22#include "ompt-specific.h"
24#define MAX_MESSAGE 512
40 if ((env = getenv(
"KMP_INITIAL_THREAD_BIND")) != NULL &&
41 __kmp_str_match_true(env)) {
42 __kmp_middle_initialize();
43 __kmp_assign_root_init_mask();
44 KC_TRACE(10, (
"__kmpc_begin: middle initialization called\n"));
45 }
else if (__kmp_ignore_mppbeg() == FALSE) {
47 __kmp_internal_begin();
48 KC_TRACE(10, (
"__kmpc_begin: called\n"));
66 if (__kmp_ignore_mppend() == FALSE) {
67 KC_TRACE(10, (
"__kmpc_end: called\n"));
68 KA_TRACE(30, (
"__kmpc_end\n"));
70 __kmp_internal_end_thread(-1);
72#if KMP_OS_WINDOWS && OMPT_SUPPORT
77 if (ompt_enabled.enabled)
78 __kmp_internal_end_library(__kmp_gtid_get_specific());
101 kmp_int32 gtid = __kmp_entry_gtid();
103 KC_TRACE(10, (
"__kmpc_global_thread_num: T#%d\n", gtid));
124 (
"__kmpc_global_num_threads: num_threads = %d\n", __kmp_all_nth));
126 return TCR_4(__kmp_all_nth);
136 KC_TRACE(10, (
"__kmpc_bound_thread_num: called\n"));
137 return __kmp_tid_from_gtid(__kmp_entry_gtid());
146 KC_TRACE(10, (
"__kmpc_bound_num_threads: called\n"));
148 return __kmp_entry_thread()->th.th_team->t.t_nproc;
168 if (__kmp_par_range == 0) {
175 semi2 = strchr(semi2,
';');
179 semi2 = strchr(semi2 + 1,
';');
183 if (__kmp_par_range_filename[0]) {
184 const char *name = semi2 - 1;
185 while ((name > loc->
psource) && (*name !=
'/') && (*name !=
';')) {
188 if ((*name ==
'/') || (*name ==
';')) {
191 if (strncmp(__kmp_par_range_filename, name, semi2 - name)) {
192 return __kmp_par_range < 0;
195 semi3 = strchr(semi2 + 1,
';');
196 if (__kmp_par_range_routine[0]) {
197 if ((semi3 != NULL) && (semi3 > semi2) &&
198 (strncmp(__kmp_par_range_routine, semi2 + 1, semi3 - semi2 - 1))) {
199 return __kmp_par_range < 0;
202 if (KMP_SSCANF(semi3 + 1,
"%d", &line_no) == 1) {
203 if ((line_no >= __kmp_par_range_lb) && (line_no <= __kmp_par_range_ub)) {
204 return __kmp_par_range > 0;
206 return __kmp_par_range < 0;
220 return __kmp_entry_thread()->th.th_root->r.r_active;
233 kmp_int32 num_threads) {
234 KA_TRACE(20, (
"__kmpc_push_num_threads: enter T#%d num_threads=%d\n",
235 global_tid, num_threads));
236 __kmp_assert_valid_gtid(global_tid);
237 __kmp_push_num_threads(loc, global_tid, num_threads);
240void __kmpc_pop_num_threads(
ident_t *loc, kmp_int32 global_tid) {
241 KA_TRACE(20, (
"__kmpc_pop_num_threads: enter\n"));
245void __kmpc_push_proc_bind(
ident_t *loc, kmp_int32 global_tid,
246 kmp_int32 proc_bind) {
247 KA_TRACE(20, (
"__kmpc_push_proc_bind: enter T#%d proc_bind=%d\n", global_tid,
249 __kmp_assert_valid_gtid(global_tid);
250 __kmp_push_proc_bind(loc, global_tid, (kmp_proc_bind_t)proc_bind);
264 int gtid = __kmp_entry_gtid();
266#if (KMP_STATS_ENABLED)
270 if (previous_state == stats_state_e::SERIAL_REGION) {
271 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_parallel_overhead);
273 KMP_PUSH_PARTITIONED_TIMER(OMP_parallel_overhead);
286 va_start(ap, microtask);
289 ompt_frame_t *ompt_frame;
290 if (ompt_enabled.enabled) {
291 kmp_info_t *master_th = __kmp_threads[gtid];
292 ompt_frame = &master_th->th.th_current_task->ompt_task_info.frame;
293 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
295 OMPT_STORE_RETURN_ADDRESS(gtid);
301 __kmp_fork_call(loc, gtid, fork_context_intel, argc,
302 VOLATILE_CAST(microtask_t) microtask,
303 VOLATILE_CAST(launch_t) __kmp_invoke_task_func,
308 __kmp_join_call(loc, gtid
318 if (ompt_enabled.enabled) {
319 ompt_frame->enter_frame = ompt_data_none;
325 if (previous_state == stats_state_e::SERIAL_REGION) {
326 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
327 KMP_SET_THREAD_STATE(previous_state);
329 KMP_POP_PARTITIONED_TIMER();
345 kmp_int32 cond,
void *args) {
346 int gtid = __kmp_entry_gtid();
356 void *exit_frame_ptr;
360 __kmp_invoke_microtask(VOLATILE_CAST(microtask_t) microtask, gtid,
369 __kmp_invoke_microtask(VOLATILE_CAST(microtask_t) microtask, gtid,
395 kmp_int32 num_teams, kmp_int32 num_threads) {
397 (
"__kmpc_push_num_teams: enter T#%d num_teams=%d num_threads=%d\n",
398 global_tid, num_teams, num_threads));
399 __kmp_assert_valid_gtid(global_tid);
400 __kmp_push_num_teams(loc, global_tid, num_teams, num_threads);
414 kmp_int32 thread_limit) {
415 __kmp_assert_valid_gtid(global_tid);
416 kmp_info_t *thread = __kmp_threads[global_tid];
417 if (thread_limit > 0)
418 thread->th.th_current_task->td_icvs.task_thread_limit = thread_limit;
438 kmp_int32 num_teams_lb, kmp_int32 num_teams_ub,
439 kmp_int32 num_threads) {
440 KA_TRACE(20, (
"__kmpc_push_num_teams_51: enter T#%d num_teams_lb=%d"
441 " num_teams_ub=%d num_threads=%d\n",
442 global_tid, num_teams_lb, num_teams_ub, num_threads));
443 __kmp_assert_valid_gtid(global_tid);
444 __kmp_push_num_teams_51(loc, global_tid, num_teams_lb, num_teams_ub,
460 int gtid = __kmp_entry_gtid();
461 kmp_info_t *this_thr = __kmp_threads[gtid];
463 va_start(ap, microtask);
468 if (previous_state == stats_state_e::SERIAL_REGION) {
469 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_teams_overhead);
471 KMP_PUSH_PARTITIONED_TIMER(OMP_teams_overhead);
476 this_thr->th.th_teams_microtask = microtask;
477 this_thr->th.th_teams_level =
478 this_thr->th.th_team->t.t_level;
481 kmp_team_t *parent_team = this_thr->th.th_team;
482 int tid = __kmp_tid_from_gtid(gtid);
483 if (ompt_enabled.enabled) {
484 parent_team->t.t_implicit_task_taskdata[tid]
485 .ompt_task_info.frame.enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
487 OMPT_STORE_RETURN_ADDRESS(gtid);
492 if (this_thr->th.th_teams_size.nteams == 0) {
493 __kmp_push_num_teams(loc, gtid, 0, 0);
495 KMP_DEBUG_ASSERT(this_thr->th.th_set_nproc >= 1);
496 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nteams >= 1);
497 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nth >= 1);
500 loc, gtid, fork_context_intel, argc,
501 VOLATILE_CAST(microtask_t) __kmp_teams_master,
502 VOLATILE_CAST(launch_t) __kmp_invoke_teams_master, kmp_va_addr_of(ap));
503 __kmp_join_call(loc, gtid
511 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
512 kmp_cg_root_t *tmp = this_thr->th.th_cg_roots;
513 this_thr->th.th_cg_roots = tmp->up;
514 KA_TRACE(100, (
"__kmpc_fork_teams: Thread %p popping node %p and moving up"
515 " to node %p. cg_nthreads was %d\n",
516 this_thr, tmp, this_thr->th.th_cg_roots, tmp->cg_nthreads));
517 KMP_DEBUG_ASSERT(tmp->cg_nthreads);
518 int i = tmp->cg_nthreads--;
523 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
524 this_thr->th.th_current_task->td_icvs.thread_limit =
525 this_thr->th.th_cg_roots->cg_thread_limit;
527 this_thr->th.th_teams_microtask = NULL;
528 this_thr->th.th_teams_level = 0;
529 *(kmp_int64 *)(&this_thr->th.th_teams_size) = 0L;
532 if (previous_state == stats_state_e::SERIAL_REGION) {
533 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
534 KMP_SET_THREAD_STATE(previous_state);
536 KMP_POP_PARTITIONED_TIMER();
545int __kmpc_invoke_task_func(
int gtid) {
return __kmp_invoke_task_func(gtid); }
563 __kmp_assert_valid_gtid(global_tid);
565 OMPT_STORE_RETURN_ADDRESS(global_tid);
567 __kmp_serialized_parallel(loc, global_tid);
578 kmp_internal_control_t *top;
579 kmp_info_t *this_thr;
580 kmp_team_t *serial_team;
583 (
"__kmpc_end_serialized_parallel: called by T#%d\n", global_tid));
591 __kmp_assert_valid_gtid(global_tid);
592 if (!TCR_4(__kmp_init_parallel))
593 __kmp_parallel_initialize();
595 __kmp_resume_if_soft_paused();
597 this_thr = __kmp_threads[global_tid];
598 serial_team = this_thr->th.th_serial_team;
600 kmp_task_team_t *task_team = this_thr->th.th_task_team;
602 if (task_team != NULL && (task_team->tt.tt_found_proxy_tasks ||
603 task_team->tt.tt_hidden_helper_task_encountered))
604 __kmp_task_team_wait(this_thr, serial_team USE_ITT_BUILD_ARG(NULL));
607 KMP_DEBUG_ASSERT(serial_team);
608 KMP_ASSERT(serial_team->t.t_serialized);
609 KMP_DEBUG_ASSERT(this_thr->th.th_team == serial_team);
610 KMP_DEBUG_ASSERT(serial_team != this_thr->th.th_root->r.r_root_team);
611 KMP_DEBUG_ASSERT(serial_team->t.t_threads);
612 KMP_DEBUG_ASSERT(serial_team->t.t_threads[0] == this_thr);
615 if (ompt_enabled.enabled &&
616 this_thr->th.ompt_thread_info.state != ompt_state_overhead) {
617 OMPT_CUR_TASK_INFO(this_thr)->frame.exit_frame = ompt_data_none;
618 if (ompt_enabled.ompt_callback_implicit_task) {
619 ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
620 ompt_scope_end, NULL, OMPT_CUR_TASK_DATA(this_thr), 1,
621 OMPT_CUR_TASK_INFO(this_thr)->thread_num, ompt_task_implicit);
625 ompt_data_t *parent_task_data;
626 __ompt_get_task_info_internal(1, NULL, &parent_task_data, NULL, NULL, NULL);
628 if (ompt_enabled.ompt_callback_parallel_end) {
629 ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
630 &(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
631 ompt_parallel_invoker_program | ompt_parallel_team,
632 OMPT_LOAD_RETURN_ADDRESS(global_tid));
634 __ompt_lw_taskteam_unlink(this_thr);
635 this_thr->th.ompt_thread_info.state = ompt_state_overhead;
641 top = serial_team->t.t_control_stack_top;
642 if (top && top->serial_nesting_level == serial_team->t.t_serialized) {
643 copy_icvs(&serial_team->t.t_threads[0]->th.th_current_task->td_icvs, top);
644 serial_team->t.t_control_stack_top = top->next;
649 KMP_DEBUG_ASSERT(serial_team->t.t_dispatch->th_disp_buffer);
651 dispatch_private_info_t *disp_buffer =
652 serial_team->t.t_dispatch->th_disp_buffer;
653 serial_team->t.t_dispatch->th_disp_buffer =
654 serial_team->t.t_dispatch->th_disp_buffer->next;
655 __kmp_free(disp_buffer);
657 this_thr->th.th_def_allocator = serial_team->t.t_def_allocator;
659 --serial_team->t.t_serialized;
660 if (serial_team->t.t_serialized == 0) {
664#if KMP_ARCH_X86 || KMP_ARCH_X86_64
665 if (__kmp_inherit_fp_control && serial_team->t.t_fp_control_saved) {
666 __kmp_clear_x87_fpu_status_word();
667 __kmp_load_x87_fpu_control_word(&serial_team->t.t_x87_fpu_control_word);
668 __kmp_load_mxcsr(&serial_team->t.t_mxcsr);
672 __kmp_pop_current_task_from_thread(this_thr);
674 if (ompd_state & OMPD_ENABLE_BP)
675 ompd_bp_parallel_end();
678 this_thr->th.th_team = serial_team->t.t_parent;
679 this_thr->th.th_info.ds.ds_tid = serial_team->t.t_master_tid;
682 this_thr->th.th_team_nproc = serial_team->t.t_parent->t.t_nproc;
683 this_thr->th.th_team_master =
684 serial_team->t.t_parent->t.t_threads[0];
685 this_thr->th.th_team_serialized = this_thr->th.th_team->t.t_serialized;
688 this_thr->th.th_dispatch =
689 &this_thr->th.th_team->t.t_dispatch[serial_team->t.t_master_tid];
691 KMP_ASSERT(this_thr->th.th_current_task->td_flags.executing == 0);
692 this_thr->th.th_current_task->td_flags.executing = 1;
694 if (__kmp_tasking_mode != tskm_immediate_exec) {
696 this_thr->th.th_task_team =
697 this_thr->th.th_team->t.t_task_team[this_thr->th.th_task_state];
699 (
"__kmpc_end_serialized_parallel: T#%d restoring task_team %p / "
701 global_tid, this_thr->th.th_task_team, this_thr->th.th_team));
703#if KMP_AFFINITY_SUPPORTED
704 if (this_thr->th.th_team->t.t_level == 0 && __kmp_affinity.flags.reset) {
705 __kmp_reset_root_init_mask(global_tid);
709 if (__kmp_tasking_mode != tskm_immediate_exec) {
710 KA_TRACE(20, (
"__kmpc_end_serialized_parallel: T#%d decreasing nesting "
711 "depth of serial team %p to %d\n",
712 global_tid, serial_team, serial_team->t.t_serialized));
716 serial_team->t.t_level--;
717 if (__kmp_env_consistency_check)
718 __kmp_pop_parallel(global_tid, NULL);
720 if (ompt_enabled.enabled)
721 this_thr->th.ompt_thread_info.state =
722 ((this_thr->th.th_team_serialized) ? ompt_state_work_serial
723 : ompt_state_work_parallel);
736 KC_TRACE(10, (
"__kmpc_flush: called\n"));
741#if OMPT_SUPPORT && OMPT_OPTIONAL
742 if (ompt_enabled.ompt_callback_flush) {
743 ompt_callbacks.ompt_callback(ompt_callback_flush)(
744 __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
759 KC_TRACE(10, (
"__kmpc_barrier: called T#%d\n", global_tid));
760 __kmp_assert_valid_gtid(global_tid);
762 if (!TCR_4(__kmp_init_parallel))
763 __kmp_parallel_initialize();
765 __kmp_resume_if_soft_paused();
767 if (__kmp_env_consistency_check) {
769 KMP_WARNING(ConstructIdentInvalid);
771 __kmp_check_barrier(global_tid, ct_barrier, loc);
775 ompt_frame_t *ompt_frame;
776 if (ompt_enabled.enabled) {
777 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
778 if (ompt_frame->enter_frame.ptr == NULL)
779 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
781 OMPT_STORE_RETURN_ADDRESS(global_tid);
783 __kmp_threads[global_tid]->th.th_ident = loc;
791 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
792#if OMPT_SUPPORT && OMPT_OPTIONAL
793 if (ompt_enabled.enabled) {
794 ompt_frame->enter_frame = ompt_data_none;
809 KC_TRACE(10, (
"__kmpc_master: called T#%d\n", global_tid));
810 __kmp_assert_valid_gtid(global_tid);
812 if (!TCR_4(__kmp_init_parallel))
813 __kmp_parallel_initialize();
815 __kmp_resume_if_soft_paused();
817 if (KMP_MASTER_GTID(global_tid)) {
819 KMP_PUSH_PARTITIONED_TIMER(OMP_master);
823#if OMPT_SUPPORT && OMPT_OPTIONAL
825 if (ompt_enabled.ompt_callback_masked) {
826 kmp_info_t *this_thr = __kmp_threads[global_tid];
827 kmp_team_t *team = this_thr->th.th_team;
829 int tid = __kmp_tid_from_gtid(global_tid);
830 ompt_callbacks.ompt_callback(ompt_callback_masked)(
831 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
832 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
833 OMPT_GET_RETURN_ADDRESS(0));
838 if (__kmp_env_consistency_check) {
839#if KMP_USE_DYNAMIC_LOCK
841 __kmp_push_sync(global_tid, ct_master, loc, NULL, 0);
843 __kmp_check_sync(global_tid, ct_master, loc, NULL, 0);
846 __kmp_push_sync(global_tid, ct_master, loc, NULL);
848 __kmp_check_sync(global_tid, ct_master, loc, NULL);
864 KC_TRACE(10, (
"__kmpc_end_master: called T#%d\n", global_tid));
865 __kmp_assert_valid_gtid(global_tid);
866 KMP_DEBUG_ASSERT(KMP_MASTER_GTID(global_tid));
867 KMP_POP_PARTITIONED_TIMER();
869#if OMPT_SUPPORT && OMPT_OPTIONAL
870 kmp_info_t *this_thr = __kmp_threads[global_tid];
871 kmp_team_t *team = this_thr->th.th_team;
872 if (ompt_enabled.ompt_callback_masked) {
873 int tid = __kmp_tid_from_gtid(global_tid);
874 ompt_callbacks.ompt_callback(ompt_callback_masked)(
875 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
876 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
877 OMPT_GET_RETURN_ADDRESS(0));
881 if (__kmp_env_consistency_check) {
882 if (KMP_MASTER_GTID(global_tid))
883 __kmp_pop_sync(global_tid, ct_master, loc);
898 KC_TRACE(10, (
"__kmpc_masked: called T#%d\n", global_tid));
899 __kmp_assert_valid_gtid(global_tid);
901 if (!TCR_4(__kmp_init_parallel))
902 __kmp_parallel_initialize();
904 __kmp_resume_if_soft_paused();
906 tid = __kmp_tid_from_gtid(global_tid);
909 KMP_PUSH_PARTITIONED_TIMER(OMP_masked);
913#if OMPT_SUPPORT && OMPT_OPTIONAL
915 if (ompt_enabled.ompt_callback_masked) {
916 kmp_info_t *this_thr = __kmp_threads[global_tid];
917 kmp_team_t *team = this_thr->th.th_team;
918 ompt_callbacks.ompt_callback(ompt_callback_masked)(
919 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
920 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
921 OMPT_GET_RETURN_ADDRESS(0));
926 if (__kmp_env_consistency_check) {
927#if KMP_USE_DYNAMIC_LOCK
929 __kmp_push_sync(global_tid, ct_masked, loc, NULL, 0);
931 __kmp_check_sync(global_tid, ct_masked, loc, NULL, 0);
934 __kmp_push_sync(global_tid, ct_masked, loc, NULL);
936 __kmp_check_sync(global_tid, ct_masked, loc, NULL);
952 KC_TRACE(10, (
"__kmpc_end_masked: called T#%d\n", global_tid));
953 __kmp_assert_valid_gtid(global_tid);
954 KMP_POP_PARTITIONED_TIMER();
956#if OMPT_SUPPORT && OMPT_OPTIONAL
957 kmp_info_t *this_thr = __kmp_threads[global_tid];
958 kmp_team_t *team = this_thr->th.th_team;
959 if (ompt_enabled.ompt_callback_masked) {
960 int tid = __kmp_tid_from_gtid(global_tid);
961 ompt_callbacks.ompt_callback(ompt_callback_masked)(
962 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
963 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
964 OMPT_GET_RETURN_ADDRESS(0));
968 if (__kmp_env_consistency_check) {
969 __kmp_pop_sync(global_tid, ct_masked, loc);
983 KMP_DEBUG_ASSERT(__kmp_init_serial);
985 KC_TRACE(10, (
"__kmpc_ordered: called T#%d\n", gtid));
986 __kmp_assert_valid_gtid(gtid);
988 if (!TCR_4(__kmp_init_parallel))
989 __kmp_parallel_initialize();
991 __kmp_resume_if_soft_paused();
994 __kmp_itt_ordered_prep(gtid);
998 th = __kmp_threads[gtid];
1000#if OMPT_SUPPORT && OMPT_OPTIONAL
1004 OMPT_STORE_RETURN_ADDRESS(gtid);
1005 if (ompt_enabled.enabled) {
1006 team = __kmp_team_from_gtid(gtid);
1007 lck = (ompt_wait_id_t)(uintptr_t)&team->t.t_ordered.dt.t_value;
1009 th->th.ompt_thread_info.wait_id = lck;
1010 th->th.ompt_thread_info.state = ompt_state_wait_ordered;
1013 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1014 if (ompt_enabled.ompt_callback_mutex_acquire) {
1015 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1016 ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin, lck,
1022 if (th->th.th_dispatch->th_deo_fcn != 0)
1023 (*th->th.th_dispatch->th_deo_fcn)(>id, &cid, loc);
1025 __kmp_parallel_deo(>id, &cid, loc);
1027#if OMPT_SUPPORT && OMPT_OPTIONAL
1028 if (ompt_enabled.enabled) {
1030 th->th.ompt_thread_info.state = ompt_state_work_parallel;
1031 th->th.ompt_thread_info.wait_id = 0;
1034 if (ompt_enabled.ompt_callback_mutex_acquired) {
1035 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1036 ompt_mutex_ordered, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1042 __kmp_itt_ordered_start(gtid);
1057 KC_TRACE(10, (
"__kmpc_end_ordered: called T#%d\n", gtid));
1058 __kmp_assert_valid_gtid(gtid);
1061 __kmp_itt_ordered_end(gtid);
1065 th = __kmp_threads[gtid];
1067 if (th->th.th_dispatch->th_dxo_fcn != 0)
1068 (*th->th.th_dispatch->th_dxo_fcn)(>id, &cid, loc);
1070 __kmp_parallel_dxo(>id, &cid, loc);
1072#if OMPT_SUPPORT && OMPT_OPTIONAL
1073 OMPT_STORE_RETURN_ADDRESS(gtid);
1074 if (ompt_enabled.ompt_callback_mutex_released) {
1075 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1077 (ompt_wait_id_t)(uintptr_t)&__kmp_team_from_gtid(gtid)
1078 ->t.t_ordered.dt.t_value,
1079 OMPT_LOAD_RETURN_ADDRESS(gtid));
1084#if KMP_USE_DYNAMIC_LOCK
1086static __forceinline
void
1087__kmp_init_indirect_csptr(kmp_critical_name *crit,
ident_t const *loc,
1088 kmp_int32 gtid, kmp_indirect_locktag_t tag) {
1092 kmp_indirect_lock_t **lck;
1093 lck = (kmp_indirect_lock_t **)crit;
1094 kmp_indirect_lock_t *ilk = __kmp_allocate_indirect_lock(&idx, gtid, tag);
1095 KMP_I_LOCK_FUNC(ilk, init)(ilk->lock);
1096 KMP_SET_I_LOCK_LOCATION(ilk, loc);
1097 KMP_SET_I_LOCK_FLAGS(ilk, kmp_lf_critical_section);
1099 (
"__kmp_init_indirect_csptr: initialized indirect lock #%d\n", tag));
1101 __kmp_itt_critical_creating(ilk->lock, loc);
1103 int status = KMP_COMPARE_AND_STORE_PTR(lck,
nullptr, ilk);
1106 __kmp_itt_critical_destroyed(ilk->lock);
1112 KMP_DEBUG_ASSERT(*lck != NULL);
1116#define KMP_ACQUIRE_TAS_LOCK(lock, gtid) \
1118 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1119 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1120 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1121 if (KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1122 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \
1124 KMP_FSYNC_PREPARE(l); \
1125 KMP_INIT_YIELD(spins); \
1126 kmp_backoff_t backoff = __kmp_spin_backoff_params; \
1128 if (TCR_4(__kmp_nth) > \
1129 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
1132 KMP_YIELD_SPIN(spins); \
1134 __kmp_spin_backoff(&backoff); \
1136 KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1137 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)); \
1139 KMP_FSYNC_ACQUIRED(l); \
1143#define KMP_TEST_TAS_LOCK(lock, gtid, rc) \
1145 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1146 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1147 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1148 rc = KMP_ATOMIC_LD_RLX(&l->lk.poll) == tas_free && \
1149 __kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy); \
1153#define KMP_RELEASE_TAS_LOCK(lock, gtid) \
1154 { KMP_ATOMIC_ST_REL(&((kmp_tas_lock_t *)lock)->lk.poll, KMP_LOCK_FREE(tas)); }
1158#include <sys/syscall.h>
1168#define KMP_ACQUIRE_FUTEX_LOCK(lock, gtid) \
1170 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1171 kmp_int32 gtid_code = (gtid + 1) << 1; \
1173 KMP_FSYNC_PREPARE(ftx); \
1174 kmp_int32 poll_val; \
1175 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( \
1176 &(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1177 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { \
1178 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; \
1180 if (!KMP_COMPARE_AND_STORE_RET32(&(ftx->lk.poll), poll_val, \
1182 KMP_LOCK_BUSY(1, futex))) { \
1185 poll_val |= KMP_LOCK_BUSY(1, futex); \
1188 if ((rc = syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAIT, poll_val, \
1189 NULL, NULL, 0)) != 0) { \
1194 KMP_FSYNC_ACQUIRED(ftx); \
1198#define KMP_TEST_FUTEX_LOCK(lock, gtid, rc) \
1200 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1201 if (KMP_COMPARE_AND_STORE_ACQ32(&(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1202 KMP_LOCK_BUSY(gtid + 1 << 1, futex))) { \
1203 KMP_FSYNC_ACQUIRED(ftx); \
1211#define KMP_RELEASE_FUTEX_LOCK(lock, gtid) \
1213 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1215 KMP_FSYNC_RELEASING(ftx); \
1216 kmp_int32 poll_val = \
1217 KMP_XCHG_FIXED32(&(ftx->lk.poll), KMP_LOCK_FREE(futex)); \
1218 if (KMP_LOCK_STRIP(poll_val) & 1) { \
1219 syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAKE, \
1220 KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \
1223 KMP_YIELD_OVERSUB(); \
1230static kmp_user_lock_p __kmp_get_critical_section_ptr(kmp_critical_name *crit,
1233 kmp_user_lock_p *lck_pp = (kmp_user_lock_p *)crit;
1236 kmp_user_lock_p lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1243 lck = __kmp_user_lock_allocate(&idx, gtid, kmp_lf_critical_section);
1244 __kmp_init_user_lock_with_checks(lck);
1245 __kmp_set_user_lock_location(lck, loc);
1247 __kmp_itt_critical_creating(lck);
1258 int status = KMP_COMPARE_AND_STORE_PTR(lck_pp, 0, lck);
1263 __kmp_itt_critical_destroyed(lck);
1267 __kmp_destroy_user_lock_with_checks(lck);
1268 __kmp_user_lock_free(&idx, gtid, lck);
1269 lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1270 KMP_DEBUG_ASSERT(lck != NULL);
1289 kmp_critical_name *crit) {
1290#if KMP_USE_DYNAMIC_LOCK
1291#if OMPT_SUPPORT && OMPT_OPTIONAL
1292 OMPT_STORE_RETURN_ADDRESS(global_tid);
1294 __kmpc_critical_with_hint(loc, global_tid, crit, omp_lock_hint_none);
1297#if OMPT_SUPPORT && OMPT_OPTIONAL
1298 ompt_state_t prev_state = ompt_state_undefined;
1299 ompt_thread_info_t ti;
1301 kmp_user_lock_p lck;
1303 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1304 __kmp_assert_valid_gtid(global_tid);
1308 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1309 KMP_CHECK_USER_LOCK_INIT();
1311 if ((__kmp_user_lock_kind == lk_tas) &&
1312 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1313 lck = (kmp_user_lock_p)crit;
1316 else if ((__kmp_user_lock_kind == lk_futex) &&
1317 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1318 lck = (kmp_user_lock_p)crit;
1322 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
1325 if (__kmp_env_consistency_check)
1326 __kmp_push_sync(global_tid, ct_critical, loc, lck);
1334 __kmp_itt_critical_acquiring(lck);
1336#if OMPT_SUPPORT && OMPT_OPTIONAL
1337 OMPT_STORE_RETURN_ADDRESS(gtid);
1338 void *codeptr_ra = NULL;
1339 if (ompt_enabled.enabled) {
1340 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1342 prev_state = ti.state;
1343 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1344 ti.state = ompt_state_wait_critical;
1347 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1348 if (ompt_enabled.ompt_callback_mutex_acquire) {
1349 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1350 ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
1351 (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1357 __kmp_acquire_user_lock_with_checks(lck, global_tid);
1360 __kmp_itt_critical_acquired(lck);
1362#if OMPT_SUPPORT && OMPT_OPTIONAL
1363 if (ompt_enabled.enabled) {
1365 ti.state = prev_state;
1369 if (ompt_enabled.ompt_callback_mutex_acquired) {
1370 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1371 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1375 KMP_POP_PARTITIONED_TIMER();
1377 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1378 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1382#if KMP_USE_DYNAMIC_LOCK
1385static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
1387#define KMP_TSX_LOCK(seq) lockseq_##seq
1389#define KMP_TSX_LOCK(seq) __kmp_user_lock_seq
1392#if KMP_ARCH_X86 || KMP_ARCH_X86_64
1393#define KMP_CPUINFO_RTM (__kmp_cpuinfo.flags.rtm)
1395#define KMP_CPUINFO_RTM 0
1399 if (hint & kmp_lock_hint_hle)
1400 return KMP_TSX_LOCK(hle);
1401 if (hint & kmp_lock_hint_rtm)
1402 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_queuing) : __kmp_user_lock_seq;
1403 if (hint & kmp_lock_hint_adaptive)
1404 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
1407 if ((hint & omp_lock_hint_contended) && (hint & omp_lock_hint_uncontended))
1408 return __kmp_user_lock_seq;
1409 if ((hint & omp_lock_hint_speculative) &&
1410 (hint & omp_lock_hint_nonspeculative))
1411 return __kmp_user_lock_seq;
1414 if (hint & omp_lock_hint_contended)
1415 return lockseq_queuing;
1418 if ((hint & omp_lock_hint_uncontended) && !(hint & omp_lock_hint_speculative))
1422 if (hint & omp_lock_hint_speculative)
1423 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_spin) : __kmp_user_lock_seq;
1425 return __kmp_user_lock_seq;
1428#if OMPT_SUPPORT && OMPT_OPTIONAL
1429#if KMP_USE_DYNAMIC_LOCK
1430static kmp_mutex_impl_t
1431__ompt_get_mutex_impl_type(
void *user_lock, kmp_indirect_lock_t *ilock = 0) {
1433 switch (KMP_EXTRACT_D_TAG(user_lock)) {
1438 return kmp_mutex_impl_queuing;
1441 return kmp_mutex_impl_spin;
1444 case locktag_rtm_spin:
1445 return kmp_mutex_impl_speculative;
1448 return kmp_mutex_impl_none;
1450 ilock = KMP_LOOKUP_I_LOCK(user_lock);
1453 switch (ilock->type) {
1455 case locktag_adaptive:
1456 case locktag_rtm_queuing:
1457 return kmp_mutex_impl_speculative;
1459 case locktag_nested_tas:
1460 return kmp_mutex_impl_spin;
1462 case locktag_nested_futex:
1464 case locktag_ticket:
1465 case locktag_queuing:
1467 case locktag_nested_ticket:
1468 case locktag_nested_queuing:
1469 case locktag_nested_drdpa:
1470 return kmp_mutex_impl_queuing;
1472 return kmp_mutex_impl_none;
1477static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
1478 switch (__kmp_user_lock_kind) {
1480 return kmp_mutex_impl_spin;
1487 return kmp_mutex_impl_queuing;
1490 case lk_rtm_queuing:
1493 return kmp_mutex_impl_speculative;
1496 return kmp_mutex_impl_none;
1515void __kmpc_critical_with_hint(
ident_t *loc, kmp_int32 global_tid,
1516 kmp_critical_name *crit, uint32_t hint) {
1518 kmp_user_lock_p lck;
1519#if OMPT_SUPPORT && OMPT_OPTIONAL
1520 ompt_state_t prev_state = ompt_state_undefined;
1521 ompt_thread_info_t ti;
1523 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(global_tid);
1525 codeptr = OMPT_GET_RETURN_ADDRESS(0);
1528 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1529 __kmp_assert_valid_gtid(global_tid);
1531 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
1533 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1534 kmp_dyna_lockseq_t lockseq = __kmp_map_hint_to_lock(hint);
1536 if (KMP_IS_D_LOCK(lockseq)) {
1537 KMP_COMPARE_AND_STORE_ACQ32(
1538 (
volatile kmp_int32 *)&((kmp_base_tas_lock_t *)crit)->poll, 0,
1539 KMP_GET_D_TAG(lockseq));
1541 __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lockseq));
1547 if (KMP_EXTRACT_D_TAG(lk) != 0) {
1548 lck = (kmp_user_lock_p)lk;
1549 if (__kmp_env_consistency_check) {
1550 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1551 __kmp_map_hint_to_lock(hint));
1554 __kmp_itt_critical_acquiring(lck);
1556#if OMPT_SUPPORT && OMPT_OPTIONAL
1557 if (ompt_enabled.enabled) {
1558 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1560 prev_state = ti.state;
1561 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1562 ti.state = ompt_state_wait_critical;
1565 if (ompt_enabled.ompt_callback_mutex_acquire) {
1566 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1567 ompt_mutex_critical, (
unsigned int)hint,
1568 __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)(uintptr_t)lck,
1573#if KMP_USE_INLINED_TAS
1574 if (lockseq == lockseq_tas && !__kmp_env_consistency_check) {
1575 KMP_ACQUIRE_TAS_LOCK(lck, global_tid);
1577#elif KMP_USE_INLINED_FUTEX
1578 if (lockseq == lockseq_futex && !__kmp_env_consistency_check) {
1579 KMP_ACQUIRE_FUTEX_LOCK(lck, global_tid);
1583 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
1586 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
1588 if (__kmp_env_consistency_check) {
1589 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1590 __kmp_map_hint_to_lock(hint));
1593 __kmp_itt_critical_acquiring(lck);
1595#if OMPT_SUPPORT && OMPT_OPTIONAL
1596 if (ompt_enabled.enabled) {
1597 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1599 prev_state = ti.state;
1600 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1601 ti.state = ompt_state_wait_critical;
1604 if (ompt_enabled.ompt_callback_mutex_acquire) {
1605 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1606 ompt_mutex_critical, (
unsigned int)hint,
1607 __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)(uintptr_t)lck,
1612 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
1614 KMP_POP_PARTITIONED_TIMER();
1617 __kmp_itt_critical_acquired(lck);
1619#if OMPT_SUPPORT && OMPT_OPTIONAL
1620 if (ompt_enabled.enabled) {
1622 ti.state = prev_state;
1626 if (ompt_enabled.ompt_callback_mutex_acquired) {
1627 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1628 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
1633 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1634 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1649 kmp_critical_name *crit) {
1650 kmp_user_lock_p lck;
1652 KC_TRACE(10, (
"__kmpc_end_critical: called T#%d\n", global_tid));
1654#if KMP_USE_DYNAMIC_LOCK
1655 int locktag = KMP_EXTRACT_D_TAG(crit);
1657 lck = (kmp_user_lock_p)crit;
1658 KMP_ASSERT(lck != NULL);
1659 if (__kmp_env_consistency_check) {
1660 __kmp_pop_sync(global_tid, ct_critical, loc);
1663 __kmp_itt_critical_releasing(lck);
1665#if KMP_USE_INLINED_TAS
1666 if (locktag == locktag_tas && !__kmp_env_consistency_check) {
1667 KMP_RELEASE_TAS_LOCK(lck, global_tid);
1669#elif KMP_USE_INLINED_FUTEX
1670 if (locktag == locktag_futex && !__kmp_env_consistency_check) {
1671 KMP_RELEASE_FUTEX_LOCK(lck, global_tid);
1675 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
1678 kmp_indirect_lock_t *ilk =
1679 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
1680 KMP_ASSERT(ilk != NULL);
1682 if (__kmp_env_consistency_check) {
1683 __kmp_pop_sync(global_tid, ct_critical, loc);
1686 __kmp_itt_critical_releasing(lck);
1688 KMP_I_LOCK_FUNC(ilk, unset)(lck, global_tid);
1693 if ((__kmp_user_lock_kind == lk_tas) &&
1694 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1695 lck = (kmp_user_lock_p)crit;
1698 else if ((__kmp_user_lock_kind == lk_futex) &&
1699 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1700 lck = (kmp_user_lock_p)crit;
1704 lck = (kmp_user_lock_p)TCR_PTR(*((kmp_user_lock_p *)crit));
1707 KMP_ASSERT(lck != NULL);
1709 if (__kmp_env_consistency_check)
1710 __kmp_pop_sync(global_tid, ct_critical, loc);
1713 __kmp_itt_critical_releasing(lck);
1717 __kmp_release_user_lock_with_checks(lck, global_tid);
1721#if OMPT_SUPPORT && OMPT_OPTIONAL
1724 OMPT_STORE_RETURN_ADDRESS(global_tid);
1725 if (ompt_enabled.ompt_callback_mutex_released) {
1726 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1727 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck,
1728 OMPT_LOAD_RETURN_ADDRESS(0));
1732 KMP_POP_PARTITIONED_TIMER();
1733 KA_TRACE(15, (
"__kmpc_end_critical: done T#%d\n", global_tid));
1747 KC_TRACE(10, (
"__kmpc_barrier_master: called T#%d\n", global_tid));
1748 __kmp_assert_valid_gtid(global_tid);
1750 if (!TCR_4(__kmp_init_parallel))
1751 __kmp_parallel_initialize();
1753 __kmp_resume_if_soft_paused();
1755 if (__kmp_env_consistency_check)
1756 __kmp_check_barrier(global_tid, ct_barrier, loc);
1759 ompt_frame_t *ompt_frame;
1760 if (ompt_enabled.enabled) {
1761 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1762 if (ompt_frame->enter_frame.ptr == NULL)
1763 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1765 OMPT_STORE_RETURN_ADDRESS(global_tid);
1768 __kmp_threads[global_tid]->th.th_ident = loc;
1770 status = __kmp_barrier(bs_plain_barrier, global_tid, TRUE, 0, NULL, NULL);
1771#if OMPT_SUPPORT && OMPT_OPTIONAL
1772 if (ompt_enabled.enabled) {
1773 ompt_frame->enter_frame = ompt_data_none;
1777 return (status != 0) ? 0 : 1;
1790 KC_TRACE(10, (
"__kmpc_end_barrier_master: called T#%d\n", global_tid));
1791 __kmp_assert_valid_gtid(global_tid);
1792 __kmp_end_split_barrier(bs_plain_barrier, global_tid);
1807 KC_TRACE(10, (
"__kmpc_barrier_master_nowait: called T#%d\n", global_tid));
1808 __kmp_assert_valid_gtid(global_tid);
1810 if (!TCR_4(__kmp_init_parallel))
1811 __kmp_parallel_initialize();
1813 __kmp_resume_if_soft_paused();
1815 if (__kmp_env_consistency_check) {
1817 KMP_WARNING(ConstructIdentInvalid);
1819 __kmp_check_barrier(global_tid, ct_barrier, loc);
1823 ompt_frame_t *ompt_frame;
1824 if (ompt_enabled.enabled) {
1825 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1826 if (ompt_frame->enter_frame.ptr == NULL)
1827 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1829 OMPT_STORE_RETURN_ADDRESS(global_tid);
1832 __kmp_threads[global_tid]->th.th_ident = loc;
1834 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
1835#if OMPT_SUPPORT && OMPT_OPTIONAL
1836 if (ompt_enabled.enabled) {
1837 ompt_frame->enter_frame = ompt_data_none;
1843 if (__kmp_env_consistency_check) {
1849 __kmp_pop_sync(global_tid, ct_master, loc);
1869 __kmp_assert_valid_gtid(global_tid);
1870 kmp_int32 rc = __kmp_enter_single(global_tid, loc, TRUE);
1875 KMP_PUSH_PARTITIONED_TIMER(OMP_single);
1878#if OMPT_SUPPORT && OMPT_OPTIONAL
1879 kmp_info_t *this_thr = __kmp_threads[global_tid];
1880 kmp_team_t *team = this_thr->th.th_team;
1881 int tid = __kmp_tid_from_gtid(global_tid);
1883 if (ompt_enabled.enabled) {
1885 if (ompt_enabled.ompt_callback_work) {
1886 ompt_callbacks.ompt_callback(ompt_callback_work)(
1887 ompt_work_single_executor, ompt_scope_begin,
1888 &(team->t.ompt_team_info.parallel_data),
1889 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1890 1, OMPT_GET_RETURN_ADDRESS(0));
1893 if (ompt_enabled.ompt_callback_work) {
1894 ompt_callbacks.ompt_callback(ompt_callback_work)(
1895 ompt_work_single_other, ompt_scope_begin,
1896 &(team->t.ompt_team_info.parallel_data),
1897 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1898 1, OMPT_GET_RETURN_ADDRESS(0));
1899 ompt_callbacks.ompt_callback(ompt_callback_work)(
1900 ompt_work_single_other, ompt_scope_end,
1901 &(team->t.ompt_team_info.parallel_data),
1902 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1903 1, OMPT_GET_RETURN_ADDRESS(0));
1922 __kmp_assert_valid_gtid(global_tid);
1923 __kmp_exit_single(global_tid);
1924 KMP_POP_PARTITIONED_TIMER();
1926#if OMPT_SUPPORT && OMPT_OPTIONAL
1927 kmp_info_t *this_thr = __kmp_threads[global_tid];
1928 kmp_team_t *team = this_thr->th.th_team;
1929 int tid = __kmp_tid_from_gtid(global_tid);
1931 if (ompt_enabled.ompt_callback_work) {
1932 ompt_callbacks.ompt_callback(ompt_callback_work)(
1933 ompt_work_single_executor, ompt_scope_end,
1934 &(team->t.ompt_team_info.parallel_data),
1935 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
1936 OMPT_GET_RETURN_ADDRESS(0));
1949 KMP_POP_PARTITIONED_TIMER();
1950 KE_TRACE(10, (
"__kmpc_for_static_fini called T#%d\n", global_tid));
1952#if OMPT_SUPPORT && OMPT_OPTIONAL
1953 if (ompt_enabled.ompt_callback_work) {
1954 ompt_work_t ompt_work_type = ompt_work_loop;
1955 ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
1956 ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
1960 ompt_work_type = ompt_work_loop;
1962 ompt_work_type = ompt_work_sections;
1964 ompt_work_type = ompt_work_distribute;
1969 KMP_DEBUG_ASSERT(ompt_work_type);
1971 ompt_callbacks.ompt_callback(ompt_callback_work)(
1972 ompt_work_type, ompt_scope_end, &(team_info->parallel_data),
1973 &(task_info->task_data), 0, OMPT_GET_RETURN_ADDRESS(0));
1976 if (__kmp_env_consistency_check)
1977 __kmp_pop_workshare(global_tid, ct_pdo, loc);
1983void ompc_set_num_threads(
int arg) {
1985 __kmp_set_num_threads(arg, __kmp_entry_gtid());
1988void ompc_set_dynamic(
int flag) {
1992 thread = __kmp_entry_thread();
1994 __kmp_save_internal_controls(thread);
1996 set__dynamic(thread, flag ?
true : false);
1999void ompc_set_nested(
int flag) {
2003 thread = __kmp_entry_thread();
2005 __kmp_save_internal_controls(thread);
2007 set__max_active_levels(thread, flag ? __kmp_dflt_max_active_levels : 1);
2010void ompc_set_max_active_levels(
int max_active_levels) {
2015 __kmp_set_max_active_levels(__kmp_entry_gtid(), max_active_levels);
2018void ompc_set_schedule(omp_sched_t kind,
int modifier) {
2020 __kmp_set_schedule(__kmp_entry_gtid(), (kmp_sched_t)kind, modifier);
2023int ompc_get_ancestor_thread_num(
int level) {
2024 return __kmp_get_ancestor_thread_num(__kmp_entry_gtid(), level);
2027int ompc_get_team_size(
int level) {
2028 return __kmp_get_team_size(__kmp_entry_gtid(), level);
2032void KMP_EXPAND_NAME(ompc_set_affinity_format)(
char const *format) {
2033 if (!__kmp_init_serial) {
2034 __kmp_serial_initialize();
2036 __kmp_strncpy_truncate(__kmp_affinity_format, KMP_AFFINITY_FORMAT_SIZE,
2037 format, KMP_STRLEN(format) + 1);
2040size_t KMP_EXPAND_NAME(ompc_get_affinity_format)(
char *buffer,
size_t size) {
2042 if (!__kmp_init_serial) {
2043 __kmp_serial_initialize();
2045 format_size = KMP_STRLEN(__kmp_affinity_format);
2046 if (buffer && size) {
2047 __kmp_strncpy_truncate(buffer, size, __kmp_affinity_format,
2053void KMP_EXPAND_NAME(ompc_display_affinity)(
char const *format) {
2055 if (!TCR_4(__kmp_init_middle)) {
2056 __kmp_middle_initialize();
2058 __kmp_assign_root_init_mask();
2059 gtid = __kmp_get_gtid();
2060#if KMP_AFFINITY_SUPPORTED
2061 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 &&
2062 __kmp_affinity.flags.reset) {
2063 __kmp_reset_root_init_mask(gtid);
2066 __kmp_aux_display_affinity(gtid, format);
2069size_t KMP_EXPAND_NAME(ompc_capture_affinity)(
char *buffer,
size_t buf_size,
2070 char const *format) {
2072 size_t num_required;
2073 kmp_str_buf_t capture_buf;
2074 if (!TCR_4(__kmp_init_middle)) {
2075 __kmp_middle_initialize();
2077 __kmp_assign_root_init_mask();
2078 gtid = __kmp_get_gtid();
2079#if KMP_AFFINITY_SUPPORTED
2080 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 &&
2081 __kmp_affinity.flags.reset) {
2082 __kmp_reset_root_init_mask(gtid);
2085 __kmp_str_buf_init(&capture_buf);
2086 num_required = __kmp_aux_capture_affinity(gtid, format, &capture_buf);
2087 if (buffer && buf_size) {
2088 __kmp_strncpy_truncate(buffer, buf_size, capture_buf.str,
2089 capture_buf.used + 1);
2091 __kmp_str_buf_free(&capture_buf);
2092 return num_required;
2095void kmpc_set_stacksize(
int arg) {
2097 __kmp_aux_set_stacksize(arg);
2100void kmpc_set_stacksize_s(
size_t arg) {
2102 __kmp_aux_set_stacksize(arg);
2105void kmpc_set_blocktime(
int arg) {
2106 int gtid, tid, bt = arg;
2109 gtid = __kmp_entry_gtid();
2110 tid = __kmp_tid_from_gtid(gtid);
2111 thread = __kmp_thread_from_gtid(gtid);
2113 __kmp_aux_convert_blocktime(&bt);
2114 __kmp_aux_set_blocktime(bt, thread, tid);
2117void kmpc_set_library(
int arg) {
2119 __kmp_user_set_library((
enum library_type)arg);
2122void kmpc_set_defaults(
char const *str) {
2124 __kmp_aux_set_defaults(str, KMP_STRLEN(str));
2127void kmpc_set_disp_num_buffers(
int arg) {
2130 if (__kmp_init_serial == FALSE && arg >= KMP_MIN_DISP_NUM_BUFF &&
2131 arg <= KMP_MAX_DISP_NUM_BUFF) {
2132 __kmp_dispatch_num_buffers = arg;
2136int kmpc_set_affinity_mask_proc(
int proc,
void **mask) {
2137#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2140 if (!TCR_4(__kmp_init_middle)) {
2141 __kmp_middle_initialize();
2143 __kmp_assign_root_init_mask();
2144 return __kmp_aux_set_affinity_mask_proc(proc, mask);
2148int kmpc_unset_affinity_mask_proc(
int proc,
void **mask) {
2149#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2152 if (!TCR_4(__kmp_init_middle)) {
2153 __kmp_middle_initialize();
2155 __kmp_assign_root_init_mask();
2156 return __kmp_aux_unset_affinity_mask_proc(proc, mask);
2160int kmpc_get_affinity_mask_proc(
int proc,
void **mask) {
2161#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2164 if (!TCR_4(__kmp_init_middle)) {
2165 __kmp_middle_initialize();
2167 __kmp_assign_root_init_mask();
2168 return __kmp_aux_get_affinity_mask_proc(proc, mask);
2218 void *cpy_data,
void (*cpy_func)(
void *,
void *),
2221 KC_TRACE(10, (
"__kmpc_copyprivate: called T#%d\n", gtid));
2222 __kmp_assert_valid_gtid(gtid);
2226 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2228 if (__kmp_env_consistency_check) {
2230 KMP_WARNING(ConstructIdentInvalid);
2237 *data_ptr = cpy_data;
2240 ompt_frame_t *ompt_frame;
2241 if (ompt_enabled.enabled) {
2242 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2243 if (ompt_frame->enter_frame.ptr == NULL)
2244 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2246 OMPT_STORE_RETURN_ADDRESS(gtid);
2250 __kmp_threads[gtid]->th.th_ident = loc;
2252 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2255 (*cpy_func)(cpy_data, *data_ptr);
2261 OMPT_STORE_RETURN_ADDRESS(gtid);
2264 __kmp_threads[gtid]->th.th_ident = loc;
2267 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2268#if OMPT_SUPPORT && OMPT_OPTIONAL
2269 if (ompt_enabled.enabled) {
2270 ompt_frame->enter_frame = ompt_data_none;
2296 KC_TRACE(10, (
"__kmpc_copyprivate_light: called T#%d\n", gtid));
2300 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2302 if (__kmp_env_consistency_check) {
2304 KMP_WARNING(ConstructIdentInvalid);
2311 *data_ptr = cpy_data;
2314 ompt_frame_t *ompt_frame;
2315 if (ompt_enabled.enabled) {
2316 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2317 if (ompt_frame->enter_frame.ptr == NULL)
2318 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2319 OMPT_STORE_RETURN_ADDRESS(gtid);
2324 __kmp_threads[gtid]->th.th_ident = loc;
2326 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2333#define INIT_LOCK __kmp_init_user_lock_with_checks
2334#define INIT_NESTED_LOCK __kmp_init_nested_user_lock_with_checks
2335#define ACQUIRE_LOCK __kmp_acquire_user_lock_with_checks
2336#define ACQUIRE_LOCK_TIMED __kmp_acquire_user_lock_with_checks_timed
2337#define ACQUIRE_NESTED_LOCK __kmp_acquire_nested_user_lock_with_checks
2338#define ACQUIRE_NESTED_LOCK_TIMED \
2339 __kmp_acquire_nested_user_lock_with_checks_timed
2340#define RELEASE_LOCK __kmp_release_user_lock_with_checks
2341#define RELEASE_NESTED_LOCK __kmp_release_nested_user_lock_with_checks
2342#define TEST_LOCK __kmp_test_user_lock_with_checks
2343#define TEST_NESTED_LOCK __kmp_test_nested_user_lock_with_checks
2344#define DESTROY_LOCK __kmp_destroy_user_lock_with_checks
2345#define DESTROY_NESTED_LOCK __kmp_destroy_nested_user_lock_with_checks
2350#if KMP_USE_DYNAMIC_LOCK
2353static __forceinline
void __kmp_init_lock_with_hint(
ident_t *loc,
void **lock,
2354 kmp_dyna_lockseq_t seq) {
2355 if (KMP_IS_D_LOCK(seq)) {
2356 KMP_INIT_D_LOCK(lock, seq);
2358 __kmp_itt_lock_creating((kmp_user_lock_p)lock, NULL);
2361 KMP_INIT_I_LOCK(lock, seq);
2363 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2364 __kmp_itt_lock_creating(ilk->lock, loc);
2370static __forceinline
void
2371__kmp_init_nest_lock_with_hint(
ident_t *loc,
void **lock,
2372 kmp_dyna_lockseq_t seq) {
2375 if (seq == lockseq_hle || seq == lockseq_rtm_queuing ||
2376 seq == lockseq_rtm_spin || seq == lockseq_adaptive)
2377 seq = __kmp_user_lock_seq;
2381 seq = lockseq_nested_tas;
2385 seq = lockseq_nested_futex;
2388 case lockseq_ticket:
2389 seq = lockseq_nested_ticket;
2391 case lockseq_queuing:
2392 seq = lockseq_nested_queuing;
2395 seq = lockseq_nested_drdpa;
2398 seq = lockseq_nested_queuing;
2400 KMP_INIT_I_LOCK(lock, seq);
2402 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2403 __kmp_itt_lock_creating(ilk->lock, loc);
2408void __kmpc_init_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
void **user_lock,
2410 KMP_DEBUG_ASSERT(__kmp_init_serial);
2411 if (__kmp_env_consistency_check && user_lock == NULL) {
2412 KMP_FATAL(LockIsUninitialized,
"omp_init_lock_with_hint");
2415 __kmp_init_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2417#if OMPT_SUPPORT && OMPT_OPTIONAL
2419 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2421 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2422 if (ompt_enabled.ompt_callback_lock_init) {
2423 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2424 ompt_mutex_lock, (omp_lock_hint_t)hint,
2425 __ompt_get_mutex_impl_type(user_lock),
2426 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2432void __kmpc_init_nest_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
2433 void **user_lock, uintptr_t hint) {
2434 KMP_DEBUG_ASSERT(__kmp_init_serial);
2435 if (__kmp_env_consistency_check && user_lock == NULL) {
2436 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock_with_hint");
2439 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2441#if OMPT_SUPPORT && OMPT_OPTIONAL
2443 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2445 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2446 if (ompt_enabled.ompt_callback_lock_init) {
2447 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2448 ompt_mutex_nest_lock, (omp_lock_hint_t)hint,
2449 __ompt_get_mutex_impl_type(user_lock),
2450 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2458void __kmpc_init_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2459#if KMP_USE_DYNAMIC_LOCK
2461 KMP_DEBUG_ASSERT(__kmp_init_serial);
2462 if (__kmp_env_consistency_check && user_lock == NULL) {
2463 KMP_FATAL(LockIsUninitialized,
"omp_init_lock");
2465 __kmp_init_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2467#if OMPT_SUPPORT && OMPT_OPTIONAL
2469 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2471 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2472 if (ompt_enabled.ompt_callback_lock_init) {
2473 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2474 ompt_mutex_lock, omp_lock_hint_none,
2475 __ompt_get_mutex_impl_type(user_lock),
2476 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2482 static char const *
const func =
"omp_init_lock";
2483 kmp_user_lock_p lck;
2484 KMP_DEBUG_ASSERT(__kmp_init_serial);
2486 if (__kmp_env_consistency_check) {
2487 if (user_lock == NULL) {
2488 KMP_FATAL(LockIsUninitialized, func);
2492 KMP_CHECK_USER_LOCK_INIT();
2494 if ((__kmp_user_lock_kind == lk_tas) &&
2495 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2496 lck = (kmp_user_lock_p)user_lock;
2499 else if ((__kmp_user_lock_kind == lk_futex) &&
2500 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2501 lck = (kmp_user_lock_p)user_lock;
2505 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2508 __kmp_set_user_lock_location(lck, loc);
2510#if OMPT_SUPPORT && OMPT_OPTIONAL
2512 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2514 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2515 if (ompt_enabled.ompt_callback_lock_init) {
2516 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2517 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2518 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2523 __kmp_itt_lock_creating(lck);
2530void __kmpc_init_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2531#if KMP_USE_DYNAMIC_LOCK
2533 KMP_DEBUG_ASSERT(__kmp_init_serial);
2534 if (__kmp_env_consistency_check && user_lock == NULL) {
2535 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock");
2537 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2539#if OMPT_SUPPORT && OMPT_OPTIONAL
2541 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2543 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2544 if (ompt_enabled.ompt_callback_lock_init) {
2545 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2546 ompt_mutex_nest_lock, omp_lock_hint_none,
2547 __ompt_get_mutex_impl_type(user_lock),
2548 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2554 static char const *
const func =
"omp_init_nest_lock";
2555 kmp_user_lock_p lck;
2556 KMP_DEBUG_ASSERT(__kmp_init_serial);
2558 if (__kmp_env_consistency_check) {
2559 if (user_lock == NULL) {
2560 KMP_FATAL(LockIsUninitialized, func);
2564 KMP_CHECK_USER_LOCK_INIT();
2566 if ((__kmp_user_lock_kind == lk_tas) &&
2567 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2568 OMP_NEST_LOCK_T_SIZE)) {
2569 lck = (kmp_user_lock_p)user_lock;
2572 else if ((__kmp_user_lock_kind == lk_futex) &&
2573 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2574 OMP_NEST_LOCK_T_SIZE)) {
2575 lck = (kmp_user_lock_p)user_lock;
2579 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2582 INIT_NESTED_LOCK(lck);
2583 __kmp_set_user_lock_location(lck, loc);
2585#if OMPT_SUPPORT && OMPT_OPTIONAL
2587 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2589 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2590 if (ompt_enabled.ompt_callback_lock_init) {
2591 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2592 ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2593 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2598 __kmp_itt_lock_creating(lck);
2604void __kmpc_destroy_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2605#if KMP_USE_DYNAMIC_LOCK
2608 kmp_user_lock_p lck;
2609 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2610 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2612 lck = (kmp_user_lock_p)user_lock;
2614 __kmp_itt_lock_destroyed(lck);
2616#if OMPT_SUPPORT && OMPT_OPTIONAL
2618 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2620 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2621 if (ompt_enabled.ompt_callback_lock_destroy) {
2622 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2623 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2626 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2628 kmp_user_lock_p lck;
2630 if ((__kmp_user_lock_kind == lk_tas) &&
2631 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2632 lck = (kmp_user_lock_p)user_lock;
2635 else if ((__kmp_user_lock_kind == lk_futex) &&
2636 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2637 lck = (kmp_user_lock_p)user_lock;
2641 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_lock");
2644#if OMPT_SUPPORT && OMPT_OPTIONAL
2646 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2648 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2649 if (ompt_enabled.ompt_callback_lock_destroy) {
2650 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2651 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2656 __kmp_itt_lock_destroyed(lck);
2660 if ((__kmp_user_lock_kind == lk_tas) &&
2661 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2665 else if ((__kmp_user_lock_kind == lk_futex) &&
2666 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2671 __kmp_user_lock_free(user_lock, gtid, lck);
2677void __kmpc_destroy_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2678#if KMP_USE_DYNAMIC_LOCK
2681 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(user_lock);
2682 __kmp_itt_lock_destroyed(ilk->lock);
2684#if OMPT_SUPPORT && OMPT_OPTIONAL
2686 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2688 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2689 if (ompt_enabled.ompt_callback_lock_destroy) {
2690 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2691 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2694 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2698 kmp_user_lock_p lck;
2700 if ((__kmp_user_lock_kind == lk_tas) &&
2701 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2702 OMP_NEST_LOCK_T_SIZE)) {
2703 lck = (kmp_user_lock_p)user_lock;
2706 else if ((__kmp_user_lock_kind == lk_futex) &&
2707 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2708 OMP_NEST_LOCK_T_SIZE)) {
2709 lck = (kmp_user_lock_p)user_lock;
2713 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_nest_lock");
2716#if OMPT_SUPPORT && OMPT_OPTIONAL
2718 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2720 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2721 if (ompt_enabled.ompt_callback_lock_destroy) {
2722 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2723 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2728 __kmp_itt_lock_destroyed(lck);
2731 DESTROY_NESTED_LOCK(lck);
2733 if ((__kmp_user_lock_kind == lk_tas) &&
2734 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2735 OMP_NEST_LOCK_T_SIZE)) {
2739 else if ((__kmp_user_lock_kind == lk_futex) &&
2740 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2741 OMP_NEST_LOCK_T_SIZE)) {
2746 __kmp_user_lock_free(user_lock, gtid, lck);
2751void __kmpc_set_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2753#if KMP_USE_DYNAMIC_LOCK
2754 int tag = KMP_EXTRACT_D_TAG(user_lock);
2756 __kmp_itt_lock_acquiring(
2760#if OMPT_SUPPORT && OMPT_OPTIONAL
2762 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2764 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2765 if (ompt_enabled.ompt_callback_mutex_acquire) {
2766 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2767 ompt_mutex_lock, omp_lock_hint_none,
2768 __ompt_get_mutex_impl_type(user_lock),
2769 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2772#if KMP_USE_INLINED_TAS
2773 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2774 KMP_ACQUIRE_TAS_LOCK(user_lock, gtid);
2776#elif KMP_USE_INLINED_FUTEX
2777 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2778 KMP_ACQUIRE_FUTEX_LOCK(user_lock, gtid);
2782 __kmp_direct_set[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2785 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2787#if OMPT_SUPPORT && OMPT_OPTIONAL
2788 if (ompt_enabled.ompt_callback_mutex_acquired) {
2789 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2790 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2796 kmp_user_lock_p lck;
2798 if ((__kmp_user_lock_kind == lk_tas) &&
2799 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2800 lck = (kmp_user_lock_p)user_lock;
2803 else if ((__kmp_user_lock_kind == lk_futex) &&
2804 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2805 lck = (kmp_user_lock_p)user_lock;
2809 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_lock");
2813 __kmp_itt_lock_acquiring(lck);
2815#if OMPT_SUPPORT && OMPT_OPTIONAL
2817 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2819 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2820 if (ompt_enabled.ompt_callback_mutex_acquire) {
2821 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2822 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2823 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2827 ACQUIRE_LOCK(lck, gtid);
2830 __kmp_itt_lock_acquired(lck);
2833#if OMPT_SUPPORT && OMPT_OPTIONAL
2834 if (ompt_enabled.ompt_callback_mutex_acquired) {
2835 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2836 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2843void __kmpc_set_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2844#if KMP_USE_DYNAMIC_LOCK
2847 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
2849#if OMPT_SUPPORT && OMPT_OPTIONAL
2851 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2853 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2854 if (ompt_enabled.enabled) {
2855 if (ompt_enabled.ompt_callback_mutex_acquire) {
2856 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2857 ompt_mutex_nest_lock, omp_lock_hint_none,
2858 __ompt_get_mutex_impl_type(user_lock),
2859 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2863 int acquire_status =
2864 KMP_D_LOCK_FUNC(user_lock, set)((kmp_dyna_lock_t *)user_lock, gtid);
2865 (void)acquire_status;
2867 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2870#if OMPT_SUPPORT && OMPT_OPTIONAL
2871 if (ompt_enabled.enabled) {
2872 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2873 if (ompt_enabled.ompt_callback_mutex_acquired) {
2875 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2876 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
2880 if (ompt_enabled.ompt_callback_nest_lock) {
2882 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2883 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2891 kmp_user_lock_p lck;
2893 if ((__kmp_user_lock_kind == lk_tas) &&
2894 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2895 OMP_NEST_LOCK_T_SIZE)) {
2896 lck = (kmp_user_lock_p)user_lock;
2899 else if ((__kmp_user_lock_kind == lk_futex) &&
2900 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2901 OMP_NEST_LOCK_T_SIZE)) {
2902 lck = (kmp_user_lock_p)user_lock;
2906 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_nest_lock");
2910 __kmp_itt_lock_acquiring(lck);
2912#if OMPT_SUPPORT && OMPT_OPTIONAL
2914 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2916 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2917 if (ompt_enabled.enabled) {
2918 if (ompt_enabled.ompt_callback_mutex_acquire) {
2919 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2920 ompt_mutex_nest_lock, omp_lock_hint_none,
2921 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
2927 ACQUIRE_NESTED_LOCK(lck, gtid, &acquire_status);
2930 __kmp_itt_lock_acquired(lck);
2933#if OMPT_SUPPORT && OMPT_OPTIONAL
2934 if (ompt_enabled.enabled) {
2935 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2936 if (ompt_enabled.ompt_callback_mutex_acquired) {
2938 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2939 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2942 if (ompt_enabled.ompt_callback_nest_lock) {
2944 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2945 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2954void __kmpc_unset_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2955#if KMP_USE_DYNAMIC_LOCK
2957 int tag = KMP_EXTRACT_D_TAG(user_lock);
2959 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2961#if KMP_USE_INLINED_TAS
2962 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2963 KMP_RELEASE_TAS_LOCK(user_lock, gtid);
2965#elif KMP_USE_INLINED_FUTEX
2966 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2967 KMP_RELEASE_FUTEX_LOCK(user_lock, gtid);
2971 __kmp_direct_unset[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2974#if OMPT_SUPPORT && OMPT_OPTIONAL
2976 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2978 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2979 if (ompt_enabled.ompt_callback_mutex_released) {
2980 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2981 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2987 kmp_user_lock_p lck;
2992 if ((__kmp_user_lock_kind == lk_tas) &&
2993 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2994#if KMP_OS_LINUX && \
2995 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
2998 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3000 TCW_4(((kmp_user_lock_p)user_lock)->tas.lk.poll, 0);
3003#if OMPT_SUPPORT && OMPT_OPTIONAL
3005 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3007 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3008 if (ompt_enabled.ompt_callback_mutex_released) {
3009 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3010 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3016 lck = (kmp_user_lock_p)user_lock;
3020 else if ((__kmp_user_lock_kind == lk_futex) &&
3021 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3022 lck = (kmp_user_lock_p)user_lock;
3026 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_lock");
3030 __kmp_itt_lock_releasing(lck);
3033 RELEASE_LOCK(lck, gtid);
3035#if OMPT_SUPPORT && OMPT_OPTIONAL
3037 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3039 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3040 if (ompt_enabled.ompt_callback_mutex_released) {
3041 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3042 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3050void __kmpc_unset_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3051#if KMP_USE_DYNAMIC_LOCK
3054 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3056 int release_status =
3057 KMP_D_LOCK_FUNC(user_lock, unset)((kmp_dyna_lock_t *)user_lock, gtid);
3058 (void)release_status;
3060#if OMPT_SUPPORT && OMPT_OPTIONAL
3062 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3064 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3065 if (ompt_enabled.enabled) {
3066 if (release_status == KMP_LOCK_RELEASED) {
3067 if (ompt_enabled.ompt_callback_mutex_released) {
3069 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3070 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3073 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3075 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3076 ompt_scope_end, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3083 kmp_user_lock_p lck;
3087 if ((__kmp_user_lock_kind == lk_tas) &&
3088 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3089 OMP_NEST_LOCK_T_SIZE)) {
3090#if KMP_OS_LINUX && \
3091 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
3093 kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
3095 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3098#if OMPT_SUPPORT && OMPT_OPTIONAL
3099 int release_status = KMP_LOCK_STILL_HELD;
3102 if (--(tl->lk.depth_locked) == 0) {
3103 TCW_4(tl->lk.poll, 0);
3104#if OMPT_SUPPORT && OMPT_OPTIONAL
3105 release_status = KMP_LOCK_RELEASED;
3110#if OMPT_SUPPORT && OMPT_OPTIONAL
3112 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3114 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3115 if (ompt_enabled.enabled) {
3116 if (release_status == KMP_LOCK_RELEASED) {
3117 if (ompt_enabled.ompt_callback_mutex_released) {
3119 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3120 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3122 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3124 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3125 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3132 lck = (kmp_user_lock_p)user_lock;
3136 else if ((__kmp_user_lock_kind == lk_futex) &&
3137 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3138 OMP_NEST_LOCK_T_SIZE)) {
3139 lck = (kmp_user_lock_p)user_lock;
3143 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_nest_lock");
3147 __kmp_itt_lock_releasing(lck);
3151 release_status = RELEASE_NESTED_LOCK(lck, gtid);
3152#if OMPT_SUPPORT && OMPT_OPTIONAL
3154 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3156 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3157 if (ompt_enabled.enabled) {
3158 if (release_status == KMP_LOCK_RELEASED) {
3159 if (ompt_enabled.ompt_callback_mutex_released) {
3161 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3162 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3164 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3166 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3167 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3176int __kmpc_test_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3179#if KMP_USE_DYNAMIC_LOCK
3181 int tag = KMP_EXTRACT_D_TAG(user_lock);
3183 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3185#if OMPT_SUPPORT && OMPT_OPTIONAL
3187 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3189 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3190 if (ompt_enabled.ompt_callback_mutex_acquire) {
3191 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3192 ompt_mutex_test_lock, omp_lock_hint_none,
3193 __ompt_get_mutex_impl_type(user_lock),
3194 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3197#if KMP_USE_INLINED_TAS
3198 if (tag == locktag_tas && !__kmp_env_consistency_check) {
3199 KMP_TEST_TAS_LOCK(user_lock, gtid, rc);
3201#elif KMP_USE_INLINED_FUTEX
3202 if (tag == locktag_futex && !__kmp_env_consistency_check) {
3203 KMP_TEST_FUTEX_LOCK(user_lock, gtid, rc);
3207 rc = __kmp_direct_test[tag]((kmp_dyna_lock_t *)user_lock, gtid);
3211 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3213#if OMPT_SUPPORT && OMPT_OPTIONAL
3214 if (ompt_enabled.ompt_callback_mutex_acquired) {
3215 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3216 ompt_mutex_test_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3222 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3229 kmp_user_lock_p lck;
3232 if ((__kmp_user_lock_kind == lk_tas) &&
3233 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
3234 lck = (kmp_user_lock_p)user_lock;
3237 else if ((__kmp_user_lock_kind == lk_futex) &&
3238 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3239 lck = (kmp_user_lock_p)user_lock;
3243 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_lock");
3247 __kmp_itt_lock_acquiring(lck);
3249#if OMPT_SUPPORT && OMPT_OPTIONAL
3251 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3253 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3254 if (ompt_enabled.ompt_callback_mutex_acquire) {
3255 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3256 ompt_mutex_test_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
3257 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3261 rc = TEST_LOCK(lck, gtid);
3264 __kmp_itt_lock_acquired(lck);
3266 __kmp_itt_lock_cancelled(lck);
3269#if OMPT_SUPPORT && OMPT_OPTIONAL
3270 if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
3271 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3272 ompt_mutex_test_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3276 return (rc ? FTN_TRUE : FTN_FALSE);
3284int __kmpc_test_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3285#if KMP_USE_DYNAMIC_LOCK
3288 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3290#if OMPT_SUPPORT && OMPT_OPTIONAL
3292 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3294 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3295 if (ompt_enabled.ompt_callback_mutex_acquire) {
3296 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3297 ompt_mutex_test_nest_lock, omp_lock_hint_none,
3298 __ompt_get_mutex_impl_type(user_lock),
3299 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3302 rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
3305 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3307 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3310#if OMPT_SUPPORT && OMPT_OPTIONAL
3311 if (ompt_enabled.enabled && rc) {
3313 if (ompt_enabled.ompt_callback_mutex_acquired) {
3315 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3316 ompt_mutex_test_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3320 if (ompt_enabled.ompt_callback_nest_lock) {
3322 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3323 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3332 kmp_user_lock_p lck;
3335 if ((__kmp_user_lock_kind == lk_tas) &&
3336 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3337 OMP_NEST_LOCK_T_SIZE)) {
3338 lck = (kmp_user_lock_p)user_lock;
3341 else if ((__kmp_user_lock_kind == lk_futex) &&
3342 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3343 OMP_NEST_LOCK_T_SIZE)) {
3344 lck = (kmp_user_lock_p)user_lock;
3348 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_nest_lock");
3352 __kmp_itt_lock_acquiring(lck);
3355#if OMPT_SUPPORT && OMPT_OPTIONAL
3357 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3359 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3360 if (ompt_enabled.enabled) &&
3361 ompt_enabled.ompt_callback_mutex_acquire) {
3362 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3363 ompt_mutex_test_nest_lock, omp_lock_hint_none,
3364 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
3369 rc = TEST_NESTED_LOCK(lck, gtid);
3372 __kmp_itt_lock_acquired(lck);
3374 __kmp_itt_lock_cancelled(lck);
3377#if OMPT_SUPPORT && OMPT_OPTIONAL
3378 if (ompt_enabled.enabled && rc) {
3380 if (ompt_enabled.ompt_callback_mutex_acquired) {
3382 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3383 ompt_mutex_test_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3386 if (ompt_enabled.ompt_callback_nest_lock) {
3388 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3389 ompt_mutex_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3408#define __KMP_SET_REDUCTION_METHOD(gtid, rmethod) \
3409 ((__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) = (rmethod))
3411#define __KMP_GET_REDUCTION_METHOD(gtid) \
3412 (__kmp_threads[(gtid)]->th.th_local.packed_reduction_method)
3418static __forceinline
void
3419__kmp_enter_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3420 kmp_critical_name *crit) {
3426 kmp_user_lock_p lck;
3428#if KMP_USE_DYNAMIC_LOCK
3430 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
3433 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3434 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
3435 KMP_GET_D_TAG(__kmp_user_lock_seq));
3437 __kmp_init_indirect_csptr(crit, loc, global_tid,
3438 KMP_GET_I_TAG(__kmp_user_lock_seq));
3444 if (KMP_EXTRACT_D_TAG(lk) != 0) {
3445 lck = (kmp_user_lock_p)lk;
3446 KMP_DEBUG_ASSERT(lck != NULL);
3447 if (__kmp_env_consistency_check) {
3448 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3450 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
3452 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
3454 KMP_DEBUG_ASSERT(lck != NULL);
3455 if (__kmp_env_consistency_check) {
3456 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3458 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
3466 if (__kmp_base_user_lock_size <= INTEL_CRITICAL_SIZE) {
3467 lck = (kmp_user_lock_p)crit;
3469 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
3471 KMP_DEBUG_ASSERT(lck != NULL);
3473 if (__kmp_env_consistency_check)
3474 __kmp_push_sync(global_tid, ct_critical, loc, lck);
3476 __kmp_acquire_user_lock_with_checks(lck, global_tid);
3482static __forceinline
void
3483__kmp_end_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3484 kmp_critical_name *crit) {
3486 kmp_user_lock_p lck;
3488#if KMP_USE_DYNAMIC_LOCK
3490 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3491 lck = (kmp_user_lock_p)crit;
3492 if (__kmp_env_consistency_check)
3493 __kmp_pop_sync(global_tid, ct_critical, loc);
3494 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
3496 kmp_indirect_lock_t *ilk =
3497 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
3498 if (__kmp_env_consistency_check)
3499 __kmp_pop_sync(global_tid, ct_critical, loc);
3500 KMP_I_LOCK_FUNC(ilk, unset)(ilk->lock, global_tid);
3508 if (__kmp_base_user_lock_size > 32) {
3509 lck = *((kmp_user_lock_p *)crit);
3510 KMP_ASSERT(lck != NULL);
3512 lck = (kmp_user_lock_p)crit;
3515 if (__kmp_env_consistency_check)
3516 __kmp_pop_sync(global_tid, ct_critical, loc);
3518 __kmp_release_user_lock_with_checks(lck, global_tid);
3523static __forceinline
int
3524__kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
3529 if (th->th.th_teams_microtask) {
3530 *team_p = team = th->th.th_team;
3531 if (team->t.t_level == th->th.th_teams_level) {
3533 KMP_DEBUG_ASSERT(!th->th.th_info.ds.ds_tid);
3535 th->th.th_info.ds.ds_tid = team->t.t_master_tid;
3536 th->th.th_team = team->t.t_parent;
3537 th->th.th_team_nproc = th->th.th_team->t.t_nproc;
3538 th->th.th_task_team = th->th.th_team->t.t_task_team[0];
3539 *task_state = th->th.th_task_state;
3540 th->th.th_task_state = 0;
3548static __forceinline
void
3549__kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team,
int task_state) {
3551 th->th.th_info.ds.ds_tid = 0;
3552 th->th.th_team = team;
3553 th->th.th_team_nproc = team->t.t_nproc;
3554 th->th.th_task_team = team->t.t_task_team[task_state];
3555 __kmp_type_convert(task_state, &(th->th.th_task_state));
3576 size_t reduce_size,
void *reduce_data,
3577 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3578 kmp_critical_name *lck) {
3582 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3585 int teams_swapped = 0, task_state;
3586 KA_TRACE(10, (
"__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
3587 __kmp_assert_valid_gtid(global_tid);
3595 if (!TCR_4(__kmp_init_parallel))
3596 __kmp_parallel_initialize();
3598 __kmp_resume_if_soft_paused();
3601#if KMP_USE_DYNAMIC_LOCK
3602 if (__kmp_env_consistency_check)
3603 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3605 if (__kmp_env_consistency_check)
3606 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3609 th = __kmp_thread_from_gtid(global_tid);
3610 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3628 packed_reduction_method = __kmp_determine_reduction_method(
3629 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3630 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3632 OMPT_REDUCTION_DECL(th, global_tid);
3633 if (packed_reduction_method == critical_reduce_block) {
3635 OMPT_REDUCTION_BEGIN;
3637 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3640 }
else if (packed_reduction_method == empty_reduce_block) {
3642 OMPT_REDUCTION_BEGIN;
3648 }
else if (packed_reduction_method == atomic_reduce_block) {
3658 if (__kmp_env_consistency_check)
3659 __kmp_pop_sync(global_tid, ct_reduce, loc);
3661 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3662 tree_reduce_block)) {
3682 ompt_frame_t *ompt_frame;
3683 if (ompt_enabled.enabled) {
3684 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3685 if (ompt_frame->enter_frame.ptr == NULL)
3686 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3688 OMPT_STORE_RETURN_ADDRESS(global_tid);
3691 __kmp_threads[global_tid]->th.th_ident = loc;
3694 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3695 global_tid, FALSE, reduce_size, reduce_data, reduce_func);
3696 retval = (retval != 0) ? (0) : (1);
3697#if OMPT_SUPPORT && OMPT_OPTIONAL
3698 if (ompt_enabled.enabled) {
3699 ompt_frame->enter_frame = ompt_data_none;
3705 if (__kmp_env_consistency_check) {
3707 __kmp_pop_sync(global_tid, ct_reduce, loc);
3716 if (teams_swapped) {
3717 __kmp_restore_swapped_teams(th, team, task_state);
3721 (
"__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
3722 global_tid, packed_reduction_method, retval));
3736 kmp_critical_name *lck) {
3738 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3740 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() enter: called T#%d\n", global_tid));
3741 __kmp_assert_valid_gtid(global_tid);
3743 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3745 OMPT_REDUCTION_DECL(__kmp_thread_from_gtid(global_tid), global_tid);
3747 if (packed_reduction_method == critical_reduce_block) {
3749 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3752 }
else if (packed_reduction_method == empty_reduce_block) {
3759 }
else if (packed_reduction_method == atomic_reduce_block) {
3766 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3767 tree_reduce_block)) {
3778 if (__kmp_env_consistency_check)
3779 __kmp_pop_sync(global_tid, ct_reduce, loc);
3781 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() exit: called T#%d: method %08x\n",
3782 global_tid, packed_reduction_method));
3805 size_t reduce_size,
void *reduce_data,
3806 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3807 kmp_critical_name *lck) {
3810 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3813 int teams_swapped = 0, task_state;
3815 KA_TRACE(10, (
"__kmpc_reduce() enter: called T#%d\n", global_tid));
3816 __kmp_assert_valid_gtid(global_tid);
3824 if (!TCR_4(__kmp_init_parallel))
3825 __kmp_parallel_initialize();
3827 __kmp_resume_if_soft_paused();
3830#if KMP_USE_DYNAMIC_LOCK
3831 if (__kmp_env_consistency_check)
3832 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3834 if (__kmp_env_consistency_check)
3835 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3838 th = __kmp_thread_from_gtid(global_tid);
3839 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3841 packed_reduction_method = __kmp_determine_reduction_method(
3842 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3843 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3845 OMPT_REDUCTION_DECL(th, global_tid);
3847 if (packed_reduction_method == critical_reduce_block) {
3849 OMPT_REDUCTION_BEGIN;
3850 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3853 }
else if (packed_reduction_method == empty_reduce_block) {
3855 OMPT_REDUCTION_BEGIN;
3860 }
else if (packed_reduction_method == atomic_reduce_block) {
3864 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3865 tree_reduce_block)) {
3871 ompt_frame_t *ompt_frame;
3872 if (ompt_enabled.enabled) {
3873 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3874 if (ompt_frame->enter_frame.ptr == NULL)
3875 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3877 OMPT_STORE_RETURN_ADDRESS(global_tid);
3880 __kmp_threads[global_tid]->th.th_ident =
3884 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3885 global_tid, TRUE, reduce_size, reduce_data, reduce_func);
3886 retval = (retval != 0) ? (0) : (1);
3887#if OMPT_SUPPORT && OMPT_OPTIONAL
3888 if (ompt_enabled.enabled) {
3889 ompt_frame->enter_frame = ompt_data_none;
3895 if (__kmp_env_consistency_check) {
3897 __kmp_pop_sync(global_tid, ct_reduce, loc);
3906 if (teams_swapped) {
3907 __kmp_restore_swapped_teams(th, team, task_state);
3911 (
"__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
3912 global_tid, packed_reduction_method, retval));
3927 kmp_critical_name *lck) {
3929 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3932 int teams_swapped = 0, task_state;
3934 KA_TRACE(10, (
"__kmpc_end_reduce() enter: called T#%d\n", global_tid));
3935 __kmp_assert_valid_gtid(global_tid);
3937 th = __kmp_thread_from_gtid(global_tid);
3938 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3940 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3944 OMPT_REDUCTION_DECL(th, global_tid);
3946 if (packed_reduction_method == critical_reduce_block) {
3947 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3953 ompt_frame_t *ompt_frame;
3954 if (ompt_enabled.enabled) {
3955 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3956 if (ompt_frame->enter_frame.ptr == NULL)
3957 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3959 OMPT_STORE_RETURN_ADDRESS(global_tid);
3962 __kmp_threads[global_tid]->th.th_ident = loc;
3964 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3965#if OMPT_SUPPORT && OMPT_OPTIONAL
3966 if (ompt_enabled.enabled) {
3967 ompt_frame->enter_frame = ompt_data_none;
3971 }
else if (packed_reduction_method == empty_reduce_block) {
3979 ompt_frame_t *ompt_frame;
3980 if (ompt_enabled.enabled) {
3981 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3982 if (ompt_frame->enter_frame.ptr == NULL)
3983 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3985 OMPT_STORE_RETURN_ADDRESS(global_tid);
3988 __kmp_threads[global_tid]->th.th_ident = loc;
3990 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3991#if OMPT_SUPPORT && OMPT_OPTIONAL
3992 if (ompt_enabled.enabled) {
3993 ompt_frame->enter_frame = ompt_data_none;
3997 }
else if (packed_reduction_method == atomic_reduce_block) {
4000 ompt_frame_t *ompt_frame;
4001 if (ompt_enabled.enabled) {
4002 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
4003 if (ompt_frame->enter_frame.ptr == NULL)
4004 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
4006 OMPT_STORE_RETURN_ADDRESS(global_tid);
4010 __kmp_threads[global_tid]->th.th_ident = loc;
4012 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
4013#if OMPT_SUPPORT && OMPT_OPTIONAL
4014 if (ompt_enabled.enabled) {
4015 ompt_frame->enter_frame = ompt_data_none;
4019 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
4020 tree_reduce_block)) {
4023 __kmp_end_split_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
4031 if (teams_swapped) {
4032 __kmp_restore_swapped_teams(th, team, task_state);
4035 if (__kmp_env_consistency_check)
4036 __kmp_pop_sync(global_tid, ct_reduce, loc);
4038 KA_TRACE(10, (
"__kmpc_end_reduce() exit: called T#%d: method %08x\n",
4039 global_tid, packed_reduction_method));
4044#undef __KMP_GET_REDUCTION_METHOD
4045#undef __KMP_SET_REDUCTION_METHOD
4049kmp_uint64 __kmpc_get_taskid() {
4054 gtid = __kmp_get_gtid();
4058 thread = __kmp_thread_from_gtid(gtid);
4059 return thread->th.th_current_task->td_task_id;
4063kmp_uint64 __kmpc_get_parent_taskid() {
4067 kmp_taskdata_t *parent_task;
4069 gtid = __kmp_get_gtid();
4073 thread = __kmp_thread_from_gtid(gtid);
4074 parent_task = thread->th.th_current_task->td_parent;
4075 return (parent_task == NULL ? 0 : parent_task->td_task_id);
4091 const struct kmp_dim *dims) {
4092 __kmp_assert_valid_gtid(gtid);
4094 kmp_int64 last, trace_count;
4095 kmp_info_t *th = __kmp_threads[gtid];
4096 kmp_team_t *team = th->th.th_team;
4098 kmp_disp_t *pr_buf = th->th.th_dispatch;
4099 dispatch_shared_info_t *sh_buf;
4103 (
"__kmpc_doacross_init() enter: called T#%d, num dims %d, active %d\n",
4104 gtid, num_dims, !team->t.t_serialized));
4105 KMP_DEBUG_ASSERT(dims != NULL);
4106 KMP_DEBUG_ASSERT(num_dims > 0);
4108 if (team->t.t_serialized) {
4109 KA_TRACE(20, (
"__kmpc_doacross_init() exit: serialized team\n"));
4112 KMP_DEBUG_ASSERT(team->t.t_nproc > 1);
4113 idx = pr_buf->th_doacross_buf_idx++;
4115 sh_buf = &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4118 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info == NULL);
4119 pr_buf->th_doacross_info = (kmp_int64 *)__kmp_thread_malloc(
4120 th,
sizeof(kmp_int64) * (4 * num_dims + 1));
4121 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4122 pr_buf->th_doacross_info[0] =
4123 (kmp_int64)num_dims;
4126 pr_buf->th_doacross_info[1] = (kmp_int64)&sh_buf->doacross_num_done;
4127 pr_buf->th_doacross_info[2] = dims[0].lo;
4128 pr_buf->th_doacross_info[3] = dims[0].up;
4129 pr_buf->th_doacross_info[4] = dims[0].st;
4131 for (j = 1; j < num_dims; ++j) {
4134 if (dims[j].st == 1) {
4136 range_length = dims[j].up - dims[j].lo + 1;
4138 if (dims[j].st > 0) {
4139 KMP_DEBUG_ASSERT(dims[j].up > dims[j].lo);
4140 range_length = (kmp_uint64)(dims[j].up - dims[j].lo) / dims[j].st + 1;
4142 KMP_DEBUG_ASSERT(dims[j].lo > dims[j].up);
4144 (kmp_uint64)(dims[j].lo - dims[j].up) / (-dims[j].st) + 1;
4147 pr_buf->th_doacross_info[last++] = range_length;
4148 pr_buf->th_doacross_info[last++] = dims[j].lo;
4149 pr_buf->th_doacross_info[last++] = dims[j].up;
4150 pr_buf->th_doacross_info[last++] = dims[j].st;
4155 if (dims[0].st == 1) {
4156 trace_count = dims[0].up - dims[0].lo + 1;
4157 }
else if (dims[0].st > 0) {
4158 KMP_DEBUG_ASSERT(dims[0].up > dims[0].lo);
4159 trace_count = (kmp_uint64)(dims[0].up - dims[0].lo) / dims[0].st + 1;
4161 KMP_DEBUG_ASSERT(dims[0].lo > dims[0].up);
4162 trace_count = (kmp_uint64)(dims[0].lo - dims[0].up) / (-dims[0].st) + 1;
4164 for (j = 1; j < num_dims; ++j) {
4165 trace_count *= pr_buf->th_doacross_info[4 * j + 1];
4167 KMP_DEBUG_ASSERT(trace_count > 0);
4171 if (idx != sh_buf->doacross_buf_idx) {
4173 __kmp_wait_4((
volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
4180 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET32(
4181 (
volatile kmp_int32 *)&sh_buf->doacross_flags, NULL, 1);
4183 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET64(
4184 (
volatile kmp_int64 *)&sh_buf->doacross_flags, NULL, 1LL);
4186 if (flags == NULL) {
4189 (size_t)trace_count / 8 + 8;
4190 flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
4192 sh_buf->doacross_flags = flags;
4193 }
else if (flags == (kmp_uint32 *)1) {
4196 while (*(
volatile kmp_int32 *)&sh_buf->doacross_flags == 1)
4198 while (*(
volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
4205 KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1);
4206 pr_buf->th_doacross_flags =
4207 sh_buf->doacross_flags;
4209 KA_TRACE(20, (
"__kmpc_doacross_init() exit: T#%d\n", gtid));
4212void __kmpc_doacross_wait(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4213 __kmp_assert_valid_gtid(gtid);
4217 kmp_int64 iter_number;
4218 kmp_info_t *th = __kmp_threads[gtid];
4219 kmp_team_t *team = th->th.th_team;
4221 kmp_int64 lo, up, st;
4223 KA_TRACE(20, (
"__kmpc_doacross_wait() enter: called T#%d\n", gtid));
4224 if (team->t.t_serialized) {
4225 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: serialized team\n"));
4230 pr_buf = th->th.th_dispatch;
4231 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4232 num_dims = (size_t)pr_buf->th_doacross_info[0];
4233 lo = pr_buf->th_doacross_info[2];
4234 up = pr_buf->th_doacross_info[3];
4235 st = pr_buf->th_doacross_info[4];
4236#if OMPT_SUPPORT && OMPT_OPTIONAL
4240 if (vec[0] < lo || vec[0] > up) {
4241 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4242 "bounds [%lld,%lld]\n",
4243 gtid, vec[0], lo, up));
4246 iter_number = vec[0] - lo;
4247 }
else if (st > 0) {
4248 if (vec[0] < lo || vec[0] > up) {
4249 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4250 "bounds [%lld,%lld]\n",
4251 gtid, vec[0], lo, up));
4254 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4256 if (vec[0] > lo || vec[0] < up) {
4257 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4258 "bounds [%lld,%lld]\n",
4259 gtid, vec[0], lo, up));
4262 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4264#if OMPT_SUPPORT && OMPT_OPTIONAL
4265 deps[0].variable.value = iter_number;
4266 deps[0].dependence_type = ompt_dependence_type_sink;
4268 for (i = 1; i < num_dims; ++i) {
4271 ln = pr_buf->th_doacross_info[j + 1];
4272 lo = pr_buf->th_doacross_info[j + 2];
4273 up = pr_buf->th_doacross_info[j + 3];
4274 st = pr_buf->th_doacross_info[j + 4];
4276 if (vec[i] < lo || vec[i] > up) {
4277 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4278 "bounds [%lld,%lld]\n",
4279 gtid, vec[i], lo, up));
4283 }
else if (st > 0) {
4284 if (vec[i] < lo || vec[i] > up) {
4285 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4286 "bounds [%lld,%lld]\n",
4287 gtid, vec[i], lo, up));
4290 iter = (kmp_uint64)(vec[i] - lo) / st;
4292 if (vec[i] > lo || vec[i] < up) {
4293 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4294 "bounds [%lld,%lld]\n",
4295 gtid, vec[i], lo, up));
4298 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4300 iter_number = iter + ln * iter_number;
4301#if OMPT_SUPPORT && OMPT_OPTIONAL
4302 deps[i].variable.value = iter;
4303 deps[i].dependence_type = ompt_dependence_type_sink;
4306 shft = iter_number % 32;
4309 while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
4313#if OMPT_SUPPORT && OMPT_OPTIONAL
4314 if (ompt_enabled.ompt_callback_dependences) {
4315 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4316 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4320 (
"__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
4321 gtid, (iter_number << 5) + shft));
4324void __kmpc_doacross_post(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4325 __kmp_assert_valid_gtid(gtid);
4329 kmp_int64 iter_number;
4330 kmp_info_t *th = __kmp_threads[gtid];
4331 kmp_team_t *team = th->th.th_team;
4335 KA_TRACE(20, (
"__kmpc_doacross_post() enter: called T#%d\n", gtid));
4336 if (team->t.t_serialized) {
4337 KA_TRACE(20, (
"__kmpc_doacross_post() exit: serialized team\n"));
4343 pr_buf = th->th.th_dispatch;
4344 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4345 num_dims = (size_t)pr_buf->th_doacross_info[0];
4346 lo = pr_buf->th_doacross_info[2];
4347 st = pr_buf->th_doacross_info[4];
4348#if OMPT_SUPPORT && OMPT_OPTIONAL
4352 iter_number = vec[0] - lo;
4353 }
else if (st > 0) {
4354 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4356 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4358#if OMPT_SUPPORT && OMPT_OPTIONAL
4359 deps[0].variable.value = iter_number;
4360 deps[0].dependence_type = ompt_dependence_type_source;
4362 for (i = 1; i < num_dims; ++i) {
4365 ln = pr_buf->th_doacross_info[j + 1];
4366 lo = pr_buf->th_doacross_info[j + 2];
4367 st = pr_buf->th_doacross_info[j + 4];
4370 }
else if (st > 0) {
4371 iter = (kmp_uint64)(vec[i] - lo) / st;
4373 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4375 iter_number = iter + ln * iter_number;
4376#if OMPT_SUPPORT && OMPT_OPTIONAL
4377 deps[i].variable.value = iter;
4378 deps[i].dependence_type = ompt_dependence_type_source;
4381#if OMPT_SUPPORT && OMPT_OPTIONAL
4382 if (ompt_enabled.ompt_callback_dependences) {
4383 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4384 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4387 shft = iter_number % 32;
4391 if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
4392 KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
4393 KA_TRACE(20, (
"__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
4394 (iter_number << 5) + shft));
4397void __kmpc_doacross_fini(
ident_t *loc,
int gtid) {
4398 __kmp_assert_valid_gtid(gtid);
4400 kmp_info_t *th = __kmp_threads[gtid];
4401 kmp_team_t *team = th->th.th_team;
4402 kmp_disp_t *pr_buf = th->th.th_dispatch;
4404 KA_TRACE(20, (
"__kmpc_doacross_fini() enter: called T#%d\n", gtid));
4405 if (team->t.t_serialized) {
4406 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: serialized team %p\n", team));
4410 KMP_TEST_THEN_INC32((kmp_uintptr_t)(pr_buf->th_doacross_info[1])) + 1;
4411 if (num_done == th->th.th_team_nproc) {
4413 int idx = pr_buf->th_doacross_buf_idx - 1;
4414 dispatch_shared_info_t *sh_buf =
4415 &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4416 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info[1] ==
4417 (kmp_int64)&sh_buf->doacross_num_done);
4418 KMP_DEBUG_ASSERT(num_done == sh_buf->doacross_num_done);
4419 KMP_DEBUG_ASSERT(idx == sh_buf->doacross_buf_idx);
4420 __kmp_thread_free(th, CCAST(kmp_uint32 *, sh_buf->doacross_flags));
4421 sh_buf->doacross_flags = NULL;
4422 sh_buf->doacross_num_done = 0;
4423 sh_buf->doacross_buf_idx +=
4424 __kmp_dispatch_num_buffers;
4427 pr_buf->th_doacross_flags = NULL;
4428 __kmp_thread_free(th, (
void *)pr_buf->th_doacross_info);
4429 pr_buf->th_doacross_info = NULL;
4430 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: T#%d\n", gtid));
4434void *omp_alloc(
size_t size, omp_allocator_handle_t allocator) {
4435 return __kmp_alloc(__kmp_entry_gtid(), 0, size, allocator);
4438void *omp_aligned_alloc(
size_t align,
size_t size,
4439 omp_allocator_handle_t allocator) {
4440 return __kmp_alloc(__kmp_entry_gtid(), align, size, allocator);
4443void *omp_calloc(
size_t nmemb,
size_t size, omp_allocator_handle_t allocator) {
4444 return __kmp_calloc(__kmp_entry_gtid(), 0, nmemb, size, allocator);
4447void *omp_aligned_calloc(
size_t align,
size_t nmemb,
size_t size,
4448 omp_allocator_handle_t allocator) {
4449 return __kmp_calloc(__kmp_entry_gtid(), align, nmemb, size, allocator);
4452void *omp_realloc(
void *ptr,
size_t size, omp_allocator_handle_t allocator,
4453 omp_allocator_handle_t free_allocator) {
4454 return __kmp_realloc(__kmp_entry_gtid(), ptr, size, allocator,
4458void omp_free(
void *ptr, omp_allocator_handle_t allocator) {
4459 ___kmpc_free(__kmp_entry_gtid(), ptr, allocator);
4463int __kmpc_get_target_offload(
void) {
4464 if (!__kmp_init_serial) {
4465 __kmp_serial_initialize();
4467 return __kmp_target_offload;
4470int __kmpc_pause_resource(kmp_pause_status_t level) {
4471 if (!__kmp_init_serial) {
4474 return __kmp_pause_resource(level);
4477void __kmpc_error(
ident_t *loc,
int severity,
const char *message) {
4478 if (!__kmp_init_serial)
4479 __kmp_serial_initialize();
4481 KMP_ASSERT(severity == severity_warning || severity == severity_fatal);
4484 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_error) {
4485 ompt_callbacks.ompt_callback(ompt_callback_error)(
4486 (ompt_severity_t)severity, message, KMP_STRLEN(message),
4487 OMPT_GET_RETURN_ADDRESS(0));
4493 kmp_str_loc_t str_loc = __kmp_str_loc_init(loc->
psource,
false);
4495 __kmp_str_format(
"%s:%d:%d", str_loc.file, str_loc.line, str_loc.col);
4496 __kmp_str_loc_free(&str_loc);
4498 src_loc = __kmp_str_format(
"unknown");
4501 if (severity == severity_warning)
4502 KMP_WARNING(UserDirectedWarning, src_loc, message);
4504 KMP_FATAL(UserDirectedError, src_loc, message);
4506 __kmp_str_free(&src_loc);
4510void __kmpc_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4512#if OMPT_SUPPORT && OMPT_OPTIONAL
4513 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4514 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4515 int tid = __kmp_tid_from_gtid(gtid);
4516 ompt_callbacks.ompt_callback(ompt_callback_work)(
4517 ompt_work_scope, ompt_scope_begin,
4518 &(team->t.ompt_team_info.parallel_data),
4519 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4520 OMPT_GET_RETURN_ADDRESS(0));
4526void __kmpc_end_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4528#if OMPT_SUPPORT && OMPT_OPTIONAL
4529 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4530 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4531 int tid = __kmp_tid_from_gtid(gtid);
4532 ompt_callbacks.ompt_callback(ompt_callback_work)(
4533 ompt_work_scope, ompt_scope_end,
4534 &(team->t.ompt_team_info.parallel_data),
4535 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4536 OMPT_GET_RETURN_ADDRESS(0));
4541#ifdef KMP_USE_VERSION_SYMBOLS
4550#ifdef omp_set_affinity_format
4551#undef omp_set_affinity_format
4553#ifdef omp_get_affinity_format
4554#undef omp_get_affinity_format
4556#ifdef omp_display_affinity
4557#undef omp_display_affinity
4559#ifdef omp_capture_affinity
4560#undef omp_capture_affinity
4562KMP_VERSION_OMPC_SYMBOL(ompc_set_affinity_format, omp_set_affinity_format, 50,
4564KMP_VERSION_OMPC_SYMBOL(ompc_get_affinity_format, omp_get_affinity_format, 50,
4566KMP_VERSION_OMPC_SYMBOL(ompc_display_affinity, omp_display_affinity, 50,
4568KMP_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)