21#include "ompt-specific.h"
23#define MAX_MESSAGE 512
39 if ((env = getenv(
"KMP_INITIAL_THREAD_BIND")) != NULL &&
40 __kmp_str_match_true(env)) {
41 __kmp_middle_initialize();
42 __kmp_assign_root_init_mask();
43 KC_TRACE(10, (
"__kmpc_begin: middle initialization called\n"));
44 }
else if (__kmp_ignore_mppbeg() == FALSE) {
46 __kmp_internal_begin();
47 KC_TRACE(10, (
"__kmpc_begin: called\n"));
65 if (__kmp_ignore_mppend() == FALSE) {
66 KC_TRACE(10, (
"__kmpc_end: called\n"));
67 KA_TRACE(30, (
"__kmpc_end\n"));
69 __kmp_internal_end_thread(-1);
71#if KMP_OS_WINDOWS && OMPT_SUPPORT
76 if (ompt_enabled.enabled)
77 __kmp_internal_end_library(__kmp_gtid_get_specific());
100 kmp_int32 gtid = __kmp_entry_gtid();
102 KC_TRACE(10, (
"__kmpc_global_thread_num: T#%d\n", gtid));
123 (
"__kmpc_global_num_threads: num_threads = %d\n", __kmp_all_nth));
125 return TCR_4(__kmp_all_nth);
135 KC_TRACE(10, (
"__kmpc_bound_thread_num: called\n"));
136 return __kmp_tid_from_gtid(__kmp_entry_gtid());
145 KC_TRACE(10, (
"__kmpc_bound_num_threads: called\n"));
147 return __kmp_entry_thread()->th.th_team->t.t_nproc;
167 if (__kmp_par_range == 0) {
174 semi2 = strchr(semi2,
';');
178 semi2 = strchr(semi2 + 1,
';');
182 if (__kmp_par_range_filename[0]) {
183 const char *name = semi2 - 1;
184 while ((name > loc->
psource) && (*name !=
'/') && (*name !=
';')) {
187 if ((*name ==
'/') || (*name ==
';')) {
190 if (strncmp(__kmp_par_range_filename, name, semi2 - name)) {
191 return __kmp_par_range < 0;
194 semi3 = strchr(semi2 + 1,
';');
195 if (__kmp_par_range_routine[0]) {
196 if ((semi3 != NULL) && (semi3 > semi2) &&
197 (strncmp(__kmp_par_range_routine, semi2 + 1, semi3 - semi2 - 1))) {
198 return __kmp_par_range < 0;
201 if (KMP_SSCANF(semi3 + 1,
"%d", &line_no) == 1) {
202 if ((line_no >= __kmp_par_range_lb) && (line_no <= __kmp_par_range_ub)) {
203 return __kmp_par_range > 0;
205 return __kmp_par_range < 0;
219 return __kmp_entry_thread()->th.th_root->r.r_active;
232 kmp_int32 num_threads) {
233 KA_TRACE(20, (
"__kmpc_push_num_threads: enter T#%d num_threads=%d\n",
234 global_tid, num_threads));
235 __kmp_assert_valid_gtid(global_tid);
236 __kmp_push_num_threads(loc, global_tid, num_threads);
239void __kmpc_pop_num_threads(
ident_t *loc, kmp_int32 global_tid) {
240 KA_TRACE(20, (
"__kmpc_pop_num_threads: enter\n"));
244void __kmpc_push_proc_bind(
ident_t *loc, kmp_int32 global_tid,
245 kmp_int32 proc_bind) {
246 KA_TRACE(20, (
"__kmpc_push_proc_bind: enter T#%d proc_bind=%d\n", global_tid,
248 __kmp_assert_valid_gtid(global_tid);
249 __kmp_push_proc_bind(loc, global_tid, (kmp_proc_bind_t)proc_bind);
263 int gtid = __kmp_entry_gtid();
265#if (KMP_STATS_ENABLED)
269 if (previous_state == stats_state_e::SERIAL_REGION) {
270 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_parallel_overhead);
272 KMP_PUSH_PARTITIONED_TIMER(OMP_parallel_overhead);
285 va_start(ap, microtask);
288 ompt_frame_t *ompt_frame;
289 if (ompt_enabled.enabled) {
290 kmp_info_t *master_th = __kmp_threads[gtid];
291 ompt_frame = &master_th->th.th_current_task->ompt_task_info.frame;
292 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
294 OMPT_STORE_RETURN_ADDRESS(gtid);
300 __kmp_fork_call(loc, gtid, fork_context_intel, argc,
301 VOLATILE_CAST(microtask_t) microtask,
302 VOLATILE_CAST(launch_t) __kmp_invoke_task_func,
307 __kmp_join_call(loc, gtid
317 if (ompt_enabled.enabled) {
318 ompt_frame->enter_frame = ompt_data_none;
324 if (previous_state == stats_state_e::SERIAL_REGION) {
325 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
326 KMP_SET_THREAD_STATE(previous_state);
328 KMP_POP_PARTITIONED_TIMER();
345 kmp_int32 num_teams, kmp_int32 num_threads) {
347 (
"__kmpc_push_num_teams: enter T#%d num_teams=%d num_threads=%d\n",
348 global_tid, num_teams, num_threads));
349 __kmp_assert_valid_gtid(global_tid);
350 __kmp_push_num_teams(loc, global_tid, num_teams, num_threads);
370 kmp_int32 num_teams_lb, kmp_int32 num_teams_ub,
371 kmp_int32 num_threads) {
372 KA_TRACE(20, (
"__kmpc_push_num_teams_51: enter T#%d num_teams_lb=%d"
373 " num_teams_ub=%d num_threads=%d\n",
374 global_tid, num_teams_lb, num_teams_ub, num_threads));
375 __kmp_assert_valid_gtid(global_tid);
376 __kmp_push_num_teams_51(loc, global_tid, num_teams_lb, num_teams_ub,
392 int gtid = __kmp_entry_gtid();
393 kmp_info_t *this_thr = __kmp_threads[gtid];
395 va_start(ap, microtask);
400 if (previous_state == stats_state_e::SERIAL_REGION) {
401 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_teams_overhead);
403 KMP_PUSH_PARTITIONED_TIMER(OMP_teams_overhead);
408 this_thr->th.th_teams_microtask = microtask;
409 this_thr->th.th_teams_level =
410 this_thr->th.th_team->t.t_level;
413 kmp_team_t *parent_team = this_thr->th.th_team;
414 int tid = __kmp_tid_from_gtid(gtid);
415 if (ompt_enabled.enabled) {
416 parent_team->t.t_implicit_task_taskdata[tid]
417 .ompt_task_info.frame.enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
419 OMPT_STORE_RETURN_ADDRESS(gtid);
424 if (this_thr->th.th_teams_size.nteams == 0) {
425 __kmp_push_num_teams(loc, gtid, 0, 0);
427 KMP_DEBUG_ASSERT(this_thr->th.th_set_nproc >= 1);
428 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nteams >= 1);
429 KMP_DEBUG_ASSERT(this_thr->th.th_teams_size.nth >= 1);
432 loc, gtid, fork_context_intel, argc,
433 VOLATILE_CAST(microtask_t) __kmp_teams_master,
434 VOLATILE_CAST(launch_t) __kmp_invoke_teams_master, kmp_va_addr_of(ap));
435 __kmp_join_call(loc, gtid
443 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
444 kmp_cg_root_t *tmp = this_thr->th.th_cg_roots;
445 this_thr->th.th_cg_roots = tmp->up;
446 KA_TRACE(100, (
"__kmpc_fork_teams: Thread %p popping node %p and moving up"
447 " to node %p. cg_nthreads was %d\n",
448 this_thr, tmp, this_thr->th.th_cg_roots, tmp->cg_nthreads));
449 KMP_DEBUG_ASSERT(tmp->cg_nthreads);
450 int i = tmp->cg_nthreads--;
455 KMP_DEBUG_ASSERT(this_thr->th.th_cg_roots);
456 this_thr->th.th_current_task->td_icvs.thread_limit =
457 this_thr->th.th_cg_roots->cg_thread_limit;
459 this_thr->th.th_teams_microtask = NULL;
460 this_thr->th.th_teams_level = 0;
461 *(kmp_int64 *)(&this_thr->th.th_teams_size) = 0L;
464 if (previous_state == stats_state_e::SERIAL_REGION) {
465 KMP_EXCHANGE_PARTITIONED_TIMER(OMP_serial);
466 KMP_SET_THREAD_STATE(previous_state);
468 KMP_POP_PARTITIONED_TIMER();
477int __kmpc_invoke_task_func(
int gtid) {
return __kmp_invoke_task_func(gtid); }
495 __kmp_assert_valid_gtid(global_tid);
497 OMPT_STORE_RETURN_ADDRESS(global_tid);
499 __kmp_serialized_parallel(loc, global_tid);
510 kmp_internal_control_t *top;
511 kmp_info_t *this_thr;
512 kmp_team_t *serial_team;
515 (
"__kmpc_end_serialized_parallel: called by T#%d\n", global_tid));
523 __kmp_assert_valid_gtid(global_tid);
524 if (!TCR_4(__kmp_init_parallel))
525 __kmp_parallel_initialize();
527 __kmp_resume_if_soft_paused();
529 this_thr = __kmp_threads[global_tid];
530 serial_team = this_thr->th.th_serial_team;
532 kmp_task_team_t *task_team = this_thr->th.th_task_team;
534 if (task_team != NULL && (task_team->tt.tt_found_proxy_tasks ||
535 task_team->tt.tt_hidden_helper_task_encountered))
536 __kmp_task_team_wait(this_thr, serial_team USE_ITT_BUILD_ARG(NULL));
539 KMP_DEBUG_ASSERT(serial_team);
540 KMP_ASSERT(serial_team->t.t_serialized);
541 KMP_DEBUG_ASSERT(this_thr->th.th_team == serial_team);
542 KMP_DEBUG_ASSERT(serial_team != this_thr->th.th_root->r.r_root_team);
543 KMP_DEBUG_ASSERT(serial_team->t.t_threads);
544 KMP_DEBUG_ASSERT(serial_team->t.t_threads[0] == this_thr);
547 if (ompt_enabled.enabled &&
548 this_thr->th.ompt_thread_info.state != ompt_state_overhead) {
549 OMPT_CUR_TASK_INFO(this_thr)->frame.exit_frame = ompt_data_none;
550 if (ompt_enabled.ompt_callback_implicit_task) {
551 ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
552 ompt_scope_end, NULL, OMPT_CUR_TASK_DATA(this_thr), 1,
553 OMPT_CUR_TASK_INFO(this_thr)->thread_num, ompt_task_implicit);
557 ompt_data_t *parent_task_data;
558 __ompt_get_task_info_internal(1, NULL, &parent_task_data, NULL, NULL, NULL);
560 if (ompt_enabled.ompt_callback_parallel_end) {
561 ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
562 &(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
563 ompt_parallel_invoker_program | ompt_parallel_team,
564 OMPT_LOAD_RETURN_ADDRESS(global_tid));
566 __ompt_lw_taskteam_unlink(this_thr);
567 this_thr->th.ompt_thread_info.state = ompt_state_overhead;
573 top = serial_team->t.t_control_stack_top;
574 if (top && top->serial_nesting_level == serial_team->t.t_serialized) {
575 copy_icvs(&serial_team->t.t_threads[0]->th.th_current_task->td_icvs, top);
576 serial_team->t.t_control_stack_top = top->next;
581 KMP_DEBUG_ASSERT(serial_team->t.t_dispatch->th_disp_buffer);
583 dispatch_private_info_t *disp_buffer =
584 serial_team->t.t_dispatch->th_disp_buffer;
585 serial_team->t.t_dispatch->th_disp_buffer =
586 serial_team->t.t_dispatch->th_disp_buffer->next;
587 __kmp_free(disp_buffer);
589 this_thr->th.th_def_allocator = serial_team->t.t_def_allocator;
591 --serial_team->t.t_serialized;
592 if (serial_team->t.t_serialized == 0) {
596#if KMP_ARCH_X86 || KMP_ARCH_X86_64
597 if (__kmp_inherit_fp_control && serial_team->t.t_fp_control_saved) {
598 __kmp_clear_x87_fpu_status_word();
599 __kmp_load_x87_fpu_control_word(&serial_team->t.t_x87_fpu_control_word);
600 __kmp_load_mxcsr(&serial_team->t.t_mxcsr);
604 __kmp_pop_current_task_from_thread(this_thr);
606 if (ompd_state & OMPD_ENABLE_BP)
607 ompd_bp_parallel_end();
610 this_thr->th.th_team = serial_team->t.t_parent;
611 this_thr->th.th_info.ds.ds_tid = serial_team->t.t_master_tid;
614 this_thr->th.th_team_nproc = serial_team->t.t_parent->t.t_nproc;
615 this_thr->th.th_team_master =
616 serial_team->t.t_parent->t.t_threads[0];
617 this_thr->th.th_team_serialized = this_thr->th.th_team->t.t_serialized;
620 this_thr->th.th_dispatch =
621 &this_thr->th.th_team->t.t_dispatch[serial_team->t.t_master_tid];
623 KMP_ASSERT(this_thr->th.th_current_task->td_flags.executing == 0);
624 this_thr->th.th_current_task->td_flags.executing = 1;
626 if (__kmp_tasking_mode != tskm_immediate_exec) {
628 this_thr->th.th_task_team =
629 this_thr->th.th_team->t.t_task_team[this_thr->th.th_task_state];
631 (
"__kmpc_end_serialized_parallel: T#%d restoring task_team %p / "
633 global_tid, this_thr->th.th_task_team, this_thr->th.th_team));
635#if KMP_AFFINITY_SUPPORTED
636 if (this_thr->th.th_team->t.t_level == 0 && __kmp_affin_reset) {
637 __kmp_reset_root_init_mask(global_tid);
641 if (__kmp_tasking_mode != tskm_immediate_exec) {
642 KA_TRACE(20, (
"__kmpc_end_serialized_parallel: T#%d decreasing nesting "
643 "depth of serial team %p to %d\n",
644 global_tid, serial_team, serial_team->t.t_serialized));
648 serial_team->t.t_level--;
649 if (__kmp_env_consistency_check)
650 __kmp_pop_parallel(global_tid, NULL);
652 if (ompt_enabled.enabled)
653 this_thr->th.ompt_thread_info.state =
654 ((this_thr->th.th_team_serialized) ? ompt_state_work_serial
655 : ompt_state_work_parallel);
668 KC_TRACE(10, (
"__kmpc_flush: called\n"));
673#if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
687 if (!__kmp_cpuinfo.initialized) {
688 __kmp_query_cpuid(&__kmp_cpuinfo);
690 if (!__kmp_cpuinfo.flags.sse2) {
693#if KMP_COMPILER_ICC || KMP_COMPILER_ICX
695#elif KMP_COMPILER_MSVC
698 __sync_synchronize();
702#elif (KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || KMP_ARCH_MIPS64 || \
708#error Unknown or unsupported architecture
711#if OMPT_SUPPORT && OMPT_OPTIONAL
712 if (ompt_enabled.ompt_callback_flush) {
713 ompt_callbacks.ompt_callback(ompt_callback_flush)(
714 __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
729 KC_TRACE(10, (
"__kmpc_barrier: called T#%d\n", global_tid));
730 __kmp_assert_valid_gtid(global_tid);
732 if (!TCR_4(__kmp_init_parallel))
733 __kmp_parallel_initialize();
735 __kmp_resume_if_soft_paused();
737 if (__kmp_env_consistency_check) {
739 KMP_WARNING(ConstructIdentInvalid);
741 __kmp_check_barrier(global_tid, ct_barrier, loc);
745 ompt_frame_t *ompt_frame;
746 if (ompt_enabled.enabled) {
747 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
748 if (ompt_frame->enter_frame.ptr == NULL)
749 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
751 OMPT_STORE_RETURN_ADDRESS(global_tid);
753 __kmp_threads[global_tid]->th.th_ident = loc;
761 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
762#if OMPT_SUPPORT && OMPT_OPTIONAL
763 if (ompt_enabled.enabled) {
764 ompt_frame->enter_frame = ompt_data_none;
779 KC_TRACE(10, (
"__kmpc_master: called T#%d\n", global_tid));
780 __kmp_assert_valid_gtid(global_tid);
782 if (!TCR_4(__kmp_init_parallel))
783 __kmp_parallel_initialize();
785 __kmp_resume_if_soft_paused();
787 if (KMP_MASTER_GTID(global_tid)) {
789 KMP_PUSH_PARTITIONED_TIMER(OMP_master);
793#if OMPT_SUPPORT && OMPT_OPTIONAL
795 if (ompt_enabled.ompt_callback_masked) {
796 kmp_info_t *this_thr = __kmp_threads[global_tid];
797 kmp_team_t *team = this_thr->th.th_team;
799 int tid = __kmp_tid_from_gtid(global_tid);
800 ompt_callbacks.ompt_callback(ompt_callback_masked)(
801 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
802 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
803 OMPT_GET_RETURN_ADDRESS(0));
808 if (__kmp_env_consistency_check) {
809#if KMP_USE_DYNAMIC_LOCK
811 __kmp_push_sync(global_tid, ct_master, loc, NULL, 0);
813 __kmp_check_sync(global_tid, ct_master, loc, NULL, 0);
816 __kmp_push_sync(global_tid, ct_master, loc, NULL);
818 __kmp_check_sync(global_tid, ct_master, loc, NULL);
834 KC_TRACE(10, (
"__kmpc_end_master: called T#%d\n", global_tid));
835 __kmp_assert_valid_gtid(global_tid);
836 KMP_DEBUG_ASSERT(KMP_MASTER_GTID(global_tid));
837 KMP_POP_PARTITIONED_TIMER();
839#if OMPT_SUPPORT && OMPT_OPTIONAL
840 kmp_info_t *this_thr = __kmp_threads[global_tid];
841 kmp_team_t *team = this_thr->th.th_team;
842 if (ompt_enabled.ompt_callback_masked) {
843 int tid = __kmp_tid_from_gtid(global_tid);
844 ompt_callbacks.ompt_callback(ompt_callback_masked)(
845 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
846 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
847 OMPT_GET_RETURN_ADDRESS(0));
851 if (__kmp_env_consistency_check) {
852 if (KMP_MASTER_GTID(global_tid))
853 __kmp_pop_sync(global_tid, ct_master, loc);
868 KC_TRACE(10, (
"__kmpc_masked: called T#%d\n", global_tid));
869 __kmp_assert_valid_gtid(global_tid);
871 if (!TCR_4(__kmp_init_parallel))
872 __kmp_parallel_initialize();
874 __kmp_resume_if_soft_paused();
876 tid = __kmp_tid_from_gtid(global_tid);
879 KMP_PUSH_PARTITIONED_TIMER(OMP_masked);
883#if OMPT_SUPPORT && OMPT_OPTIONAL
885 if (ompt_enabled.ompt_callback_masked) {
886 kmp_info_t *this_thr = __kmp_threads[global_tid];
887 kmp_team_t *team = this_thr->th.th_team;
888 ompt_callbacks.ompt_callback(ompt_callback_masked)(
889 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
890 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
891 OMPT_GET_RETURN_ADDRESS(0));
896 if (__kmp_env_consistency_check) {
897#if KMP_USE_DYNAMIC_LOCK
899 __kmp_push_sync(global_tid, ct_masked, loc, NULL, 0);
901 __kmp_check_sync(global_tid, ct_masked, loc, NULL, 0);
904 __kmp_push_sync(global_tid, ct_masked, loc, NULL);
906 __kmp_check_sync(global_tid, ct_masked, loc, NULL);
922 KC_TRACE(10, (
"__kmpc_end_masked: called T#%d\n", global_tid));
923 __kmp_assert_valid_gtid(global_tid);
924 KMP_POP_PARTITIONED_TIMER();
926#if OMPT_SUPPORT && OMPT_OPTIONAL
927 kmp_info_t *this_thr = __kmp_threads[global_tid];
928 kmp_team_t *team = this_thr->th.th_team;
929 if (ompt_enabled.ompt_callback_masked) {
930 int tid = __kmp_tid_from_gtid(global_tid);
931 ompt_callbacks.ompt_callback(ompt_callback_masked)(
932 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
933 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
934 OMPT_GET_RETURN_ADDRESS(0));
938 if (__kmp_env_consistency_check) {
939 __kmp_pop_sync(global_tid, ct_masked, loc);
953 KMP_DEBUG_ASSERT(__kmp_init_serial);
955 KC_TRACE(10, (
"__kmpc_ordered: called T#%d\n", gtid));
956 __kmp_assert_valid_gtid(gtid);
958 if (!TCR_4(__kmp_init_parallel))
959 __kmp_parallel_initialize();
961 __kmp_resume_if_soft_paused();
964 __kmp_itt_ordered_prep(gtid);
968 th = __kmp_threads[gtid];
970#if OMPT_SUPPORT && OMPT_OPTIONAL
974 OMPT_STORE_RETURN_ADDRESS(gtid);
975 if (ompt_enabled.enabled) {
976 team = __kmp_team_from_gtid(gtid);
977 lck = (ompt_wait_id_t)(uintptr_t)&team->t.t_ordered.dt.t_value;
979 th->th.ompt_thread_info.wait_id = lck;
980 th->th.ompt_thread_info.state = ompt_state_wait_ordered;
983 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
984 if (ompt_enabled.ompt_callback_mutex_acquire) {
985 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
986 ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin, lck,
992 if (th->th.th_dispatch->th_deo_fcn != 0)
993 (*th->th.th_dispatch->th_deo_fcn)(>id, &cid, loc);
995 __kmp_parallel_deo(>id, &cid, loc);
997#if OMPT_SUPPORT && OMPT_OPTIONAL
998 if (ompt_enabled.enabled) {
1000 th->th.ompt_thread_info.state = ompt_state_work_parallel;
1001 th->th.ompt_thread_info.wait_id = 0;
1004 if (ompt_enabled.ompt_callback_mutex_acquired) {
1005 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1006 ompt_mutex_ordered, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1012 __kmp_itt_ordered_start(gtid);
1027 KC_TRACE(10, (
"__kmpc_end_ordered: called T#%d\n", gtid));
1028 __kmp_assert_valid_gtid(gtid);
1031 __kmp_itt_ordered_end(gtid);
1035 th = __kmp_threads[gtid];
1037 if (th->th.th_dispatch->th_dxo_fcn != 0)
1038 (*th->th.th_dispatch->th_dxo_fcn)(>id, &cid, loc);
1040 __kmp_parallel_dxo(>id, &cid, loc);
1042#if OMPT_SUPPORT && OMPT_OPTIONAL
1043 OMPT_STORE_RETURN_ADDRESS(gtid);
1044 if (ompt_enabled.ompt_callback_mutex_released) {
1045 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1047 (ompt_wait_id_t)(uintptr_t)&__kmp_team_from_gtid(gtid)
1048 ->t.t_ordered.dt.t_value,
1049 OMPT_LOAD_RETURN_ADDRESS(gtid));
1054#if KMP_USE_DYNAMIC_LOCK
1056static __forceinline
void
1057__kmp_init_indirect_csptr(kmp_critical_name *crit,
ident_t const *loc,
1058 kmp_int32 gtid, kmp_indirect_locktag_t tag) {
1062 kmp_indirect_lock_t **lck;
1063 lck = (kmp_indirect_lock_t **)crit;
1064 kmp_indirect_lock_t *ilk = __kmp_allocate_indirect_lock(&idx, gtid, tag);
1065 KMP_I_LOCK_FUNC(ilk, init)(ilk->lock);
1066 KMP_SET_I_LOCK_LOCATION(ilk, loc);
1067 KMP_SET_I_LOCK_FLAGS(ilk, kmp_lf_critical_section);
1069 (
"__kmp_init_indirect_csptr: initialized indirect lock #%d\n", tag));
1071 __kmp_itt_critical_creating(ilk->lock, loc);
1073 int status = KMP_COMPARE_AND_STORE_PTR(lck,
nullptr, ilk);
1076 __kmp_itt_critical_destroyed(ilk->lock);
1082 KMP_DEBUG_ASSERT(*lck != NULL);
1086#define KMP_ACQUIRE_TAS_LOCK(lock, gtid) \
1088 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1089 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1090 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1091 if (KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1092 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \
1094 KMP_FSYNC_PREPARE(l); \
1095 KMP_INIT_YIELD(spins); \
1096 kmp_backoff_t backoff = __kmp_spin_backoff_params; \
1098 if (TCR_4(__kmp_nth) > \
1099 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
1102 KMP_YIELD_SPIN(spins); \
1104 __kmp_spin_backoff(&backoff); \
1106 KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1107 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)); \
1109 KMP_FSYNC_ACQUIRED(l); \
1113#define KMP_TEST_TAS_LOCK(lock, gtid, rc) \
1115 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1116 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1117 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1118 rc = KMP_ATOMIC_LD_RLX(&l->lk.poll) == tas_free && \
1119 __kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy); \
1123#define KMP_RELEASE_TAS_LOCK(lock, gtid) \
1124 { KMP_ATOMIC_ST_REL(&((kmp_tas_lock_t *)lock)->lk.poll, KMP_LOCK_FREE(tas)); }
1128#include <sys/syscall.h>
1138#define KMP_ACQUIRE_FUTEX_LOCK(lock, gtid) \
1140 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1141 kmp_int32 gtid_code = (gtid + 1) << 1; \
1143 KMP_FSYNC_PREPARE(ftx); \
1144 kmp_int32 poll_val; \
1145 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( \
1146 &(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1147 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { \
1148 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; \
1150 if (!KMP_COMPARE_AND_STORE_RET32(&(ftx->lk.poll), poll_val, \
1152 KMP_LOCK_BUSY(1, futex))) { \
1155 poll_val |= KMP_LOCK_BUSY(1, futex); \
1158 if ((rc = syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAIT, poll_val, \
1159 NULL, NULL, 0)) != 0) { \
1164 KMP_FSYNC_ACQUIRED(ftx); \
1168#define KMP_TEST_FUTEX_LOCK(lock, gtid, rc) \
1170 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1171 if (KMP_COMPARE_AND_STORE_ACQ32(&(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1172 KMP_LOCK_BUSY(gtid + 1 << 1, futex))) { \
1173 KMP_FSYNC_ACQUIRED(ftx); \
1181#define KMP_RELEASE_FUTEX_LOCK(lock, gtid) \
1183 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1185 KMP_FSYNC_RELEASING(ftx); \
1186 kmp_int32 poll_val = \
1187 KMP_XCHG_FIXED32(&(ftx->lk.poll), KMP_LOCK_FREE(futex)); \
1188 if (KMP_LOCK_STRIP(poll_val) & 1) { \
1189 syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAKE, \
1190 KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \
1193 KMP_YIELD_OVERSUB(); \
1200static kmp_user_lock_p __kmp_get_critical_section_ptr(kmp_critical_name *crit,
1203 kmp_user_lock_p *lck_pp = (kmp_user_lock_p *)crit;
1206 kmp_user_lock_p lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1213 lck = __kmp_user_lock_allocate(&idx, gtid, kmp_lf_critical_section);
1214 __kmp_init_user_lock_with_checks(lck);
1215 __kmp_set_user_lock_location(lck, loc);
1217 __kmp_itt_critical_creating(lck);
1228 int status = KMP_COMPARE_AND_STORE_PTR(lck_pp, 0, lck);
1233 __kmp_itt_critical_destroyed(lck);
1237 __kmp_destroy_user_lock_with_checks(lck);
1238 __kmp_user_lock_free(&idx, gtid, lck);
1239 lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1240 KMP_DEBUG_ASSERT(lck != NULL);
1259 kmp_critical_name *crit) {
1260#if KMP_USE_DYNAMIC_LOCK
1261#if OMPT_SUPPORT && OMPT_OPTIONAL
1262 OMPT_STORE_RETURN_ADDRESS(global_tid);
1264 __kmpc_critical_with_hint(loc, global_tid, crit, omp_lock_hint_none);
1267#if OMPT_SUPPORT && OMPT_OPTIONAL
1268 ompt_state_t prev_state = ompt_state_undefined;
1269 ompt_thread_info_t ti;
1271 kmp_user_lock_p lck;
1273 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1274 __kmp_assert_valid_gtid(global_tid);
1278 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1279 KMP_CHECK_USER_LOCK_INIT();
1281 if ((__kmp_user_lock_kind == lk_tas) &&
1282 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1283 lck = (kmp_user_lock_p)crit;
1286 else if ((__kmp_user_lock_kind == lk_futex) &&
1287 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1288 lck = (kmp_user_lock_p)crit;
1292 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
1295 if (__kmp_env_consistency_check)
1296 __kmp_push_sync(global_tid, ct_critical, loc, lck);
1304 __kmp_itt_critical_acquiring(lck);
1306#if OMPT_SUPPORT && OMPT_OPTIONAL
1307 OMPT_STORE_RETURN_ADDRESS(gtid);
1308 void *codeptr_ra = NULL;
1309 if (ompt_enabled.enabled) {
1310 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1312 prev_state = ti.state;
1313 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1314 ti.state = ompt_state_wait_critical;
1317 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1318 if (ompt_enabled.ompt_callback_mutex_acquire) {
1319 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1320 ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
1321 (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1327 __kmp_acquire_user_lock_with_checks(lck, global_tid);
1330 __kmp_itt_critical_acquired(lck);
1332#if OMPT_SUPPORT && OMPT_OPTIONAL
1333 if (ompt_enabled.enabled) {
1335 ti.state = prev_state;
1339 if (ompt_enabled.ompt_callback_mutex_acquired) {
1340 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1341 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1345 KMP_POP_PARTITIONED_TIMER();
1347 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1348 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1352#if KMP_USE_DYNAMIC_LOCK
1355static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
1357#define KMP_TSX_LOCK(seq) lockseq_##seq
1359#define KMP_TSX_LOCK(seq) __kmp_user_lock_seq
1362#if KMP_ARCH_X86 || KMP_ARCH_X86_64
1363#define KMP_CPUINFO_RTM (__kmp_cpuinfo.flags.rtm)
1365#define KMP_CPUINFO_RTM 0
1369 if (hint & kmp_lock_hint_hle)
1370 return KMP_TSX_LOCK(hle);
1371 if (hint & kmp_lock_hint_rtm)
1372 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_queuing) : __kmp_user_lock_seq;
1373 if (hint & kmp_lock_hint_adaptive)
1374 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
1377 if ((hint & omp_lock_hint_contended) && (hint & omp_lock_hint_uncontended))
1378 return __kmp_user_lock_seq;
1379 if ((hint & omp_lock_hint_speculative) &&
1380 (hint & omp_lock_hint_nonspeculative))
1381 return __kmp_user_lock_seq;
1384 if (hint & omp_lock_hint_contended)
1385 return lockseq_queuing;
1388 if ((hint & omp_lock_hint_uncontended) && !(hint & omp_lock_hint_speculative))
1392 if (hint & omp_lock_hint_speculative)
1393 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_spin) : __kmp_user_lock_seq;
1395 return __kmp_user_lock_seq;
1398#if OMPT_SUPPORT && OMPT_OPTIONAL
1399#if KMP_USE_DYNAMIC_LOCK
1400static kmp_mutex_impl_t
1401__ompt_get_mutex_impl_type(
void *user_lock, kmp_indirect_lock_t *ilock = 0) {
1403 switch (KMP_EXTRACT_D_TAG(user_lock)) {
1408 return kmp_mutex_impl_queuing;
1411 return kmp_mutex_impl_spin;
1414 case locktag_rtm_spin:
1415 return kmp_mutex_impl_speculative;
1418 return kmp_mutex_impl_none;
1420 ilock = KMP_LOOKUP_I_LOCK(user_lock);
1423 switch (ilock->type) {
1425 case locktag_adaptive:
1426 case locktag_rtm_queuing:
1427 return kmp_mutex_impl_speculative;
1429 case locktag_nested_tas:
1430 return kmp_mutex_impl_spin;
1432 case locktag_nested_futex:
1434 case locktag_ticket:
1435 case locktag_queuing:
1437 case locktag_nested_ticket:
1438 case locktag_nested_queuing:
1439 case locktag_nested_drdpa:
1440 return kmp_mutex_impl_queuing;
1442 return kmp_mutex_impl_none;
1447static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
1448 switch (__kmp_user_lock_kind) {
1450 return kmp_mutex_impl_spin;
1457 return kmp_mutex_impl_queuing;
1460 case lk_rtm_queuing:
1463 return kmp_mutex_impl_speculative;
1466 return kmp_mutex_impl_none;
1485void __kmpc_critical_with_hint(
ident_t *loc, kmp_int32 global_tid,
1486 kmp_critical_name *crit, uint32_t hint) {
1488 kmp_user_lock_p lck;
1489#if OMPT_SUPPORT && OMPT_OPTIONAL
1490 ompt_state_t prev_state = ompt_state_undefined;
1491 ompt_thread_info_t ti;
1493 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(global_tid);
1495 codeptr = OMPT_GET_RETURN_ADDRESS(0);
1498 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1499 __kmp_assert_valid_gtid(global_tid);
1501 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
1503 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1504 kmp_dyna_lockseq_t lockseq = __kmp_map_hint_to_lock(hint);
1506 if (KMP_IS_D_LOCK(lockseq)) {
1507 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
1508 KMP_GET_D_TAG(lockseq));
1510 __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lockseq));
1516 if (KMP_EXTRACT_D_TAG(lk) != 0) {
1517 lck = (kmp_user_lock_p)lk;
1518 if (__kmp_env_consistency_check) {
1519 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1520 __kmp_map_hint_to_lock(hint));
1523 __kmp_itt_critical_acquiring(lck);
1525#if OMPT_SUPPORT && OMPT_OPTIONAL
1526 if (ompt_enabled.enabled) {
1527 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1529 prev_state = ti.state;
1530 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1531 ti.state = ompt_state_wait_critical;
1534 if (ompt_enabled.ompt_callback_mutex_acquire) {
1535 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1536 ompt_mutex_critical, (
unsigned int)hint,
1537 __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)(uintptr_t)lck,
1542#if KMP_USE_INLINED_TAS
1543 if (lockseq == lockseq_tas && !__kmp_env_consistency_check) {
1544 KMP_ACQUIRE_TAS_LOCK(lck, global_tid);
1546#elif KMP_USE_INLINED_FUTEX
1547 if (lockseq == lockseq_futex && !__kmp_env_consistency_check) {
1548 KMP_ACQUIRE_FUTEX_LOCK(lck, global_tid);
1552 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
1555 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
1557 if (__kmp_env_consistency_check) {
1558 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1559 __kmp_map_hint_to_lock(hint));
1562 __kmp_itt_critical_acquiring(lck);
1564#if OMPT_SUPPORT && OMPT_OPTIONAL
1565 if (ompt_enabled.enabled) {
1566 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1568 prev_state = ti.state;
1569 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1570 ti.state = ompt_state_wait_critical;
1573 if (ompt_enabled.ompt_callback_mutex_acquire) {
1574 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1575 ompt_mutex_critical, (
unsigned int)hint,
1576 __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)(uintptr_t)lck,
1581 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
1583 KMP_POP_PARTITIONED_TIMER();
1586 __kmp_itt_critical_acquired(lck);
1588#if OMPT_SUPPORT && OMPT_OPTIONAL
1589 if (ompt_enabled.enabled) {
1591 ti.state = prev_state;
1595 if (ompt_enabled.ompt_callback_mutex_acquired) {
1596 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1597 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
1602 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1603 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1618 kmp_critical_name *crit) {
1619 kmp_user_lock_p lck;
1621 KC_TRACE(10, (
"__kmpc_end_critical: called T#%d\n", global_tid));
1623#if KMP_USE_DYNAMIC_LOCK
1624 int locktag = KMP_EXTRACT_D_TAG(crit);
1626 lck = (kmp_user_lock_p)crit;
1627 KMP_ASSERT(lck != NULL);
1628 if (__kmp_env_consistency_check) {
1629 __kmp_pop_sync(global_tid, ct_critical, loc);
1632 __kmp_itt_critical_releasing(lck);
1634#if KMP_USE_INLINED_TAS
1635 if (locktag == locktag_tas && !__kmp_env_consistency_check) {
1636 KMP_RELEASE_TAS_LOCK(lck, global_tid);
1638#elif KMP_USE_INLINED_FUTEX
1639 if (locktag == locktag_futex && !__kmp_env_consistency_check) {
1640 KMP_RELEASE_FUTEX_LOCK(lck, global_tid);
1644 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
1647 kmp_indirect_lock_t *ilk =
1648 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
1649 KMP_ASSERT(ilk != NULL);
1651 if (__kmp_env_consistency_check) {
1652 __kmp_pop_sync(global_tid, ct_critical, loc);
1655 __kmp_itt_critical_releasing(lck);
1657 KMP_I_LOCK_FUNC(ilk, unset)(lck, global_tid);
1662 if ((__kmp_user_lock_kind == lk_tas) &&
1663 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1664 lck = (kmp_user_lock_p)crit;
1667 else if ((__kmp_user_lock_kind == lk_futex) &&
1668 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1669 lck = (kmp_user_lock_p)crit;
1673 lck = (kmp_user_lock_p)TCR_PTR(*((kmp_user_lock_p *)crit));
1676 KMP_ASSERT(lck != NULL);
1678 if (__kmp_env_consistency_check)
1679 __kmp_pop_sync(global_tid, ct_critical, loc);
1682 __kmp_itt_critical_releasing(lck);
1686 __kmp_release_user_lock_with_checks(lck, global_tid);
1690#if OMPT_SUPPORT && OMPT_OPTIONAL
1693 OMPT_STORE_RETURN_ADDRESS(global_tid);
1694 if (ompt_enabled.ompt_callback_mutex_released) {
1695 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1696 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck,
1697 OMPT_LOAD_RETURN_ADDRESS(0));
1701 KMP_POP_PARTITIONED_TIMER();
1702 KA_TRACE(15, (
"__kmpc_end_critical: done T#%d\n", global_tid));
1716 KC_TRACE(10, (
"__kmpc_barrier_master: called T#%d\n", global_tid));
1717 __kmp_assert_valid_gtid(global_tid);
1719 if (!TCR_4(__kmp_init_parallel))
1720 __kmp_parallel_initialize();
1722 __kmp_resume_if_soft_paused();
1724 if (__kmp_env_consistency_check)
1725 __kmp_check_barrier(global_tid, ct_barrier, loc);
1728 ompt_frame_t *ompt_frame;
1729 if (ompt_enabled.enabled) {
1730 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1731 if (ompt_frame->enter_frame.ptr == NULL)
1732 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1734 OMPT_STORE_RETURN_ADDRESS(global_tid);
1737 __kmp_threads[global_tid]->th.th_ident = loc;
1739 status = __kmp_barrier(bs_plain_barrier, global_tid, TRUE, 0, NULL, NULL);
1740#if OMPT_SUPPORT && OMPT_OPTIONAL
1741 if (ompt_enabled.enabled) {
1742 ompt_frame->enter_frame = ompt_data_none;
1746 return (status != 0) ? 0 : 1;
1759 KC_TRACE(10, (
"__kmpc_end_barrier_master: called T#%d\n", global_tid));
1760 __kmp_assert_valid_gtid(global_tid);
1761 __kmp_end_split_barrier(bs_plain_barrier, global_tid);
1776 KC_TRACE(10, (
"__kmpc_barrier_master_nowait: called T#%d\n", global_tid));
1777 __kmp_assert_valid_gtid(global_tid);
1779 if (!TCR_4(__kmp_init_parallel))
1780 __kmp_parallel_initialize();
1782 __kmp_resume_if_soft_paused();
1784 if (__kmp_env_consistency_check) {
1786 KMP_WARNING(ConstructIdentInvalid);
1788 __kmp_check_barrier(global_tid, ct_barrier, loc);
1792 ompt_frame_t *ompt_frame;
1793 if (ompt_enabled.enabled) {
1794 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1795 if (ompt_frame->enter_frame.ptr == NULL)
1796 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1798 OMPT_STORE_RETURN_ADDRESS(global_tid);
1801 __kmp_threads[global_tid]->th.th_ident = loc;
1803 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
1804#if OMPT_SUPPORT && OMPT_OPTIONAL
1805 if (ompt_enabled.enabled) {
1806 ompt_frame->enter_frame = ompt_data_none;
1812 if (__kmp_env_consistency_check) {
1818 __kmp_pop_sync(global_tid, ct_master, loc);
1838 __kmp_assert_valid_gtid(global_tid);
1839 kmp_int32 rc = __kmp_enter_single(global_tid, loc, TRUE);
1844 KMP_PUSH_PARTITIONED_TIMER(OMP_single);
1847#if OMPT_SUPPORT && OMPT_OPTIONAL
1848 kmp_info_t *this_thr = __kmp_threads[global_tid];
1849 kmp_team_t *team = this_thr->th.th_team;
1850 int tid = __kmp_tid_from_gtid(global_tid);
1852 if (ompt_enabled.enabled) {
1854 if (ompt_enabled.ompt_callback_work) {
1855 ompt_callbacks.ompt_callback(ompt_callback_work)(
1856 ompt_work_single_executor, ompt_scope_begin,
1857 &(team->t.ompt_team_info.parallel_data),
1858 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1859 1, OMPT_GET_RETURN_ADDRESS(0));
1862 if (ompt_enabled.ompt_callback_work) {
1863 ompt_callbacks.ompt_callback(ompt_callback_work)(
1864 ompt_work_single_other, ompt_scope_begin,
1865 &(team->t.ompt_team_info.parallel_data),
1866 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1867 1, OMPT_GET_RETURN_ADDRESS(0));
1868 ompt_callbacks.ompt_callback(ompt_callback_work)(
1869 ompt_work_single_other, ompt_scope_end,
1870 &(team->t.ompt_team_info.parallel_data),
1871 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1872 1, OMPT_GET_RETURN_ADDRESS(0));
1891 __kmp_assert_valid_gtid(global_tid);
1892 __kmp_exit_single(global_tid);
1893 KMP_POP_PARTITIONED_TIMER();
1895#if OMPT_SUPPORT && OMPT_OPTIONAL
1896 kmp_info_t *this_thr = __kmp_threads[global_tid];
1897 kmp_team_t *team = this_thr->th.th_team;
1898 int tid = __kmp_tid_from_gtid(global_tid);
1900 if (ompt_enabled.ompt_callback_work) {
1901 ompt_callbacks.ompt_callback(ompt_callback_work)(
1902 ompt_work_single_executor, ompt_scope_end,
1903 &(team->t.ompt_team_info.parallel_data),
1904 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
1905 OMPT_GET_RETURN_ADDRESS(0));
1918 KMP_POP_PARTITIONED_TIMER();
1919 KE_TRACE(10, (
"__kmpc_for_static_fini called T#%d\n", global_tid));
1921#if OMPT_SUPPORT && OMPT_OPTIONAL
1922 if (ompt_enabled.ompt_callback_work) {
1923 ompt_work_t ompt_work_type = ompt_work_loop;
1924 ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
1925 ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
1929 ompt_work_type = ompt_work_loop;
1931 ompt_work_type = ompt_work_sections;
1933 ompt_work_type = ompt_work_distribute;
1938 KMP_DEBUG_ASSERT(ompt_work_type);
1940 ompt_callbacks.ompt_callback(ompt_callback_work)(
1941 ompt_work_type, ompt_scope_end, &(team_info->parallel_data),
1942 &(task_info->task_data), 0, OMPT_GET_RETURN_ADDRESS(0));
1945 if (__kmp_env_consistency_check)
1946 __kmp_pop_workshare(global_tid, ct_pdo, loc);
1952void ompc_set_num_threads(
int arg) {
1954 __kmp_set_num_threads(arg, __kmp_entry_gtid());
1957void ompc_set_dynamic(
int flag) {
1961 thread = __kmp_entry_thread();
1963 __kmp_save_internal_controls(thread);
1965 set__dynamic(thread, flag ?
true : false);
1968void ompc_set_nested(
int flag) {
1972 thread = __kmp_entry_thread();
1974 __kmp_save_internal_controls(thread);
1976 set__max_active_levels(thread, flag ? __kmp_dflt_max_active_levels : 1);
1979void ompc_set_max_active_levels(
int max_active_levels) {
1984 __kmp_set_max_active_levels(__kmp_entry_gtid(), max_active_levels);
1987void ompc_set_schedule(omp_sched_t kind,
int modifier) {
1989 __kmp_set_schedule(__kmp_entry_gtid(), (kmp_sched_t)kind, modifier);
1992int ompc_get_ancestor_thread_num(
int level) {
1993 return __kmp_get_ancestor_thread_num(__kmp_entry_gtid(), level);
1996int ompc_get_team_size(
int level) {
1997 return __kmp_get_team_size(__kmp_entry_gtid(), level);
2001void KMP_EXPAND_NAME(ompc_set_affinity_format)(
char const *format) {
2002 if (!__kmp_init_serial) {
2003 __kmp_serial_initialize();
2005 __kmp_strncpy_truncate(__kmp_affinity_format, KMP_AFFINITY_FORMAT_SIZE,
2006 format, KMP_STRLEN(format) + 1);
2009size_t KMP_EXPAND_NAME(ompc_get_affinity_format)(
char *buffer,
size_t size) {
2011 if (!__kmp_init_serial) {
2012 __kmp_serial_initialize();
2014 format_size = KMP_STRLEN(__kmp_affinity_format);
2015 if (buffer && size) {
2016 __kmp_strncpy_truncate(buffer, size, __kmp_affinity_format,
2022void KMP_EXPAND_NAME(ompc_display_affinity)(
char const *format) {
2024 if (!TCR_4(__kmp_init_middle)) {
2025 __kmp_middle_initialize();
2027 __kmp_assign_root_init_mask();
2028 gtid = __kmp_get_gtid();
2029#if KMP_AFFINITY_SUPPORTED
2030 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 && __kmp_affin_reset) {
2031 __kmp_reset_root_init_mask(gtid);
2034 __kmp_aux_display_affinity(gtid, format);
2037size_t KMP_EXPAND_NAME(ompc_capture_affinity)(
char *buffer,
size_t buf_size,
2038 char const *format) {
2040 size_t num_required;
2041 kmp_str_buf_t capture_buf;
2042 if (!TCR_4(__kmp_init_middle)) {
2043 __kmp_middle_initialize();
2045 __kmp_assign_root_init_mask();
2046 gtid = __kmp_get_gtid();
2047#if KMP_AFFINITY_SUPPORTED
2048 if (__kmp_threads[gtid]->th.th_team->t.t_level == 0 && __kmp_affin_reset) {
2049 __kmp_reset_root_init_mask(gtid);
2052 __kmp_str_buf_init(&capture_buf);
2053 num_required = __kmp_aux_capture_affinity(gtid, format, &capture_buf);
2054 if (buffer && buf_size) {
2055 __kmp_strncpy_truncate(buffer, buf_size, capture_buf.str,
2056 capture_buf.used + 1);
2058 __kmp_str_buf_free(&capture_buf);
2059 return num_required;
2062void kmpc_set_stacksize(
int arg) {
2064 __kmp_aux_set_stacksize(arg);
2067void kmpc_set_stacksize_s(
size_t arg) {
2069 __kmp_aux_set_stacksize(arg);
2072void kmpc_set_blocktime(
int arg) {
2076 gtid = __kmp_entry_gtid();
2077 tid = __kmp_tid_from_gtid(gtid);
2078 thread = __kmp_thread_from_gtid(gtid);
2080 __kmp_aux_set_blocktime(arg, thread, tid);
2083void kmpc_set_library(
int arg) {
2085 __kmp_user_set_library((
enum library_type)arg);
2088void kmpc_set_defaults(
char const *str) {
2090 __kmp_aux_set_defaults(str, KMP_STRLEN(str));
2093void kmpc_set_disp_num_buffers(
int arg) {
2096 if (__kmp_init_serial == FALSE && arg >= KMP_MIN_DISP_NUM_BUFF &&
2097 arg <= KMP_MAX_DISP_NUM_BUFF) {
2098 __kmp_dispatch_num_buffers = arg;
2102int kmpc_set_affinity_mask_proc(
int proc,
void **mask) {
2103#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2106 if (!TCR_4(__kmp_init_middle)) {
2107 __kmp_middle_initialize();
2109 __kmp_assign_root_init_mask();
2110 return __kmp_aux_set_affinity_mask_proc(proc, mask);
2114int kmpc_unset_affinity_mask_proc(
int proc,
void **mask) {
2115#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2118 if (!TCR_4(__kmp_init_middle)) {
2119 __kmp_middle_initialize();
2121 __kmp_assign_root_init_mask();
2122 return __kmp_aux_unset_affinity_mask_proc(proc, mask);
2126int kmpc_get_affinity_mask_proc(
int proc,
void **mask) {
2127#if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2130 if (!TCR_4(__kmp_init_middle)) {
2131 __kmp_middle_initialize();
2133 __kmp_assign_root_init_mask();
2134 return __kmp_aux_get_affinity_mask_proc(proc, mask);
2184 void *cpy_data,
void (*cpy_func)(
void *,
void *),
2187 KC_TRACE(10, (
"__kmpc_copyprivate: called T#%d\n", gtid));
2188 __kmp_assert_valid_gtid(gtid);
2192 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2194 if (__kmp_env_consistency_check) {
2196 KMP_WARNING(ConstructIdentInvalid);
2203 *data_ptr = cpy_data;
2206 ompt_frame_t *ompt_frame;
2207 if (ompt_enabled.enabled) {
2208 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2209 if (ompt_frame->enter_frame.ptr == NULL)
2210 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2212 OMPT_STORE_RETURN_ADDRESS(gtid);
2216 __kmp_threads[gtid]->th.th_ident = loc;
2218 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2221 (*cpy_func)(cpy_data, *data_ptr);
2227 OMPT_STORE_RETURN_ADDRESS(gtid);
2230 __kmp_threads[gtid]->th.th_ident = loc;
2233 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2234#if OMPT_SUPPORT && OMPT_OPTIONAL
2235 if (ompt_enabled.enabled) {
2236 ompt_frame->enter_frame = ompt_data_none;
2262 KC_TRACE(10, (
"__kmpc_copyprivate_light: called T#%d\n", gtid));
2266 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2268 if (__kmp_env_consistency_check) {
2270 KMP_WARNING(ConstructIdentInvalid);
2277 *data_ptr = cpy_data;
2280 ompt_frame_t *ompt_frame;
2281 if (ompt_enabled.enabled) {
2282 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2283 if (ompt_frame->enter_frame.ptr == NULL)
2284 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2285 OMPT_STORE_RETURN_ADDRESS(gtid);
2290 __kmp_threads[gtid]->th.th_ident = loc;
2292 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2299#define INIT_LOCK __kmp_init_user_lock_with_checks
2300#define INIT_NESTED_LOCK __kmp_init_nested_user_lock_with_checks
2301#define ACQUIRE_LOCK __kmp_acquire_user_lock_with_checks
2302#define ACQUIRE_LOCK_TIMED __kmp_acquire_user_lock_with_checks_timed
2303#define ACQUIRE_NESTED_LOCK __kmp_acquire_nested_user_lock_with_checks
2304#define ACQUIRE_NESTED_LOCK_TIMED \
2305 __kmp_acquire_nested_user_lock_with_checks_timed
2306#define RELEASE_LOCK __kmp_release_user_lock_with_checks
2307#define RELEASE_NESTED_LOCK __kmp_release_nested_user_lock_with_checks
2308#define TEST_LOCK __kmp_test_user_lock_with_checks
2309#define TEST_NESTED_LOCK __kmp_test_nested_user_lock_with_checks
2310#define DESTROY_LOCK __kmp_destroy_user_lock_with_checks
2311#define DESTROY_NESTED_LOCK __kmp_destroy_nested_user_lock_with_checks
2316#if KMP_USE_DYNAMIC_LOCK
2319static __forceinline
void __kmp_init_lock_with_hint(
ident_t *loc,
void **lock,
2320 kmp_dyna_lockseq_t seq) {
2321 if (KMP_IS_D_LOCK(seq)) {
2322 KMP_INIT_D_LOCK(lock, seq);
2324 __kmp_itt_lock_creating((kmp_user_lock_p)lock, NULL);
2327 KMP_INIT_I_LOCK(lock, seq);
2329 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2330 __kmp_itt_lock_creating(ilk->lock, loc);
2336static __forceinline
void
2337__kmp_init_nest_lock_with_hint(
ident_t *loc,
void **lock,
2338 kmp_dyna_lockseq_t seq) {
2341 if (seq == lockseq_hle || seq == lockseq_rtm_queuing ||
2342 seq == lockseq_rtm_spin || seq == lockseq_adaptive)
2343 seq = __kmp_user_lock_seq;
2347 seq = lockseq_nested_tas;
2351 seq = lockseq_nested_futex;
2354 case lockseq_ticket:
2355 seq = lockseq_nested_ticket;
2357 case lockseq_queuing:
2358 seq = lockseq_nested_queuing;
2361 seq = lockseq_nested_drdpa;
2364 seq = lockseq_nested_queuing;
2366 KMP_INIT_I_LOCK(lock, seq);
2368 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2369 __kmp_itt_lock_creating(ilk->lock, loc);
2374void __kmpc_init_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
void **user_lock,
2376 KMP_DEBUG_ASSERT(__kmp_init_serial);
2377 if (__kmp_env_consistency_check && user_lock == NULL) {
2378 KMP_FATAL(LockIsUninitialized,
"omp_init_lock_with_hint");
2381 __kmp_init_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2383#if OMPT_SUPPORT && OMPT_OPTIONAL
2385 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2387 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2388 if (ompt_enabled.ompt_callback_lock_init) {
2389 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2390 ompt_mutex_lock, (omp_lock_hint_t)hint,
2391 __ompt_get_mutex_impl_type(user_lock),
2392 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2398void __kmpc_init_nest_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
2399 void **user_lock, uintptr_t hint) {
2400 KMP_DEBUG_ASSERT(__kmp_init_serial);
2401 if (__kmp_env_consistency_check && user_lock == NULL) {
2402 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock_with_hint");
2405 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2407#if OMPT_SUPPORT && OMPT_OPTIONAL
2409 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2411 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2412 if (ompt_enabled.ompt_callback_lock_init) {
2413 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2414 ompt_mutex_nest_lock, (omp_lock_hint_t)hint,
2415 __ompt_get_mutex_impl_type(user_lock),
2416 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2424void __kmpc_init_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2425#if KMP_USE_DYNAMIC_LOCK
2427 KMP_DEBUG_ASSERT(__kmp_init_serial);
2428 if (__kmp_env_consistency_check && user_lock == NULL) {
2429 KMP_FATAL(LockIsUninitialized,
"omp_init_lock");
2431 __kmp_init_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2433#if OMPT_SUPPORT && OMPT_OPTIONAL
2435 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2437 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2438 if (ompt_enabled.ompt_callback_lock_init) {
2439 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2440 ompt_mutex_lock, omp_lock_hint_none,
2441 __ompt_get_mutex_impl_type(user_lock),
2442 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2448 static char const *
const func =
"omp_init_lock";
2449 kmp_user_lock_p lck;
2450 KMP_DEBUG_ASSERT(__kmp_init_serial);
2452 if (__kmp_env_consistency_check) {
2453 if (user_lock == NULL) {
2454 KMP_FATAL(LockIsUninitialized, func);
2458 KMP_CHECK_USER_LOCK_INIT();
2460 if ((__kmp_user_lock_kind == lk_tas) &&
2461 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2462 lck = (kmp_user_lock_p)user_lock;
2465 else if ((__kmp_user_lock_kind == lk_futex) &&
2466 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2467 lck = (kmp_user_lock_p)user_lock;
2471 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2474 __kmp_set_user_lock_location(lck, loc);
2476#if OMPT_SUPPORT && OMPT_OPTIONAL
2478 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2480 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2481 if (ompt_enabled.ompt_callback_lock_init) {
2482 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2483 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2484 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2489 __kmp_itt_lock_creating(lck);
2496void __kmpc_init_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2497#if KMP_USE_DYNAMIC_LOCK
2499 KMP_DEBUG_ASSERT(__kmp_init_serial);
2500 if (__kmp_env_consistency_check && user_lock == NULL) {
2501 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock");
2503 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2505#if OMPT_SUPPORT && OMPT_OPTIONAL
2507 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2509 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2510 if (ompt_enabled.ompt_callback_lock_init) {
2511 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2512 ompt_mutex_nest_lock, omp_lock_hint_none,
2513 __ompt_get_mutex_impl_type(user_lock),
2514 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2520 static char const *
const func =
"omp_init_nest_lock";
2521 kmp_user_lock_p lck;
2522 KMP_DEBUG_ASSERT(__kmp_init_serial);
2524 if (__kmp_env_consistency_check) {
2525 if (user_lock == NULL) {
2526 KMP_FATAL(LockIsUninitialized, func);
2530 KMP_CHECK_USER_LOCK_INIT();
2532 if ((__kmp_user_lock_kind == lk_tas) &&
2533 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2534 OMP_NEST_LOCK_T_SIZE)) {
2535 lck = (kmp_user_lock_p)user_lock;
2538 else if ((__kmp_user_lock_kind == lk_futex) &&
2539 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2540 OMP_NEST_LOCK_T_SIZE)) {
2541 lck = (kmp_user_lock_p)user_lock;
2545 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2548 INIT_NESTED_LOCK(lck);
2549 __kmp_set_user_lock_location(lck, loc);
2551#if OMPT_SUPPORT && OMPT_OPTIONAL
2553 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2555 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2556 if (ompt_enabled.ompt_callback_lock_init) {
2557 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2558 ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2559 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2564 __kmp_itt_lock_creating(lck);
2570void __kmpc_destroy_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2571#if KMP_USE_DYNAMIC_LOCK
2574 kmp_user_lock_p lck;
2575 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2576 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2578 lck = (kmp_user_lock_p)user_lock;
2580 __kmp_itt_lock_destroyed(lck);
2582#if OMPT_SUPPORT && OMPT_OPTIONAL
2584 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2586 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2587 if (ompt_enabled.ompt_callback_lock_destroy) {
2588 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2589 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2592 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2594 kmp_user_lock_p lck;
2596 if ((__kmp_user_lock_kind == lk_tas) &&
2597 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2598 lck = (kmp_user_lock_p)user_lock;
2601 else if ((__kmp_user_lock_kind == lk_futex) &&
2602 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2603 lck = (kmp_user_lock_p)user_lock;
2607 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_lock");
2610#if OMPT_SUPPORT && OMPT_OPTIONAL
2612 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2614 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2615 if (ompt_enabled.ompt_callback_lock_destroy) {
2616 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2617 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2622 __kmp_itt_lock_destroyed(lck);
2626 if ((__kmp_user_lock_kind == lk_tas) &&
2627 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2631 else if ((__kmp_user_lock_kind == lk_futex) &&
2632 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2637 __kmp_user_lock_free(user_lock, gtid, lck);
2643void __kmpc_destroy_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2644#if KMP_USE_DYNAMIC_LOCK
2647 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(user_lock);
2648 __kmp_itt_lock_destroyed(ilk->lock);
2650#if OMPT_SUPPORT && OMPT_OPTIONAL
2652 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2654 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2655 if (ompt_enabled.ompt_callback_lock_destroy) {
2656 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2657 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2660 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2664 kmp_user_lock_p lck;
2666 if ((__kmp_user_lock_kind == lk_tas) &&
2667 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2668 OMP_NEST_LOCK_T_SIZE)) {
2669 lck = (kmp_user_lock_p)user_lock;
2672 else if ((__kmp_user_lock_kind == lk_futex) &&
2673 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2674 OMP_NEST_LOCK_T_SIZE)) {
2675 lck = (kmp_user_lock_p)user_lock;
2679 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_nest_lock");
2682#if OMPT_SUPPORT && OMPT_OPTIONAL
2684 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2686 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2687 if (ompt_enabled.ompt_callback_lock_destroy) {
2688 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2689 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2694 __kmp_itt_lock_destroyed(lck);
2697 DESTROY_NESTED_LOCK(lck);
2699 if ((__kmp_user_lock_kind == lk_tas) &&
2700 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2701 OMP_NEST_LOCK_T_SIZE)) {
2705 else if ((__kmp_user_lock_kind == lk_futex) &&
2706 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2707 OMP_NEST_LOCK_T_SIZE)) {
2712 __kmp_user_lock_free(user_lock, gtid, lck);
2717void __kmpc_set_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2719#if KMP_USE_DYNAMIC_LOCK
2720 int tag = KMP_EXTRACT_D_TAG(user_lock);
2722 __kmp_itt_lock_acquiring(
2726#if OMPT_SUPPORT && OMPT_OPTIONAL
2728 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2730 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2731 if (ompt_enabled.ompt_callback_mutex_acquire) {
2732 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2733 ompt_mutex_lock, omp_lock_hint_none,
2734 __ompt_get_mutex_impl_type(user_lock),
2735 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2738#if KMP_USE_INLINED_TAS
2739 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2740 KMP_ACQUIRE_TAS_LOCK(user_lock, gtid);
2742#elif KMP_USE_INLINED_FUTEX
2743 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2744 KMP_ACQUIRE_FUTEX_LOCK(user_lock, gtid);
2748 __kmp_direct_set[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2751 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2753#if OMPT_SUPPORT && OMPT_OPTIONAL
2754 if (ompt_enabled.ompt_callback_mutex_acquired) {
2755 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2756 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2762 kmp_user_lock_p lck;
2764 if ((__kmp_user_lock_kind == lk_tas) &&
2765 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2766 lck = (kmp_user_lock_p)user_lock;
2769 else if ((__kmp_user_lock_kind == lk_futex) &&
2770 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2771 lck = (kmp_user_lock_p)user_lock;
2775 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_lock");
2779 __kmp_itt_lock_acquiring(lck);
2781#if OMPT_SUPPORT && OMPT_OPTIONAL
2783 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2785 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2786 if (ompt_enabled.ompt_callback_mutex_acquire) {
2787 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2788 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2789 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2793 ACQUIRE_LOCK(lck, gtid);
2796 __kmp_itt_lock_acquired(lck);
2799#if OMPT_SUPPORT && OMPT_OPTIONAL
2800 if (ompt_enabled.ompt_callback_mutex_acquired) {
2801 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2802 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2809void __kmpc_set_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2810#if KMP_USE_DYNAMIC_LOCK
2813 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
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.enabled) {
2821 if (ompt_enabled.ompt_callback_mutex_acquire) {
2822 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2823 ompt_mutex_nest_lock, omp_lock_hint_none,
2824 __ompt_get_mutex_impl_type(user_lock),
2825 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2829 int acquire_status =
2830 KMP_D_LOCK_FUNC(user_lock, set)((kmp_dyna_lock_t *)user_lock, gtid);
2831 (void)acquire_status;
2833 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2836#if OMPT_SUPPORT && OMPT_OPTIONAL
2837 if (ompt_enabled.enabled) {
2838 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2839 if (ompt_enabled.ompt_callback_mutex_acquired) {
2841 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2842 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
2846 if (ompt_enabled.ompt_callback_nest_lock) {
2848 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2849 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2857 kmp_user_lock_p lck;
2859 if ((__kmp_user_lock_kind == lk_tas) &&
2860 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2861 OMP_NEST_LOCK_T_SIZE)) {
2862 lck = (kmp_user_lock_p)user_lock;
2865 else if ((__kmp_user_lock_kind == lk_futex) &&
2866 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2867 OMP_NEST_LOCK_T_SIZE)) {
2868 lck = (kmp_user_lock_p)user_lock;
2872 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_nest_lock");
2876 __kmp_itt_lock_acquiring(lck);
2878#if OMPT_SUPPORT && OMPT_OPTIONAL
2880 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2882 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2883 if (ompt_enabled.enabled) {
2884 if (ompt_enabled.ompt_callback_mutex_acquire) {
2885 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2886 ompt_mutex_nest_lock, omp_lock_hint_none,
2887 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
2893 ACQUIRE_NESTED_LOCK(lck, gtid, &acquire_status);
2896 __kmp_itt_lock_acquired(lck);
2899#if OMPT_SUPPORT && OMPT_OPTIONAL
2900 if (ompt_enabled.enabled) {
2901 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2902 if (ompt_enabled.ompt_callback_mutex_acquired) {
2904 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2905 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2908 if (ompt_enabled.ompt_callback_nest_lock) {
2910 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2911 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2920void __kmpc_unset_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2921#if KMP_USE_DYNAMIC_LOCK
2923 int tag = KMP_EXTRACT_D_TAG(user_lock);
2925 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2927#if KMP_USE_INLINED_TAS
2928 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2929 KMP_RELEASE_TAS_LOCK(user_lock, gtid);
2931#elif KMP_USE_INLINED_FUTEX
2932 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2933 KMP_RELEASE_FUTEX_LOCK(user_lock, gtid);
2937 __kmp_direct_unset[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2940#if OMPT_SUPPORT && OMPT_OPTIONAL
2942 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2944 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2945 if (ompt_enabled.ompt_callback_mutex_released) {
2946 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2947 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2953 kmp_user_lock_p lck;
2958 if ((__kmp_user_lock_kind == lk_tas) &&
2959 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2960#if KMP_OS_LINUX && \
2961 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
2964 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2966 TCW_4(((kmp_user_lock_p)user_lock)->tas.lk.poll, 0);
2969#if OMPT_SUPPORT && OMPT_OPTIONAL
2971 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2973 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2974 if (ompt_enabled.ompt_callback_mutex_released) {
2975 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2976 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2982 lck = (kmp_user_lock_p)user_lock;
2986 else if ((__kmp_user_lock_kind == lk_futex) &&
2987 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2988 lck = (kmp_user_lock_p)user_lock;
2992 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_lock");
2996 __kmp_itt_lock_releasing(lck);
2999 RELEASE_LOCK(lck, gtid);
3001#if OMPT_SUPPORT && OMPT_OPTIONAL
3003 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3005 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3006 if (ompt_enabled.ompt_callback_mutex_released) {
3007 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3008 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3016void __kmpc_unset_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3017#if KMP_USE_DYNAMIC_LOCK
3020 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3022 int release_status =
3023 KMP_D_LOCK_FUNC(user_lock, unset)((kmp_dyna_lock_t *)user_lock, gtid);
3024 (void)release_status;
3026#if OMPT_SUPPORT && OMPT_OPTIONAL
3028 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3030 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3031 if (ompt_enabled.enabled) {
3032 if (release_status == KMP_LOCK_RELEASED) {
3033 if (ompt_enabled.ompt_callback_mutex_released) {
3035 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3036 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3039 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3041 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3042 ompt_scope_end, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3049 kmp_user_lock_p lck;
3053 if ((__kmp_user_lock_kind == lk_tas) &&
3054 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3055 OMP_NEST_LOCK_T_SIZE)) {
3056#if KMP_OS_LINUX && \
3057 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
3059 kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
3061 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
3064#if OMPT_SUPPORT && OMPT_OPTIONAL
3065 int release_status = KMP_LOCK_STILL_HELD;
3068 if (--(tl->lk.depth_locked) == 0) {
3069 TCW_4(tl->lk.poll, 0);
3070#if OMPT_SUPPORT && OMPT_OPTIONAL
3071 release_status = KMP_LOCK_RELEASED;
3076#if OMPT_SUPPORT && OMPT_OPTIONAL
3078 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3080 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3081 if (ompt_enabled.enabled) {
3082 if (release_status == KMP_LOCK_RELEASED) {
3083 if (ompt_enabled.ompt_callback_mutex_released) {
3085 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3086 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3088 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3090 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3091 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3098 lck = (kmp_user_lock_p)user_lock;
3102 else if ((__kmp_user_lock_kind == lk_futex) &&
3103 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3104 OMP_NEST_LOCK_T_SIZE)) {
3105 lck = (kmp_user_lock_p)user_lock;
3109 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_nest_lock");
3113 __kmp_itt_lock_releasing(lck);
3117 release_status = RELEASE_NESTED_LOCK(lck, gtid);
3118#if OMPT_SUPPORT && OMPT_OPTIONAL
3120 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3122 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3123 if (ompt_enabled.enabled) {
3124 if (release_status == KMP_LOCK_RELEASED) {
3125 if (ompt_enabled.ompt_callback_mutex_released) {
3127 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3128 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3130 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3132 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3133 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3142int __kmpc_test_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3145#if KMP_USE_DYNAMIC_LOCK
3147 int tag = KMP_EXTRACT_D_TAG(user_lock);
3149 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3151#if OMPT_SUPPORT && OMPT_OPTIONAL
3153 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3155 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3156 if (ompt_enabled.ompt_callback_mutex_acquire) {
3157 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3158 ompt_mutex_lock, omp_lock_hint_none,
3159 __ompt_get_mutex_impl_type(user_lock),
3160 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3163#if KMP_USE_INLINED_TAS
3164 if (tag == locktag_tas && !__kmp_env_consistency_check) {
3165 KMP_TEST_TAS_LOCK(user_lock, gtid, rc);
3167#elif KMP_USE_INLINED_FUTEX
3168 if (tag == locktag_futex && !__kmp_env_consistency_check) {
3169 KMP_TEST_FUTEX_LOCK(user_lock, gtid, rc);
3173 rc = __kmp_direct_test[tag]((kmp_dyna_lock_t *)user_lock, gtid);
3177 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3179#if OMPT_SUPPORT && OMPT_OPTIONAL
3180 if (ompt_enabled.ompt_callback_mutex_acquired) {
3181 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3182 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3188 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3195 kmp_user_lock_p lck;
3198 if ((__kmp_user_lock_kind == lk_tas) &&
3199 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
3200 lck = (kmp_user_lock_p)user_lock;
3203 else if ((__kmp_user_lock_kind == lk_futex) &&
3204 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3205 lck = (kmp_user_lock_p)user_lock;
3209 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_lock");
3213 __kmp_itt_lock_acquiring(lck);
3215#if OMPT_SUPPORT && OMPT_OPTIONAL
3217 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3219 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3220 if (ompt_enabled.ompt_callback_mutex_acquire) {
3221 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3222 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
3223 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3227 rc = TEST_LOCK(lck, gtid);
3230 __kmp_itt_lock_acquired(lck);
3232 __kmp_itt_lock_cancelled(lck);
3235#if OMPT_SUPPORT && OMPT_OPTIONAL
3236 if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
3237 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3238 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3242 return (rc ? FTN_TRUE : FTN_FALSE);
3250int __kmpc_test_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3251#if KMP_USE_DYNAMIC_LOCK
3254 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3256#if OMPT_SUPPORT && OMPT_OPTIONAL
3258 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3260 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3261 if (ompt_enabled.ompt_callback_mutex_acquire) {
3262 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3263 ompt_mutex_nest_lock, omp_lock_hint_none,
3264 __ompt_get_mutex_impl_type(user_lock),
3265 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3268 rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
3271 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3273 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3276#if OMPT_SUPPORT && OMPT_OPTIONAL
3277 if (ompt_enabled.enabled && rc) {
3279 if (ompt_enabled.ompt_callback_mutex_acquired) {
3281 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3282 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3286 if (ompt_enabled.ompt_callback_nest_lock) {
3288 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3289 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3298 kmp_user_lock_p lck;
3301 if ((__kmp_user_lock_kind == lk_tas) &&
3302 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3303 OMP_NEST_LOCK_T_SIZE)) {
3304 lck = (kmp_user_lock_p)user_lock;
3307 else if ((__kmp_user_lock_kind == lk_futex) &&
3308 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3309 OMP_NEST_LOCK_T_SIZE)) {
3310 lck = (kmp_user_lock_p)user_lock;
3314 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_nest_lock");
3318 __kmp_itt_lock_acquiring(lck);
3321#if OMPT_SUPPORT && OMPT_OPTIONAL
3323 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3325 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3326 if (ompt_enabled.enabled) &&
3327 ompt_enabled.ompt_callback_mutex_acquire) {
3328 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3329 ompt_mutex_nest_lock, omp_lock_hint_none,
3330 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
3335 rc = TEST_NESTED_LOCK(lck, gtid);
3338 __kmp_itt_lock_acquired(lck);
3340 __kmp_itt_lock_cancelled(lck);
3343#if OMPT_SUPPORT && OMPT_OPTIONAL
3344 if (ompt_enabled.enabled && rc) {
3346 if (ompt_enabled.ompt_callback_mutex_acquired) {
3348 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3349 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3352 if (ompt_enabled.ompt_callback_nest_lock) {
3354 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3355 ompt_mutex_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3374#define __KMP_SET_REDUCTION_METHOD(gtid, rmethod) \
3375 ((__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) = (rmethod))
3377#define __KMP_GET_REDUCTION_METHOD(gtid) \
3378 (__kmp_threads[(gtid)]->th.th_local.packed_reduction_method)
3384static __forceinline
void
3385__kmp_enter_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3386 kmp_critical_name *crit) {
3392 kmp_user_lock_p lck;
3394#if KMP_USE_DYNAMIC_LOCK
3396 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
3399 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3400 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
3401 KMP_GET_D_TAG(__kmp_user_lock_seq));
3403 __kmp_init_indirect_csptr(crit, loc, global_tid,
3404 KMP_GET_I_TAG(__kmp_user_lock_seq));
3410 if (KMP_EXTRACT_D_TAG(lk) != 0) {
3411 lck = (kmp_user_lock_p)lk;
3412 KMP_DEBUG_ASSERT(lck != NULL);
3413 if (__kmp_env_consistency_check) {
3414 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3416 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
3418 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
3420 KMP_DEBUG_ASSERT(lck != NULL);
3421 if (__kmp_env_consistency_check) {
3422 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3424 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
3432 if (__kmp_base_user_lock_size <= INTEL_CRITICAL_SIZE) {
3433 lck = (kmp_user_lock_p)crit;
3435 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
3437 KMP_DEBUG_ASSERT(lck != NULL);
3439 if (__kmp_env_consistency_check)
3440 __kmp_push_sync(global_tid, ct_critical, loc, lck);
3442 __kmp_acquire_user_lock_with_checks(lck, global_tid);
3448static __forceinline
void
3449__kmp_end_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3450 kmp_critical_name *crit) {
3452 kmp_user_lock_p lck;
3454#if KMP_USE_DYNAMIC_LOCK
3456 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3457 lck = (kmp_user_lock_p)crit;
3458 if (__kmp_env_consistency_check)
3459 __kmp_pop_sync(global_tid, ct_critical, loc);
3460 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
3462 kmp_indirect_lock_t *ilk =
3463 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
3464 if (__kmp_env_consistency_check)
3465 __kmp_pop_sync(global_tid, ct_critical, loc);
3466 KMP_I_LOCK_FUNC(ilk, unset)(ilk->lock, global_tid);
3474 if (__kmp_base_user_lock_size > 32) {
3475 lck = *((kmp_user_lock_p *)crit);
3476 KMP_ASSERT(lck != NULL);
3478 lck = (kmp_user_lock_p)crit;
3481 if (__kmp_env_consistency_check)
3482 __kmp_pop_sync(global_tid, ct_critical, loc);
3484 __kmp_release_user_lock_with_checks(lck, global_tid);
3489static __forceinline
int
3490__kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
3495 if (th->th.th_teams_microtask) {
3496 *team_p = team = th->th.th_team;
3497 if (team->t.t_level == th->th.th_teams_level) {
3499 KMP_DEBUG_ASSERT(!th->th.th_info.ds.ds_tid);
3501 th->th.th_info.ds.ds_tid = team->t.t_master_tid;
3502 th->th.th_team = team->t.t_parent;
3503 th->th.th_team_nproc = th->th.th_team->t.t_nproc;
3504 th->th.th_task_team = th->th.th_team->t.t_task_team[0];
3505 *task_state = th->th.th_task_state;
3506 th->th.th_task_state = 0;
3514static __forceinline
void
3515__kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team,
int task_state) {
3517 th->th.th_info.ds.ds_tid = 0;
3518 th->th.th_team = team;
3519 th->th.th_team_nproc = team->t.t_nproc;
3520 th->th.th_task_team = team->t.t_task_team[task_state];
3521 __kmp_type_convert(task_state, &(th->th.th_task_state));
3542 size_t reduce_size,
void *reduce_data,
3543 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3544 kmp_critical_name *lck) {
3548 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3551 int teams_swapped = 0, task_state;
3552 KA_TRACE(10, (
"__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
3553 __kmp_assert_valid_gtid(global_tid);
3561 if (!TCR_4(__kmp_init_parallel))
3562 __kmp_parallel_initialize();
3564 __kmp_resume_if_soft_paused();
3567#if KMP_USE_DYNAMIC_LOCK
3568 if (__kmp_env_consistency_check)
3569 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3571 if (__kmp_env_consistency_check)
3572 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3575 th = __kmp_thread_from_gtid(global_tid);
3576 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3594 packed_reduction_method = __kmp_determine_reduction_method(
3595 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3596 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3598 OMPT_REDUCTION_DECL(th, global_tid);
3599 if (packed_reduction_method == critical_reduce_block) {
3601 OMPT_REDUCTION_BEGIN;
3603 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3606 }
else if (packed_reduction_method == empty_reduce_block) {
3608 OMPT_REDUCTION_BEGIN;
3614 }
else if (packed_reduction_method == atomic_reduce_block) {
3624 if (__kmp_env_consistency_check)
3625 __kmp_pop_sync(global_tid, ct_reduce, loc);
3627 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3628 tree_reduce_block)) {
3648 ompt_frame_t *ompt_frame;
3649 if (ompt_enabled.enabled) {
3650 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3651 if (ompt_frame->enter_frame.ptr == NULL)
3652 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3654 OMPT_STORE_RETURN_ADDRESS(global_tid);
3657 __kmp_threads[global_tid]->th.th_ident = loc;
3660 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3661 global_tid, FALSE, reduce_size, reduce_data, reduce_func);
3662 retval = (retval != 0) ? (0) : (1);
3663#if OMPT_SUPPORT && OMPT_OPTIONAL
3664 if (ompt_enabled.enabled) {
3665 ompt_frame->enter_frame = ompt_data_none;
3671 if (__kmp_env_consistency_check) {
3673 __kmp_pop_sync(global_tid, ct_reduce, loc);
3682 if (teams_swapped) {
3683 __kmp_restore_swapped_teams(th, team, task_state);
3687 (
"__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
3688 global_tid, packed_reduction_method, retval));
3702 kmp_critical_name *lck) {
3704 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3706 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() enter: called T#%d\n", global_tid));
3707 __kmp_assert_valid_gtid(global_tid);
3709 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3711 OMPT_REDUCTION_DECL(__kmp_thread_from_gtid(global_tid), global_tid);
3713 if (packed_reduction_method == critical_reduce_block) {
3715 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3718 }
else if (packed_reduction_method == empty_reduce_block) {
3725 }
else if (packed_reduction_method == atomic_reduce_block) {
3732 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3733 tree_reduce_block)) {
3744 if (__kmp_env_consistency_check)
3745 __kmp_pop_sync(global_tid, ct_reduce, loc);
3747 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() exit: called T#%d: method %08x\n",
3748 global_tid, packed_reduction_method));
3771 size_t reduce_size,
void *reduce_data,
3772 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3773 kmp_critical_name *lck) {
3776 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3779 int teams_swapped = 0, task_state;
3781 KA_TRACE(10, (
"__kmpc_reduce() enter: called T#%d\n", global_tid));
3782 __kmp_assert_valid_gtid(global_tid);
3790 if (!TCR_4(__kmp_init_parallel))
3791 __kmp_parallel_initialize();
3793 __kmp_resume_if_soft_paused();
3796#if KMP_USE_DYNAMIC_LOCK
3797 if (__kmp_env_consistency_check)
3798 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3800 if (__kmp_env_consistency_check)
3801 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3804 th = __kmp_thread_from_gtid(global_tid);
3805 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3807 packed_reduction_method = __kmp_determine_reduction_method(
3808 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3809 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3811 OMPT_REDUCTION_DECL(th, global_tid);
3813 if (packed_reduction_method == critical_reduce_block) {
3815 OMPT_REDUCTION_BEGIN;
3816 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3819 }
else if (packed_reduction_method == empty_reduce_block) {
3821 OMPT_REDUCTION_BEGIN;
3826 }
else if (packed_reduction_method == atomic_reduce_block) {
3830 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3831 tree_reduce_block)) {
3837 ompt_frame_t *ompt_frame;
3838 if (ompt_enabled.enabled) {
3839 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3840 if (ompt_frame->enter_frame.ptr == NULL)
3841 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3843 OMPT_STORE_RETURN_ADDRESS(global_tid);
3846 __kmp_threads[global_tid]->th.th_ident =
3850 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3851 global_tid, TRUE, reduce_size, reduce_data, reduce_func);
3852 retval = (retval != 0) ? (0) : (1);
3853#if OMPT_SUPPORT && OMPT_OPTIONAL
3854 if (ompt_enabled.enabled) {
3855 ompt_frame->enter_frame = ompt_data_none;
3861 if (__kmp_env_consistency_check) {
3863 __kmp_pop_sync(global_tid, ct_reduce, loc);
3872 if (teams_swapped) {
3873 __kmp_restore_swapped_teams(th, team, task_state);
3877 (
"__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
3878 global_tid, packed_reduction_method, retval));
3893 kmp_critical_name *lck) {
3895 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3898 int teams_swapped = 0, task_state;
3900 KA_TRACE(10, (
"__kmpc_end_reduce() enter: called T#%d\n", global_tid));
3901 __kmp_assert_valid_gtid(global_tid);
3903 th = __kmp_thread_from_gtid(global_tid);
3904 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3906 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3910 OMPT_REDUCTION_DECL(th, global_tid);
3912 if (packed_reduction_method == critical_reduce_block) {
3913 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3919 ompt_frame_t *ompt_frame;
3920 if (ompt_enabled.enabled) {
3921 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3922 if (ompt_frame->enter_frame.ptr == NULL)
3923 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3925 OMPT_STORE_RETURN_ADDRESS(global_tid);
3928 __kmp_threads[global_tid]->th.th_ident = loc;
3930 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3931#if OMPT_SUPPORT && OMPT_OPTIONAL
3932 if (ompt_enabled.enabled) {
3933 ompt_frame->enter_frame = ompt_data_none;
3937 }
else if (packed_reduction_method == empty_reduce_block) {
3945 ompt_frame_t *ompt_frame;
3946 if (ompt_enabled.enabled) {
3947 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3948 if (ompt_frame->enter_frame.ptr == NULL)
3949 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3951 OMPT_STORE_RETURN_ADDRESS(global_tid);
3954 __kmp_threads[global_tid]->th.th_ident = loc;
3956 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3957#if OMPT_SUPPORT && OMPT_OPTIONAL
3958 if (ompt_enabled.enabled) {
3959 ompt_frame->enter_frame = ompt_data_none;
3963 }
else if (packed_reduction_method == atomic_reduce_block) {
3966 ompt_frame_t *ompt_frame;
3967 if (ompt_enabled.enabled) {
3968 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3969 if (ompt_frame->enter_frame.ptr == NULL)
3970 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3972 OMPT_STORE_RETURN_ADDRESS(global_tid);
3976 __kmp_threads[global_tid]->th.th_ident = loc;
3978 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3979#if OMPT_SUPPORT && OMPT_OPTIONAL
3980 if (ompt_enabled.enabled) {
3981 ompt_frame->enter_frame = ompt_data_none;
3985 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3986 tree_reduce_block)) {
3989 __kmp_end_split_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3997 if (teams_swapped) {
3998 __kmp_restore_swapped_teams(th, team, task_state);
4001 if (__kmp_env_consistency_check)
4002 __kmp_pop_sync(global_tid, ct_reduce, loc);
4004 KA_TRACE(10, (
"__kmpc_end_reduce() exit: called T#%d: method %08x\n",
4005 global_tid, packed_reduction_method));
4010#undef __KMP_GET_REDUCTION_METHOD
4011#undef __KMP_SET_REDUCTION_METHOD
4015kmp_uint64 __kmpc_get_taskid() {
4020 gtid = __kmp_get_gtid();
4024 thread = __kmp_thread_from_gtid(gtid);
4025 return thread->th.th_current_task->td_task_id;
4029kmp_uint64 __kmpc_get_parent_taskid() {
4033 kmp_taskdata_t *parent_task;
4035 gtid = __kmp_get_gtid();
4039 thread = __kmp_thread_from_gtid(gtid);
4040 parent_task = thread->th.th_current_task->td_parent;
4041 return (parent_task == NULL ? 0 : parent_task->td_task_id);
4057 const struct kmp_dim *dims) {
4058 __kmp_assert_valid_gtid(gtid);
4060 kmp_int64 last, trace_count;
4061 kmp_info_t *th = __kmp_threads[gtid];
4062 kmp_team_t *team = th->th.th_team;
4064 kmp_disp_t *pr_buf = th->th.th_dispatch;
4065 dispatch_shared_info_t *sh_buf;
4069 (
"__kmpc_doacross_init() enter: called T#%d, num dims %d, active %d\n",
4070 gtid, num_dims, !team->t.t_serialized));
4071 KMP_DEBUG_ASSERT(dims != NULL);
4072 KMP_DEBUG_ASSERT(num_dims > 0);
4074 if (team->t.t_serialized) {
4075 KA_TRACE(20, (
"__kmpc_doacross_init() exit: serialized team\n"));
4078 KMP_DEBUG_ASSERT(team->t.t_nproc > 1);
4079 idx = pr_buf->th_doacross_buf_idx++;
4081 sh_buf = &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4084 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info == NULL);
4085 pr_buf->th_doacross_info = (kmp_int64 *)__kmp_thread_malloc(
4086 th,
sizeof(kmp_int64) * (4 * num_dims + 1));
4087 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4088 pr_buf->th_doacross_info[0] =
4089 (kmp_int64)num_dims;
4092 pr_buf->th_doacross_info[1] = (kmp_int64)&sh_buf->doacross_num_done;
4093 pr_buf->th_doacross_info[2] = dims[0].lo;
4094 pr_buf->th_doacross_info[3] = dims[0].up;
4095 pr_buf->th_doacross_info[4] = dims[0].st;
4097 for (j = 1; j < num_dims; ++j) {
4100 if (dims[j].st == 1) {
4102 range_length = dims[j].up - dims[j].lo + 1;
4104 if (dims[j].st > 0) {
4105 KMP_DEBUG_ASSERT(dims[j].up > dims[j].lo);
4106 range_length = (kmp_uint64)(dims[j].up - dims[j].lo) / dims[j].st + 1;
4108 KMP_DEBUG_ASSERT(dims[j].lo > dims[j].up);
4110 (kmp_uint64)(dims[j].lo - dims[j].up) / (-dims[j].st) + 1;
4113 pr_buf->th_doacross_info[last++] = range_length;
4114 pr_buf->th_doacross_info[last++] = dims[j].lo;
4115 pr_buf->th_doacross_info[last++] = dims[j].up;
4116 pr_buf->th_doacross_info[last++] = dims[j].st;
4121 if (dims[0].st == 1) {
4122 trace_count = dims[0].up - dims[0].lo + 1;
4123 }
else if (dims[0].st > 0) {
4124 KMP_DEBUG_ASSERT(dims[0].up > dims[0].lo);
4125 trace_count = (kmp_uint64)(dims[0].up - dims[0].lo) / dims[0].st + 1;
4127 KMP_DEBUG_ASSERT(dims[0].lo > dims[0].up);
4128 trace_count = (kmp_uint64)(dims[0].lo - dims[0].up) / (-dims[0].st) + 1;
4130 for (j = 1; j < num_dims; ++j) {
4131 trace_count *= pr_buf->th_doacross_info[4 * j + 1];
4133 KMP_DEBUG_ASSERT(trace_count > 0);
4137 if (idx != sh_buf->doacross_buf_idx) {
4139 __kmp_wait_4((
volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
4146 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET32(
4147 (
volatile kmp_int32 *)&sh_buf->doacross_flags, NULL, 1);
4149 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET64(
4150 (
volatile kmp_int64 *)&sh_buf->doacross_flags, NULL, 1LL);
4152 if (flags == NULL) {
4155 (size_t)trace_count / 8 + 8;
4156 flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
4158 sh_buf->doacross_flags = flags;
4159 }
else if (flags == (kmp_uint32 *)1) {
4162 while (*(
volatile kmp_int32 *)&sh_buf->doacross_flags == 1)
4164 while (*(
volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
4171 KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1);
4172 pr_buf->th_doacross_flags =
4173 sh_buf->doacross_flags;
4175 KA_TRACE(20, (
"__kmpc_doacross_init() exit: T#%d\n", gtid));
4178void __kmpc_doacross_wait(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4179 __kmp_assert_valid_gtid(gtid);
4183 kmp_int64 iter_number;
4184 kmp_info_t *th = __kmp_threads[gtid];
4185 kmp_team_t *team = th->th.th_team;
4187 kmp_int64 lo, up, st;
4189 KA_TRACE(20, (
"__kmpc_doacross_wait() enter: called T#%d\n", gtid));
4190 if (team->t.t_serialized) {
4191 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: serialized team\n"));
4196 pr_buf = th->th.th_dispatch;
4197 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4198 num_dims = (size_t)pr_buf->th_doacross_info[0];
4199 lo = pr_buf->th_doacross_info[2];
4200 up = pr_buf->th_doacross_info[3];
4201 st = pr_buf->th_doacross_info[4];
4202#if OMPT_SUPPORT && OMPT_OPTIONAL
4203 ompt_dependence_t deps[num_dims];
4206 if (vec[0] < lo || vec[0] > up) {
4207 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4208 "bounds [%lld,%lld]\n",
4209 gtid, vec[0], lo, up));
4212 iter_number = vec[0] - lo;
4213 }
else if (st > 0) {
4214 if (vec[0] < lo || vec[0] > up) {
4215 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4216 "bounds [%lld,%lld]\n",
4217 gtid, vec[0], lo, up));
4220 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4222 if (vec[0] > lo || vec[0] < up) {
4223 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4224 "bounds [%lld,%lld]\n",
4225 gtid, vec[0], lo, up));
4228 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4230#if OMPT_SUPPORT && OMPT_OPTIONAL
4231 deps[0].variable.value = iter_number;
4232 deps[0].dependence_type = ompt_dependence_type_sink;
4234 for (i = 1; i < num_dims; ++i) {
4237 ln = pr_buf->th_doacross_info[j + 1];
4238 lo = pr_buf->th_doacross_info[j + 2];
4239 up = pr_buf->th_doacross_info[j + 3];
4240 st = pr_buf->th_doacross_info[j + 4];
4242 if (vec[i] < lo || vec[i] > up) {
4243 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4244 "bounds [%lld,%lld]\n",
4245 gtid, vec[i], lo, up));
4249 }
else if (st > 0) {
4250 if (vec[i] < lo || vec[i] > up) {
4251 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4252 "bounds [%lld,%lld]\n",
4253 gtid, vec[i], lo, up));
4256 iter = (kmp_uint64)(vec[i] - lo) / st;
4258 if (vec[i] > lo || vec[i] < up) {
4259 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4260 "bounds [%lld,%lld]\n",
4261 gtid, vec[i], lo, up));
4264 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4266 iter_number = iter + ln * iter_number;
4267#if OMPT_SUPPORT && OMPT_OPTIONAL
4268 deps[i].variable.value = iter;
4269 deps[i].dependence_type = ompt_dependence_type_sink;
4272 shft = iter_number % 32;
4275 while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
4279#if OMPT_SUPPORT && OMPT_OPTIONAL
4280 if (ompt_enabled.ompt_callback_dependences) {
4281 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4282 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4286 (
"__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
4287 gtid, (iter_number << 5) + shft));
4290void __kmpc_doacross_post(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4291 __kmp_assert_valid_gtid(gtid);
4295 kmp_int64 iter_number;
4296 kmp_info_t *th = __kmp_threads[gtid];
4297 kmp_team_t *team = th->th.th_team;
4301 KA_TRACE(20, (
"__kmpc_doacross_post() enter: called T#%d\n", gtid));
4302 if (team->t.t_serialized) {
4303 KA_TRACE(20, (
"__kmpc_doacross_post() exit: serialized team\n"));
4309 pr_buf = th->th.th_dispatch;
4310 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4311 num_dims = (size_t)pr_buf->th_doacross_info[0];
4312 lo = pr_buf->th_doacross_info[2];
4313 st = pr_buf->th_doacross_info[4];
4314#if OMPT_SUPPORT && OMPT_OPTIONAL
4315 ompt_dependence_t deps[num_dims];
4318 iter_number = vec[0] - lo;
4319 }
else if (st > 0) {
4320 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4322 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4324#if OMPT_SUPPORT && OMPT_OPTIONAL
4325 deps[0].variable.value = iter_number;
4326 deps[0].dependence_type = ompt_dependence_type_source;
4328 for (i = 1; i < num_dims; ++i) {
4331 ln = pr_buf->th_doacross_info[j + 1];
4332 lo = pr_buf->th_doacross_info[j + 2];
4333 st = pr_buf->th_doacross_info[j + 4];
4336 }
else if (st > 0) {
4337 iter = (kmp_uint64)(vec[i] - lo) / st;
4339 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4341 iter_number = iter + ln * iter_number;
4342#if OMPT_SUPPORT && OMPT_OPTIONAL
4343 deps[i].variable.value = iter;
4344 deps[i].dependence_type = ompt_dependence_type_source;
4347#if OMPT_SUPPORT && OMPT_OPTIONAL
4348 if (ompt_enabled.ompt_callback_dependences) {
4349 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4350 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4353 shft = iter_number % 32;
4357 if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
4358 KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
4359 KA_TRACE(20, (
"__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
4360 (iter_number << 5) + shft));
4363void __kmpc_doacross_fini(
ident_t *loc,
int gtid) {
4364 __kmp_assert_valid_gtid(gtid);
4366 kmp_info_t *th = __kmp_threads[gtid];
4367 kmp_team_t *team = th->th.th_team;
4368 kmp_disp_t *pr_buf = th->th.th_dispatch;
4370 KA_TRACE(20, (
"__kmpc_doacross_fini() enter: called T#%d\n", gtid));
4371 if (team->t.t_serialized) {
4372 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: serialized team %p\n", team));
4376 KMP_TEST_THEN_INC32((kmp_uintptr_t)(pr_buf->th_doacross_info[1])) + 1;
4377 if (num_done == th->th.th_team_nproc) {
4379 int idx = pr_buf->th_doacross_buf_idx - 1;
4380 dispatch_shared_info_t *sh_buf =
4381 &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4382 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info[1] ==
4383 (kmp_int64)&sh_buf->doacross_num_done);
4384 KMP_DEBUG_ASSERT(num_done == sh_buf->doacross_num_done);
4385 KMP_DEBUG_ASSERT(idx == sh_buf->doacross_buf_idx);
4386 __kmp_thread_free(th, CCAST(kmp_uint32 *, sh_buf->doacross_flags));
4387 sh_buf->doacross_flags = NULL;
4388 sh_buf->doacross_num_done = 0;
4389 sh_buf->doacross_buf_idx +=
4390 __kmp_dispatch_num_buffers;
4393 pr_buf->th_doacross_flags = NULL;
4394 __kmp_thread_free(th, (
void *)pr_buf->th_doacross_info);
4395 pr_buf->th_doacross_info = NULL;
4396 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: T#%d\n", gtid));
4400void *omp_alloc(
size_t size, omp_allocator_handle_t allocator) {
4401 return __kmp_alloc(__kmp_entry_gtid(), 0, size, allocator);
4404void *omp_aligned_alloc(
size_t align,
size_t size,
4405 omp_allocator_handle_t allocator) {
4406 return __kmp_alloc(__kmp_entry_gtid(), align, size, allocator);
4409void *omp_calloc(
size_t nmemb,
size_t size, omp_allocator_handle_t allocator) {
4410 return __kmp_calloc(__kmp_entry_gtid(), 0, nmemb, size, allocator);
4413void *omp_aligned_calloc(
size_t align,
size_t nmemb,
size_t size,
4414 omp_allocator_handle_t allocator) {
4415 return __kmp_calloc(__kmp_entry_gtid(), align, nmemb, size, allocator);
4418void *omp_realloc(
void *ptr,
size_t size, omp_allocator_handle_t allocator,
4419 omp_allocator_handle_t free_allocator) {
4420 return __kmp_realloc(__kmp_entry_gtid(), ptr, size, allocator,
4424void omp_free(
void *ptr, omp_allocator_handle_t allocator) {
4425 ___kmpc_free(__kmp_entry_gtid(), ptr, allocator);
4429int __kmpc_get_target_offload(
void) {
4430 if (!__kmp_init_serial) {
4431 __kmp_serial_initialize();
4433 return __kmp_target_offload;
4436int __kmpc_pause_resource(kmp_pause_status_t level) {
4437 if (!__kmp_init_serial) {
4440 return __kmp_pause_resource(level);
4443void __kmpc_error(
ident_t *loc,
int severity,
const char *message) {
4444 if (!__kmp_init_serial)
4445 __kmp_serial_initialize();
4447 KMP_ASSERT(severity == severity_warning || severity == severity_fatal);
4450 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_error) {
4451 ompt_callbacks.ompt_callback(ompt_callback_error)(
4452 (ompt_severity_t)severity, message, KMP_STRLEN(message),
4453 OMPT_GET_RETURN_ADDRESS(0));
4459 kmp_str_loc_t str_loc = __kmp_str_loc_init(loc->
psource,
false);
4461 __kmp_str_format(
"%s:%s:%s", str_loc.file, str_loc.line, str_loc.col);
4462 __kmp_str_loc_free(&str_loc);
4464 src_loc = __kmp_str_format(
"unknown");
4467 if (severity == severity_warning)
4468 KMP_WARNING(UserDirectedWarning, src_loc, message);
4470 KMP_FATAL(UserDirectedError, src_loc, message);
4472 __kmp_str_free(&src_loc);
4476void __kmpc_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4478#if OMPT_SUPPORT && OMPT_OPTIONAL
4479 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4480 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4481 int tid = __kmp_tid_from_gtid(gtid);
4482 ompt_callbacks.ompt_callback(ompt_callback_work)(
4483 ompt_work_scope, ompt_scope_begin,
4484 &(team->t.ompt_team_info.parallel_data),
4485 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4486 OMPT_GET_RETURN_ADDRESS(0));
4492void __kmpc_end_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4494#if OMPT_SUPPORT && OMPT_OPTIONAL
4495 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4496 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4497 int tid = __kmp_tid_from_gtid(gtid);
4498 ompt_callbacks.ompt_callback(ompt_callback_work)(
4499 ompt_work_scope, ompt_scope_end,
4500 &(team->t.ompt_team_info.parallel_data),
4501 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4502 OMPT_GET_RETURN_ADDRESS(0));
4507#ifdef KMP_USE_VERSION_SYMBOLS
4516#ifdef omp_set_affinity_format
4517#undef omp_set_affinity_format
4519#ifdef omp_get_affinity_format
4520#undef omp_get_affinity_format
4522#ifdef omp_display_affinity
4523#undef omp_display_affinity
4525#ifdef omp_capture_affinity
4526#undef omp_capture_affinity
4528KMP_VERSION_OMPC_SYMBOL(ompc_set_affinity_format, omp_set_affinity_format, 50,
4530KMP_VERSION_OMPC_SYMBOL(ompc_get_affinity_format, omp_get_affinity_format, 50,
4532KMP_VERSION_OMPC_SYMBOL(ompc_display_affinity, omp_display_affinity, 50,
4534KMP_VERSION_OMPC_SYMBOL(ompc_capture_affinity, omp_capture_affinity, 50,
@ KMP_IDENT_WORK_SECTIONS
@ KMP_IDENT_WORK_DISTRIBUTE
kmp_int32 __kmpc_ok_to_fork(ident_t *loc)
void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)
void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_threads)
void __kmpc_serialized_parallel(ident_t *loc, kmp_int32 global_tid)
void __kmpc_push_num_teams(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams, kmp_int32 num_threads)
void __kmpc_fork_call(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)
void __kmpc_end_serialized_parallel(ident_t *loc, kmp_int32 global_tid)
void(* kmpc_micro)(kmp_int32 *global_tid, kmp_int32 *bound_tid,...)
void __kmpc_push_num_teams_51(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams_lb, kmp_int32 num_teams_ub, kmp_int32 num_threads)
void __kmpc_begin(ident_t *loc, kmp_int32 flags)
void __kmpc_end(ident_t *loc)
#define KMP_COUNT_BLOCK(name)
Increments specified counter (name).
stats_state_e
the states which a thread can be in
void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
void __kmpc_end_barrier_master(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_barrier_master_nowait(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void(*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck)
void __kmpc_barrier(ident_t *loc, kmp_int32 global_tid)
void __kmpc_flush(ident_t *loc)
kmp_int32 __kmpc_barrier_master(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void(*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck)
void __kmpc_copyprivate(ident_t *loc, kmp_int32 gtid, size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *), kmp_int32 didit)
void * __kmpc_copyprivate_light(ident_t *loc, kmp_int32 gtid, void *cpy_data)
kmp_int32 __kmpc_global_num_threads(ident_t *loc)
kmp_int32 __kmpc_global_thread_num(ident_t *loc)
kmp_int32 __kmpc_in_parallel(ident_t *loc)
kmp_int32 __kmpc_bound_thread_num(ident_t *loc)
kmp_int32 __kmpc_bound_num_threads(ident_t *loc)
void __kmpc_end_ordered(ident_t *loc, kmp_int32 gtid)
void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)
void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_masked(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_master(ident_t *loc, kmp_int32 global_tid)
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)