hint_none,
3363 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
3368 rc = TEST_NESTED_LOCK(lck, gtid);
3371 __kmp_itt_lock_acquired(lck);
3373 __kmp_itt_lock_cancelled(lck);
3376#if OMPT_SUPPORT && OMPT_OPTIONAL
3377 if (ompt_enabled.enabled && rc) {
3379 if (ompt_enabled.ompt_callback_mutex_acquired) {
3381 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3382 ompt_mutex_test_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3385 if (ompt_enabled.ompt_callback_nest_lock) {
3387 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3388 ompt_mutex_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3407#define __KMP_SET_REDUCTION_METHOD(gtid, rmethod) \
3408 ((__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) = (rmethod))
3410#define __KMP_GET_REDUCTION_METHOD(gtid) \
3411 (__kmp_threads[(gtid)]->th.th_local.packed_reduction_method)
3417static __forceinline
void
3418__kmp_enter_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3419 kmp_critical_name *crit) {
3425 kmp_user_lock_p lck;
3427#if KMP_USE_DYNAMIC_LOCK
3429 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
3432 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3433 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
3434 KMP_GET_D_TAG(__kmp_user_lock_seq));
3436 __kmp_init_indirect_csptr(crit, loc, global_tid,
3437 KMP_GET_I_TAG(__kmp_user_lock_seq));
3443 if (KMP_EXTRACT_D_TAG(lk) != 0) {
3444 lck = (kmp_user_lock_p)lk;
3445 KMP_DEBUG_ASSERT(lck != NULL);
3446 if (__kmp_env_consistency_check) {
3447 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3449 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
3451 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
3453 KMP_DEBUG_ASSERT(lck != NULL);
3454 if (__kmp_env_consistency_check) {
3455 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3457 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
3465 if (__kmp_base_user_lock_size <= INTEL_CRITICAL_SIZE) {
3466 lck = (kmp_user_lock_p)crit;
3468 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
3470 KMP_DEBUG_ASSERT(lck != NULL);
3472 if (__kmp_env_consistency_check)
3473 __kmp_push_sync(global_tid, ct_critical, loc, lck);
3475 __kmp_acquire_user_lock_with_checks(lck, global_tid);
3481static __forceinline
void
3482__kmp_end_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3483 kmp_critical_name *crit) {
3485 kmp_user_lock_p lck;
3487#if KMP_USE_DYNAMIC_LOCK
3489 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3490 lck = (kmp_user_lock_p)crit;
3491 if (__kmp_env_consistency_check)
3492 __kmp_pop_sync(global_tid, ct_critical, loc);
3493 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
3495 kmp_indirect_lock_t *ilk =
3496 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
3497 if (__kmp_env_consistency_check)
3498 __kmp_pop_sync(global_tid, ct_critical, loc);
3499 KMP_I_LOCK_FUNC(ilk, unset)(ilk->lock, global_tid);
3507 if (__kmp_base_user_lock_size > 32) {
3508 lck = *((kmp_user_lock_p *)crit);
3509 KMP_ASSERT(lck != NULL);
3511 lck = (kmp_user_lock_p)crit;
3514 if (__kmp_env_consistency_check)
3515 __kmp_pop_sync(global_tid, ct_critical, loc);
3517 __kmp_release_user_lock_with_checks(lck, global_tid);
3522static __forceinline
int
3523__kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
3528 if (th->th.th_teams_microtask) {
3529 *team_p = team = th->th.th_team;
3530 if (team->t.t_level == th->th.th_teams_level) {
3532 KMP_DEBUG_ASSERT(!th->th.th_info.ds.ds_tid);
3534 th->th.th_info.ds.ds_tid = team->t.t_master_tid;
3535 th->th.th_team = team->t.t_parent;
3536 th->th.th_team_nproc = th->th.th_team->t.t_nproc;
3537 th->th.th_task_team = th->th.th_team->t.t_task_team[0];
3538 *task_state = th->th.th_task_state;
3539 th->th.th_task_state = 0;
3547static __forceinline
void
3548__kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team,
int task_state) {
3550 th->th.th_info.ds.ds_tid = 0;
3551 th->th.th_team = team;
3552 th->th.th_team_nproc = team->t.t_nproc;
3553 th->th.th_task_team = team->t.t_task_team[task_state];
3554 __kmp_type_convert(task_state, &(th->th.th_task_state));
3575 size_t reduce_size,
void *reduce_data,
3576 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3577 kmp_critical_name *lck) {
3581 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3584 int teams_swapped = 0, task_state;
3585 KA_TRACE(10, (
"__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
3586 __kmp_assert_valid_gtid(global_tid);
3594 if (!TCR_4(__kmp_init_parallel))
3595 __kmp_parallel_initialize();
3597 __kmp_resume_if_soft_paused();
3600#if KMP_USE_DYNAMIC_LOCK
3601 if (__kmp_env_consistency_check)
3602 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3604 if (__kmp_env_consistency_check)
3605 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3608 th = __kmp_thread_from_gtid(global_tid);
3609 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3627 packed_reduction_method = __kmp_determine_reduction_method(
3628 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3629 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3631 OMPT_REDUCTION_DECL(th, global_tid);
3632 if (packed_reduction_method == critical_reduce_block) {
3634 OMPT_REDUCTION_BEGIN;
3636 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3639 }
else if (packed_reduction_method == empty_reduce_block) {
3641 OMPT_REDUCTION_BEGIN;
3647 }
else if (packed_reduction_method == atomic_reduce_block) {
3657 if (__kmp_env_consistency_check)
3658 __kmp_pop_sync(global_tid, ct_reduce, loc);
3660 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3661 tree_reduce_block)) {
3681 ompt_frame_t *ompt_frame;
3682 if (ompt_enabled.enabled) {
3683 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3684 if (ompt_frame->enter_frame.ptr == NULL)
3685 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3687 OMPT_STORE_RETURN_ADDRESS(global_tid);
3690 __kmp_threads[global_tid]->th.th_ident = loc;
3693 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3694 global_tid, FALSE, reduce_size, reduce_data, reduce_func);
3695 retval = (retval != 0) ? (0) : (1);
3696#if OMPT_SUPPORT && OMPT_OPTIONAL
3697 if (ompt_enabled.enabled) {
3698 ompt_frame->enter_frame = ompt_data_none;
3704 if (__kmp_env_consistency_check) {
3706 __kmp_pop_sync(global_tid, ct_reduce, loc);
3715 if (teams_swapped) {
3716 __kmp_restore_swapped_teams(th, team, task_state);
3720 (
"__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
3721 global_tid, packed_reduction_method, retval));
3735 kmp_critical_name *lck) {
3737 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3739 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() enter: called T#%d\n", global_tid));
3740 __kmp_assert_valid_gtid(global_tid);
3742 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3744 OMPT_REDUCTION_DECL(__kmp_thread_from_gtid(global_tid), global_tid);
3746 if (packed_reduction_method == critical_reduce_block) {
3748 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3751 }
else if (packed_reduction_method == empty_reduce_block) {
3758 }
else if (packed_reduction_method == atomic_reduce_block) {
3765 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3766 tree_reduce_block)) {
3777 if (__kmp_env_consistency_check)
3778 __kmp_pop_sync(global_tid, ct_reduce, loc);
3780 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() exit: called T#%d: method %08x\n",
3781 global_tid, packed_reduction_method));
3804 size_t reduce_size,
void *reduce_data,
3805 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3806 kmp_critical_name *lck) {
3809 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3812 int teams_swapped = 0, task_state;
3814 KA_TRACE(10, (
"__kmpc_reduce() enter: called T#%d\n", global_tid));
3815 __kmp_assert_valid_gtid(global_tid);
3823 if (!TCR_4(__kmp_init_parallel))
3824 __kmp_parallel_initialize();
3826 __kmp_resume_if_soft_paused();
3829#if KMP_USE_DYNAMIC_LOCK
3830 if (__kmp_env_consistency_check)
3831 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3833 if (__kmp_env_consistency_check)
3834 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3837 th = __kmp_thread_from_gtid(global_tid);
3838 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3840 packed_reduction_method = __kmp_determine_reduction_method(
3841 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3842 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3844 OMPT_REDUCTION_DECL(th, global_tid);
3846 if (packed_reduction_method == critical_reduce_block) {
3848 OMPT_REDUCTION_BEGIN;
3849 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3852 }
else if (packed_reduction_method == empty_reduce_block) {
3854 OMPT_REDUCTION_BEGIN;
3859 }
else if (packed_reduction_method == atomic_reduce_block) {
3863 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3864 tree_reduce_block)) {
3870 ompt_frame_t *ompt_frame;
3871 if (ompt_enabled.enabled) {
3872 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3873 if (ompt_frame->enter_frame.ptr == NULL)
3874 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3876 OMPT_STORE_RETURN_ADDRESS(global_tid);
3879 __kmp_threads[global_tid]->th.th_ident =
3883 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3884 global_tid, TRUE, reduce_size, reduce_data, reduce_func);
3885 retval = (retval != 0) ? (0) : (1);
3886#if OMPT_SUPPORT && OMPT_OPTIONAL
3887 if (ompt_enabled.enabled) {
3888 ompt_frame->enter_frame = ompt_data_none;
3894 if (__kmp_env_consistency_check) {
3896 __kmp_pop_sync(global_tid, ct_reduce, loc);
3905 if (teams_swapped) {
3906 __kmp_restore_swapped_teams(th, team, task_state);
3910 (
"__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
3911 global_tid, packed_reduction_method, retval));
3926 kmp_critical_name *lck) {
3928 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3931 int teams_swapped = 0, task_state;
3933 KA_TRACE(10, (
"__kmpc_end_reduce() enter: called T#%d\n", global_tid));
3934 __kmp_assert_valid_gtid(global_tid);
3936 th = __kmp_thread_from_gtid(global_tid);
3937 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3939 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3943 OMPT_REDUCTION_DECL(th, global_tid);
3945 if (packed_reduction_method == critical_reduce_block) {
3946 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3952 ompt_frame_t *ompt_frame;
3953 if (ompt_enabled.enabled) {
3954 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3955 if (ompt_frame->enter_frame.ptr == NULL)
3956 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3958 OMPT_STORE_RETURN_ADDRESS(global_tid);
3961 __kmp_threads[global_tid]->th.th_ident = loc;
3963 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3964#if OMPT_SUPPORT && OMPT_OPTIONAL
3965 if (ompt_enabled.enabled) {
3966 ompt_frame->enter_frame = ompt_data_none;
3970 }
else if (packed_reduction_method == empty_reduce_block) {
3978 ompt_frame_t *ompt_frame;
3979 if (ompt_enabled.enabled) {
3980 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3981 if (ompt_frame->enter_frame.ptr == NULL)
3982 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3984 OMPT_STORE_RETURN_ADDRESS(global_tid);
3987 __kmp_threads[global_tid]->th.th_ident = loc;
3989 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3990#if OMPT_SUPPORT && OMPT_OPTIONAL
3991 if (ompt_enabled.enabled) {
3992 ompt_frame->enter_frame = ompt_data_none;
3996 }
else if (packed_reduction_method == atomic_reduce_block) {
3999 ompt_frame_t *ompt_frame;
4000 if (ompt_enabled.enabled) {
4001 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
4002 if (ompt_frame->enter_frame.ptr == NULL)
4003 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
4005 OMPT_STORE_RETURN_ADDRESS(global_tid);
4009 __kmp_threads[global_tid]->th.th_ident = loc;
4011 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
4012#if OMPT_SUPPORT && OMPT_OPTIONAL
4013 if (ompt_enabled.enabled) {
4014 ompt_frame->enter_frame = ompt_data_none;
4018 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
4019 tree_reduce_block)) {
4022 __kmp_end_split_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
4030 if (teams_swapped) {
4031 __kmp_restore_swapped_teams(th, team, task_state);
4034 if (__kmp_env_consistency_check)
4035 __kmp_pop_sync(global_tid, ct_reduce, loc);
4037 KA_TRACE(10, (
"__kmpc_end_reduce() exit: called T#%d: method %08x\n",
4038 global_tid, packed_reduction_method));
4043#undef __KMP_GET_REDUCTION_METHOD
4044#undef __KMP_SET_REDUCTION_METHOD
4048kmp_uint64 __kmpc_get_taskid() {
4053 gtid = __kmp_get_gtid();
4057 thread = __kmp_thread_from_gtid(gtid);
4058 return thread->th.th_current_task->td_task_id;
4062kmp_uint64 __kmpc_get_parent_taskid() {
4066 kmp_taskdata_t *parent_task;
4068 gtid = __kmp_get_gtid();
4072 thread = __kmp_thread_from_gtid(gtid);
4073 parent_task = thread->th.th_current_task->td_parent;
4074 return (parent_task == NULL ? 0 : parent_task->td_task_id);
4090 const struct kmp_dim *dims) {
4091 __kmp_assert_valid_gtid(gtid);
4093 kmp_int64 last, trace_count;
4094 kmp_info_t *th = __kmp_threads[gtid];
4095 kmp_team_t *team = th->th.th_team;
4097 kmp_disp_t *pr_buf = th->th.th_dispatch;
4098 dispatch_shared_info_t *sh_buf;
4102 (
"__kmpc_doacross_init() enter: called T#%d, num dims %d, active %d\n",
4103 gtid, num_dims, !team->t.t_serialized));
4104 KMP_DEBUG_ASSERT(dims != NULL);
4105 KMP_DEBUG_ASSERT(num_dims > 0);
4107 if (team->t.t_serialized) {
4108 KA_TRACE(20, (
"__kmpc_doacross_init() exit: serialized team\n"));
4111 KMP_DEBUG_ASSERT(team->t.t_nproc > 1);
4112 idx = pr_buf->th_doacross_buf_idx++;
4114 sh_buf = &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4117 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info == NULL);
4118 pr_buf->th_doacross_info = (kmp_int64 *)__kmp_thread_malloc(
4119 th,
sizeof(kmp_int64) * (4 * num_dims + 1));
4120 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4121 pr_buf->th_doacross_info[0] =
4122 (kmp_int64)num_dims;
4125 pr_buf->th_doacross_info[1] = (kmp_int64)&sh_buf->doacross_num_done;
4126 pr_buf->th_doacross_info[2] = dims[0].lo;
4127 pr_buf->th_doacross_info[3] = dims[0].up;
4128 pr_buf->th_doacross_info[4] = dims[0].st;
4130 for (j = 1; j < num_dims; ++j) {
4133 if (dims[j].st == 1) {
4135 range_length = dims[j].up - dims[j].lo + 1;
4137 if (dims[j].st > 0) {
4138 KMP_DEBUG_ASSERT(dims[j].up > dims[j].lo);
4139 range_length = (kmp_uint64)(dims[j].up - dims[j].lo) / dims[j].st + 1;
4141 KMP_DEBUG_ASSERT(dims[j].lo > dims[j].up);
4143 (kmp_uint64)(dims[j].lo - dims[j].up) / (-dims[j].st) + 1;
4146 pr_buf->th_doacross_info[last++] = range_length;
4147 pr_buf->th_doacross_info[last++] = dims[j].lo;
4148 pr_buf->th_doacross_info[last++] = dims[j].up;
4149 pr_buf->th_doacross_info[last++] = dims[j].st;
4154 if (dims[0].st == 1) {
4155 trace_count = dims[0].up - dims[0].lo + 1;
4156 }
else if (dims[0].st > 0) {
4157 KMP_DEBUG_ASSERT(dims[0].up > dims[0].lo);
4158 trace_count = (kmp_uint64)(dims[0].up - dims[0].lo) / dims[0].st + 1;
4160 KMP_DEBUG_ASSERT(dims[0].lo > dims[0].up);
4161 trace_count = (kmp_uint64)(dims[0].lo - dims[0].up) / (-dims[0].st) + 1;
4163 for (j = 1; j < num_dims; ++j) {
4164 trace_count *= pr_buf->th_doacross_info[4 * j + 1];
4166 KMP_DEBUG_ASSERT(trace_count > 0);
4170 if (idx != sh_buf->doacross_buf_idx) {
4172 __kmp_wait_4((
volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
4179 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET32(
4180 (
volatile kmp_int32 *)&sh_buf->doacross_flags, NULL, 1);
4182 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET64(
4183 (
volatile kmp_int64 *)&sh_buf->doacross_flags, NULL, 1LL);
4185 if (flags == NULL) {
4188 (size_t)trace_count / 8 + 8;
4189 flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
4191 sh_buf->doacross_flags = flags;
4192 }
else if (flags == (kmp_uint32 *)1) {
4195 while (*(
volatile kmp_int32 *)&sh_buf->doacross_flags == 1)
4197 while (*(
volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
4204 KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1);
4205 pr_buf->th_doacross_flags =
4206 sh_buf->doacross_flags;
4208 KA_TRACE(20, (
"__kmpc_doacross_init() exit: T#%d\n", gtid));
4211void __kmpc_doacross_wait(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4212 __kmp_assert_valid_gtid(gtid);
4216 kmp_int64 iter_number;
4217 kmp_info_t *th = __kmp_threads[gtid];
4218 kmp_team_t *team = th->th.th_team;
4220 kmp_int64 lo, up, st;
4222 KA_TRACE(20, (
"__kmpc_doacross_wait() enter: called T#%d\n", gtid));
4223 if (team->t.t_serialized) {
4224 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: serialized team\n"));
4229 pr_buf = th->th.th_dispatch;
4230 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4231 num_dims = (size_t)pr_buf->th_doacross_info[0];
4232 lo = pr_buf->th_doacross_info[2];
4233 up = pr_buf->th_doacross_info[3];
4234 st = pr_buf->th_doacross_info[4];
4235#if OMPT_SUPPORT && OMPT_OPTIONAL
4236 ompt_dependence_t deps[num_dims];
4239 if (vec[0] < lo || vec[0] > up) {
4240 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4241 "bounds [%lld,%lld]\n",
4242 gtid, vec[0], lo, up));
4245 iter_number = vec[0] - lo;
4246 }
else if (st > 0) {
4247 if (vec[0] < lo || vec[0] > up) {
4248 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4249 "bounds [%lld,%lld]\n",
4250 gtid, vec[0], lo, up));
4253 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4255 if (vec[0] > lo || vec[0] < up) {
4256 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4257 "bounds [%lld,%lld]\n",
4258 gtid, vec[0], lo, up));
4261 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4263#if OMPT_SUPPORT && OMPT_OPTIONAL
4264 deps[0].variable.value = iter_number;
4265 deps[0].dependence_type = ompt_dependence_type_sink;
4267 for (i = 1; i < num_dims; ++i) {
4270 ln = pr_buf->th_doacross_info[j + 1];
4271 lo = pr_buf->th_doacross_info[j + 2];
4272 up = pr_buf->th_doacross_info[j + 3];
4273 st = pr_buf->th_doacross_info[j + 4];
4275 if (vec[i] < lo || vec[i] > up) {
4276 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4277 "bounds [%lld,%lld]\n",
4278 gtid, vec[i], lo, up));
4282 }
else if (st > 0) {
4283 if (vec[i] < lo || vec[i] > up) {
4284 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4285 "bounds [%lld,%lld]\n",
4286 gtid, vec[i], lo, up));
4289 iter = (kmp_uint64)(vec[i] - lo) / st;
4291 if (vec[i] > lo || vec[i] < up) {
4292 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4293 "bounds [%lld,%lld]\n",
4294 gtid, vec[i], lo, up));
4297 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4299 iter_number = iter + ln * iter_number;
4300#if OMPT_SUPPORT && OMPT_OPTIONAL
4301 deps[i].variable.value = iter;
4302 deps[i].dependence_type = ompt_dependence_type_sink;
4305 shft = iter_number % 32;
4308 while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
4312#if OMPT_SUPPORT && OMPT_OPTIONAL
4313 if (ompt_enabled.ompt_callback_dependences) {
4314 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4315 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4319 (
"__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
4320 gtid, (iter_number << 5) + shft));
4323void __kmpc_doacross_post(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4324 __kmp_assert_valid_gtid(gtid);
4328 kmp_int64 iter_number;
4329 kmp_info_t *th = __kmp_threads[gtid];
4330 kmp_team_t *team = th->th.th_team;
4334 KA_TRACE(20, (
"__kmpc_doacross_post() enter: called T#%d\n", gtid));
4335 if (team->t.t_serialized) {
4336 KA_TRACE(20, (
"__kmpc_doacross_post() exit: serialized team\n"));
4342 pr_buf = th->th.th_dispatch;
4343 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4344 num_dims = (size_t)pr_buf->th_doacross_info[0];
4345 lo = pr_buf->th_doacross_info[2];
4346 st = pr_buf->th_doacross_info[4];
4347#if OMPT_SUPPORT && OMPT_OPTIONAL
4348 ompt_dependence_t deps[num_dims];
4351 iter_number = vec[0] - lo;
4352 }
else if (st > 0) {
4353 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4355 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4357#if OMPT_SUPPORT && OMPT_OPTIONAL
4358 deps[0].variable.value = iter_number;
4359 deps[0].dependence_type = ompt_dependence_type_source;
4361 for (i = 1; i < num_dims; ++i) {
4364 ln = pr_buf->th_doacross_info[j + 1];
4365 lo = pr_buf->th_doacross_info[j + 2];
4366 st = pr_buf->th_doacross_info[j + 4];
4369 }
else if (st > 0) {
4370 iter = (kmp_uint64)(vec[i] - lo) / st;
4372 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4374 iter_number = iter + ln * iter_number;
4375#if OMPT_SUPPORT && OMPT_OPTIONAL
4376 deps[i].variable.value = iter;
4377 deps[i].dependence_type = ompt_dependence_type_source;
4380#if OMPT_SUPPORT && OMPT_OPTIONAL
4381 if (ompt_enabled.ompt_callback_dependences) {
4382 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4383 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4386 shft = iter_number % 32;
4390 if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
4391 KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
4392 KA_TRACE(20, (
"__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
4393 (iter_number << 5) + shft));
4396void __kmpc_doacross_fini(
ident_t *loc,
int gtid) {
4397 __kmp_assert_valid_gtid(gtid);
4399 kmp_info_t *th = __kmp_threads[gtid];
4400 kmp_team_t *team = th->th.th_team;
4401 kmp_disp_t *pr_buf = th->th.th_dispatch;
4403 KA_TRACE(20, (
"__kmpc_doacross_fini() enter: called T#%d\n", gtid));
4404 if (team->t.t_serialized) {
4405 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: serialized team %p\n", team));
4409 KMP_TEST_THEN_INC32((kmp_uintptr_t)(pr_buf->th_doacross_info[1])) + 1;
4410 if (num_done == th->th.th_team_nproc) {
4412 int idx = pr_buf->th_doacross_buf_idx - 1;
4413 dispatch_shared_info_t *sh_buf =
4414 &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4415 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info[1] ==
4416 (kmp_int64)&sh_buf->doacross_num_done);
4417 KMP_DEBUG_ASSERT(num_done == sh_buf->doacross_num_done);
4418 KMP_DEBUG_ASSERT(idx == sh_buf->doacross_buf_idx);
4419 __kmp_thread_free(th, CCAST(kmp_uint32 *, sh_buf->doacross_flags));
4420 sh_buf->doacross_flags = NULL;
4421 sh_buf->doacross_num_done = 0;
4422 sh_buf->doacross_buf_idx +=
4423 __kmp_dispatch_num_buffers;
4426 pr_buf->th_doacross_flags = NULL;
4427 __kmp_thread_free(th, (
void *)pr_buf->th_doacross_info);
4428 pr_buf->th_doacross_info = NULL;
4429 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: T#%d\n", gtid));
4433void *omp_alloc(
size_t size, omp_allocator_handle_t allocator) {
4434 return __kmp_alloc(__kmp_entry_gtid(), 0, size, allocator);
4437void *omp_aligned_alloc(
size_t align,
size_t size,
4438 omp_allocator_handle_t allocator) {
4439 return __kmp_alloc(__kmp_entry_gtid(), align, size, allocator);
4442void *omp_calloc(
size_t nmemb,
size_t size, omp_allocator_handle_t allocator) {
4443 return __kmp_calloc(__kmp_entry_gtid(), 0, nmemb, size, allocator);
4446void *omp_aligned_calloc(
size_t align,
size_t nmemb,
size_t size,
4447 omp_allocator_handle_t allocator) {
4448 return __kmp_calloc(__kmp_entry_gtid(), align, nmemb, size, allocator);
4451void *omp_realloc(
void *ptr,
size_t size, omp_allocator_handle_t allocator,
4452 omp_allocator_handle_t free_allocator) {
4453 return __kmp_realloc(__kmp_entry_gtid(), ptr, size, allocator,
4457void omp_free(
void *ptr, omp_allocator_handle_t allocator) {
4458 ___kmpc_free(__kmp_entry_gtid(), ptr, allocator);
4462int __kmpc_get_target_offload(
void) {
4463 if (!__kmp_init_serial) {
4464 __kmp_serial_initialize();
4466 return __kmp_target_offload;
4469int __kmpc_pause_resource(kmp_pause_status_t level) {
4470 if (!__kmp_init_serial) {
4473 return __kmp_pause_resource(level);
4476void __kmpc_error(
ident_t *loc,
int severity,
const char *message) {
4477 if (!__kmp_init_serial)
4478 __kmp_serial_initialize();
4480 KMP_ASSERT(severity == severity_warning || severity == severity_fatal);
4483 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_error) {
4484 ompt_callbacks.ompt_callback(ompt_callback_error)(
4485 (ompt_severity_t)severity, message, KMP_STRLEN(message),
4486 OMPT_GET_RETURN_ADDRESS(0));
4492 kmp_str_loc_t str_loc = __kmp_str_loc_init(loc->
psource,
false);
4494 __kmp_str_format(
"%s:%d:%d", str_loc.file, str_loc.line, str_loc.col);
4495 __kmp_str_loc_free(&str_loc);
4497 src_loc = __kmp_str_format(
"unknown");
4500 if (severity == severity_warning)
4501 KMP_WARNING(UserDirectedWarning, src_loc, message);
4503 KMP_FATAL(UserDirectedError, src_loc, message);
4505 __kmp_str_free(&src_loc);
4509void __kmpc_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4511#if OMPT_SUPPORT && OMPT_OPTIONAL
4512 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4513 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4514 int tid = __kmp_tid_from_gtid(gtid);
4515 ompt_callbacks.ompt_callback(ompt_callback_work)(
4516 ompt_work_scope, ompt_scope_begin,
4517 &(team->t.ompt_team_info.parallel_data),
4518 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4519 OMPT_GET_RETURN_ADDRESS(0));
4525void __kmpc_end_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4527#if OMPT_SUPPORT && OMPT_OPTIONAL
4528 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4529 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4530 int tid = __kmp_tid_from_gtid(gtid);
4531 ompt_callbacks.ompt_callback(ompt_callback_work)(
4532 ompt_work_scope, ompt_scope_end,
4533 &(team->t.ompt_team_info.parallel_data),
4534 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4535 OMPT_GET_RETURN_ADDRESS(0));
4540#ifdef KMP_USE_VERSION_SYMBOLS
4549#ifdef omp_set_affinity_format
4550#undef omp_set_affinity_format
4552#ifdef omp_get_affinity_format
4553#undef omp_get_affinity_format
4555#ifdef omp_display_affinity
4556#undef omp_display_affinity
4558#ifdef omp_capture_affinity
4559#undef omp_capture_affinity
4561KMP_VERSION_OMPC_SYMBOL(ompc_set_affinity_format, omp_set_affinity_format, 50,
4563KMP_VERSION_OMPC_SYMBOL(ompc_get_affinity_format, omp_get_affinity_format, 50,
4565KMP_VERSION_OMPC_SYMBOL(ompc_display_affinity, omp_display_affinity, 50,
4567KMP_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_fork_call_if(ident_t *loc, kmp_int32 argc, kmpc_micro microtask, kmp_int32 cond, void *args)
void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_threads)
void __kmpc_set_thread_limit(ident_t *loc, kmp_int32 global_tid, kmp_int32 thread_limit)
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)
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)