16 #include "kmp_error.h"
20 #include "kmp_stats.h"
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);
239 void __kmpc_pop_num_threads(
ident_t *loc, kmp_int32 global_tid) {
240 KA_TRACE(20, (
"__kmpc_pop_num_threads: enter\n"));
244 void __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);
297 #if INCLUDE_SSC_MARKS
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,
304 #if INCLUDE_SSC_MARKS
307 __kmp_join_call(loc, gtid
317 if (ompt_enabled.enabled) {
318 ompt_frame->enter_frame = ompt_data_none;
323 #if KMP_STATS_ENABLED
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);
397 #if KMP_STATS_ENABLED
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;
463 #if KMP_STATS_ENABLED
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();
477 int __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));
636 if (__kmp_tasking_mode != tskm_immediate_exec) {
637 KA_TRACE(20, (
"__kmpc_end_serialized_parallel: T#%d decreasing nesting "
638 "depth of serial team %p to %d\n",
639 global_tid, serial_team, serial_team->t.t_serialized));
643 serial_team->t.t_level--;
644 if (__kmp_env_consistency_check)
645 __kmp_pop_parallel(global_tid, NULL);
647 if (ompt_enabled.enabled)
648 this_thr->th.ompt_thread_info.state =
649 ((this_thr->th.th_team_serialized) ? ompt_state_work_serial
650 : ompt_state_work_parallel);
663 KC_TRACE(10, (
"__kmpc_flush: called\n"));
668 #if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
682 if (!__kmp_cpuinfo.initialized) {
683 __kmp_query_cpuid(&__kmp_cpuinfo);
685 if (!__kmp_cpuinfo.flags.sse2) {
690 #elif KMP_COMPILER_MSVC
693 __sync_synchronize();
697 #elif (KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || KMP_ARCH_MIPS64 || \
703 #error Unknown or unsupported architecture
706 #if OMPT_SUPPORT && OMPT_OPTIONAL
707 if (ompt_enabled.ompt_callback_flush) {
708 ompt_callbacks.ompt_callback(ompt_callback_flush)(
709 __ompt_get_thread_data_internal(), OMPT_GET_RETURN_ADDRESS(0));
724 KC_TRACE(10, (
"__kmpc_barrier: called T#%d\n", global_tid));
725 __kmp_assert_valid_gtid(global_tid);
727 if (!TCR_4(__kmp_init_parallel))
728 __kmp_parallel_initialize();
730 __kmp_resume_if_soft_paused();
732 if (__kmp_env_consistency_check) {
734 KMP_WARNING(ConstructIdentInvalid);
736 __kmp_check_barrier(global_tid, ct_barrier, loc);
740 ompt_frame_t *ompt_frame;
741 if (ompt_enabled.enabled) {
742 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
743 if (ompt_frame->enter_frame.ptr == NULL)
744 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
746 OMPT_STORE_RETURN_ADDRESS(global_tid);
748 __kmp_threads[global_tid]->th.th_ident = loc;
756 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
757 #if OMPT_SUPPORT && OMPT_OPTIONAL
758 if (ompt_enabled.enabled) {
759 ompt_frame->enter_frame = ompt_data_none;
774 KC_TRACE(10, (
"__kmpc_master: called T#%d\n", global_tid));
775 __kmp_assert_valid_gtid(global_tid);
777 if (!TCR_4(__kmp_init_parallel))
778 __kmp_parallel_initialize();
780 __kmp_resume_if_soft_paused();
782 if (KMP_MASTER_GTID(global_tid)) {
784 KMP_PUSH_PARTITIONED_TIMER(OMP_master);
788 #if OMPT_SUPPORT && OMPT_OPTIONAL
790 if (ompt_enabled.ompt_callback_masked) {
791 kmp_info_t *this_thr = __kmp_threads[global_tid];
792 kmp_team_t *team = this_thr->th.th_team;
794 int tid = __kmp_tid_from_gtid(global_tid);
795 ompt_callbacks.ompt_callback(ompt_callback_masked)(
796 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
797 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
798 OMPT_GET_RETURN_ADDRESS(0));
803 if (__kmp_env_consistency_check) {
804 #if KMP_USE_DYNAMIC_LOCK
806 __kmp_push_sync(global_tid, ct_master, loc, NULL, 0);
808 __kmp_check_sync(global_tid, ct_master, loc, NULL, 0);
811 __kmp_push_sync(global_tid, ct_master, loc, NULL);
813 __kmp_check_sync(global_tid, ct_master, loc, NULL);
829 KC_TRACE(10, (
"__kmpc_end_master: called T#%d\n", global_tid));
830 __kmp_assert_valid_gtid(global_tid);
831 KMP_DEBUG_ASSERT(KMP_MASTER_GTID(global_tid));
832 KMP_POP_PARTITIONED_TIMER();
834 #if OMPT_SUPPORT && OMPT_OPTIONAL
835 kmp_info_t *this_thr = __kmp_threads[global_tid];
836 kmp_team_t *team = this_thr->th.th_team;
837 if (ompt_enabled.ompt_callback_masked) {
838 int tid = __kmp_tid_from_gtid(global_tid);
839 ompt_callbacks.ompt_callback(ompt_callback_masked)(
840 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
841 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
842 OMPT_GET_RETURN_ADDRESS(0));
846 if (__kmp_env_consistency_check) {
847 if (KMP_MASTER_GTID(global_tid))
848 __kmp_pop_sync(global_tid, ct_master, loc);
863 KC_TRACE(10, (
"__kmpc_masked: called T#%d\n", global_tid));
864 __kmp_assert_valid_gtid(global_tid);
866 if (!TCR_4(__kmp_init_parallel))
867 __kmp_parallel_initialize();
869 __kmp_resume_if_soft_paused();
871 tid = __kmp_tid_from_gtid(global_tid);
874 KMP_PUSH_PARTITIONED_TIMER(OMP_masked);
878 #if OMPT_SUPPORT && OMPT_OPTIONAL
880 if (ompt_enabled.ompt_callback_masked) {
881 kmp_info_t *this_thr = __kmp_threads[global_tid];
882 kmp_team_t *team = this_thr->th.th_team;
883 ompt_callbacks.ompt_callback(ompt_callback_masked)(
884 ompt_scope_begin, &(team->t.ompt_team_info.parallel_data),
885 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
886 OMPT_GET_RETURN_ADDRESS(0));
891 if (__kmp_env_consistency_check) {
892 #if KMP_USE_DYNAMIC_LOCK
894 __kmp_push_sync(global_tid, ct_masked, loc, NULL, 0);
896 __kmp_check_sync(global_tid, ct_masked, loc, NULL, 0);
899 __kmp_push_sync(global_tid, ct_masked, loc, NULL);
901 __kmp_check_sync(global_tid, ct_masked, loc, NULL);
917 KC_TRACE(10, (
"__kmpc_end_masked: called T#%d\n", global_tid));
918 __kmp_assert_valid_gtid(global_tid);
919 KMP_POP_PARTITIONED_TIMER();
921 #if OMPT_SUPPORT && OMPT_OPTIONAL
922 kmp_info_t *this_thr = __kmp_threads[global_tid];
923 kmp_team_t *team = this_thr->th.th_team;
924 if (ompt_enabled.ompt_callback_masked) {
925 int tid = __kmp_tid_from_gtid(global_tid);
926 ompt_callbacks.ompt_callback(ompt_callback_masked)(
927 ompt_scope_end, &(team->t.ompt_team_info.parallel_data),
928 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
929 OMPT_GET_RETURN_ADDRESS(0));
933 if (__kmp_env_consistency_check) {
934 __kmp_pop_sync(global_tid, ct_masked, loc);
948 KMP_DEBUG_ASSERT(__kmp_init_serial);
950 KC_TRACE(10, (
"__kmpc_ordered: called T#%d\n", gtid));
951 __kmp_assert_valid_gtid(gtid);
953 if (!TCR_4(__kmp_init_parallel))
954 __kmp_parallel_initialize();
956 __kmp_resume_if_soft_paused();
959 __kmp_itt_ordered_prep(gtid);
963 th = __kmp_threads[gtid];
965 #if OMPT_SUPPORT && OMPT_OPTIONAL
969 OMPT_STORE_RETURN_ADDRESS(gtid);
970 if (ompt_enabled.enabled) {
971 team = __kmp_team_from_gtid(gtid);
972 lck = (ompt_wait_id_t)(uintptr_t)&team->t.t_ordered.dt.t_value;
974 th->th.ompt_thread_info.wait_id = lck;
975 th->th.ompt_thread_info.state = ompt_state_wait_ordered;
978 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
979 if (ompt_enabled.ompt_callback_mutex_acquire) {
980 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
981 ompt_mutex_ordered, omp_lock_hint_none, kmp_mutex_impl_spin, lck,
987 if (th->th.th_dispatch->th_deo_fcn != 0)
988 (*th->th.th_dispatch->th_deo_fcn)(>id, &cid, loc);
990 __kmp_parallel_deo(>id, &cid, loc);
992 #if OMPT_SUPPORT && OMPT_OPTIONAL
993 if (ompt_enabled.enabled) {
995 th->th.ompt_thread_info.state = ompt_state_work_parallel;
996 th->th.ompt_thread_info.wait_id = 0;
999 if (ompt_enabled.ompt_callback_mutex_acquired) {
1000 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1001 ompt_mutex_ordered, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1007 __kmp_itt_ordered_start(gtid);
1022 KC_TRACE(10, (
"__kmpc_end_ordered: called T#%d\n", gtid));
1023 __kmp_assert_valid_gtid(gtid);
1026 __kmp_itt_ordered_end(gtid);
1030 th = __kmp_threads[gtid];
1032 if (th->th.th_dispatch->th_dxo_fcn != 0)
1033 (*th->th.th_dispatch->th_dxo_fcn)(>id, &cid, loc);
1035 __kmp_parallel_dxo(>id, &cid, loc);
1037 #if OMPT_SUPPORT && OMPT_OPTIONAL
1038 OMPT_STORE_RETURN_ADDRESS(gtid);
1039 if (ompt_enabled.ompt_callback_mutex_released) {
1040 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1042 (ompt_wait_id_t)(uintptr_t)&__kmp_team_from_gtid(gtid)
1043 ->t.t_ordered.dt.t_value,
1044 OMPT_LOAD_RETURN_ADDRESS(gtid));
1049 #if KMP_USE_DYNAMIC_LOCK
1051 static __forceinline
void
1052 __kmp_init_indirect_csptr(kmp_critical_name *crit,
ident_t const *loc,
1053 kmp_int32 gtid, kmp_indirect_locktag_t tag) {
1057 kmp_indirect_lock_t **lck;
1058 lck = (kmp_indirect_lock_t **)crit;
1059 kmp_indirect_lock_t *ilk = __kmp_allocate_indirect_lock(&idx, gtid, tag);
1060 KMP_I_LOCK_FUNC(ilk, init)(ilk->lock);
1061 KMP_SET_I_LOCK_LOCATION(ilk, loc);
1062 KMP_SET_I_LOCK_FLAGS(ilk, kmp_lf_critical_section);
1064 (
"__kmp_init_indirect_csptr: initialized indirect lock #%d\n", tag));
1066 __kmp_itt_critical_creating(ilk->lock, loc);
1068 int status = KMP_COMPARE_AND_STORE_PTR(lck,
nullptr, ilk);
1071 __kmp_itt_critical_destroyed(ilk->lock);
1077 KMP_DEBUG_ASSERT(*lck != NULL);
1081 #define KMP_ACQUIRE_TAS_LOCK(lock, gtid) \
1083 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1084 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1085 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1086 if (KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1087 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)) { \
1089 KMP_FSYNC_PREPARE(l); \
1090 KMP_INIT_YIELD(spins); \
1091 kmp_backoff_t backoff = __kmp_spin_backoff_params; \
1093 if (TCR_4(__kmp_nth) > \
1094 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { \
1097 KMP_YIELD_SPIN(spins); \
1099 __kmp_spin_backoff(&backoff); \
1101 KMP_ATOMIC_LD_RLX(&l->lk.poll) != tas_free || \
1102 !__kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy)); \
1104 KMP_FSYNC_ACQUIRED(l); \
1108 #define KMP_TEST_TAS_LOCK(lock, gtid, rc) \
1110 kmp_tas_lock_t *l = (kmp_tas_lock_t *)lock; \
1111 kmp_int32 tas_free = KMP_LOCK_FREE(tas); \
1112 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); \
1113 rc = KMP_ATOMIC_LD_RLX(&l->lk.poll) == tas_free && \
1114 __kmp_atomic_compare_store_acq(&l->lk.poll, tas_free, tas_busy); \
1118 #define KMP_RELEASE_TAS_LOCK(lock, gtid) \
1119 { KMP_ATOMIC_ST_REL(&((kmp_tas_lock_t *)lock)->lk.poll, KMP_LOCK_FREE(tas)); }
1123 #include <sys/syscall.h>
1126 #define FUTEX_WAIT 0
1129 #define FUTEX_WAKE 1
1133 #define KMP_ACQUIRE_FUTEX_LOCK(lock, gtid) \
1135 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1136 kmp_int32 gtid_code = (gtid + 1) << 1; \
1138 KMP_FSYNC_PREPARE(ftx); \
1139 kmp_int32 poll_val; \
1140 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( \
1141 &(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1142 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { \
1143 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; \
1145 if (!KMP_COMPARE_AND_STORE_RET32(&(ftx->lk.poll), poll_val, \
1147 KMP_LOCK_BUSY(1, futex))) { \
1150 poll_val |= KMP_LOCK_BUSY(1, futex); \
1153 if ((rc = syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAIT, poll_val, \
1154 NULL, NULL, 0)) != 0) { \
1159 KMP_FSYNC_ACQUIRED(ftx); \
1163 #define KMP_TEST_FUTEX_LOCK(lock, gtid, rc) \
1165 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1166 if (KMP_COMPARE_AND_STORE_ACQ32(&(ftx->lk.poll), KMP_LOCK_FREE(futex), \
1167 KMP_LOCK_BUSY(gtid + 1 << 1, futex))) { \
1168 KMP_FSYNC_ACQUIRED(ftx); \
1176 #define KMP_RELEASE_FUTEX_LOCK(lock, gtid) \
1178 kmp_futex_lock_t *ftx = (kmp_futex_lock_t *)lock; \
1180 KMP_FSYNC_RELEASING(ftx); \
1181 kmp_int32 poll_val = \
1182 KMP_XCHG_FIXED32(&(ftx->lk.poll), KMP_LOCK_FREE(futex)); \
1183 if (KMP_LOCK_STRIP(poll_val) & 1) { \
1184 syscall(__NR_futex, &(ftx->lk.poll), FUTEX_WAKE, \
1185 KMP_LOCK_BUSY(1, futex), NULL, NULL, 0); \
1188 KMP_YIELD_OVERSUB(); \
1195 static kmp_user_lock_p __kmp_get_critical_section_ptr(kmp_critical_name *crit,
1198 kmp_user_lock_p *lck_pp = (kmp_user_lock_p *)crit;
1201 kmp_user_lock_p lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1208 lck = __kmp_user_lock_allocate(&idx, gtid, kmp_lf_critical_section);
1209 __kmp_init_user_lock_with_checks(lck);
1210 __kmp_set_user_lock_location(lck, loc);
1212 __kmp_itt_critical_creating(lck);
1223 int status = KMP_COMPARE_AND_STORE_PTR(lck_pp, 0, lck);
1228 __kmp_itt_critical_destroyed(lck);
1232 __kmp_destroy_user_lock_with_checks(lck);
1233 __kmp_user_lock_free(&idx, gtid, lck);
1234 lck = (kmp_user_lock_p)TCR_PTR(*lck_pp);
1235 KMP_DEBUG_ASSERT(lck != NULL);
1254 kmp_critical_name *crit) {
1255 #if KMP_USE_DYNAMIC_LOCK
1256 #if OMPT_SUPPORT && OMPT_OPTIONAL
1257 OMPT_STORE_RETURN_ADDRESS(global_tid);
1262 #if OMPT_SUPPORT && OMPT_OPTIONAL
1263 ompt_state_t prev_state = ompt_state_undefined;
1264 ompt_thread_info_t ti;
1266 kmp_user_lock_p lck;
1268 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1269 __kmp_assert_valid_gtid(global_tid);
1273 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1274 KMP_CHECK_USER_LOCK_INIT();
1276 if ((__kmp_user_lock_kind == lk_tas) &&
1277 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1278 lck = (kmp_user_lock_p)crit;
1281 else if ((__kmp_user_lock_kind == lk_futex) &&
1282 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1283 lck = (kmp_user_lock_p)crit;
1287 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
1290 if (__kmp_env_consistency_check)
1291 __kmp_push_sync(global_tid, ct_critical, loc, lck);
1299 __kmp_itt_critical_acquiring(lck);
1301 #if OMPT_SUPPORT && OMPT_OPTIONAL
1302 OMPT_STORE_RETURN_ADDRESS(gtid);
1303 void *codeptr_ra = NULL;
1304 if (ompt_enabled.enabled) {
1305 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1307 prev_state = ti.state;
1308 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1309 ti.state = ompt_state_wait_critical;
1312 codeptr_ra = OMPT_LOAD_RETURN_ADDRESS(gtid);
1313 if (ompt_enabled.ompt_callback_mutex_acquire) {
1314 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1315 ompt_mutex_critical, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
1316 (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1322 __kmp_acquire_user_lock_with_checks(lck, global_tid);
1325 __kmp_itt_critical_acquired(lck);
1327 #if OMPT_SUPPORT && OMPT_OPTIONAL
1328 if (ompt_enabled.enabled) {
1330 ti.state = prev_state;
1334 if (ompt_enabled.ompt_callback_mutex_acquired) {
1335 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1336 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr_ra);
1340 KMP_POP_PARTITIONED_TIMER();
1342 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1343 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1347 #if KMP_USE_DYNAMIC_LOCK
1350 static __forceinline kmp_dyna_lockseq_t __kmp_map_hint_to_lock(uintptr_t hint) {
1352 #define KMP_TSX_LOCK(seq) lockseq_##seq
1354 #define KMP_TSX_LOCK(seq) __kmp_user_lock_seq
1357 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
1358 #define KMP_CPUINFO_RTM (__kmp_cpuinfo.flags.rtm)
1360 #define KMP_CPUINFO_RTM 0
1364 if (hint & kmp_lock_hint_hle)
1365 return KMP_TSX_LOCK(hle);
1366 if (hint & kmp_lock_hint_rtm)
1367 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_queuing) : __kmp_user_lock_seq;
1368 if (hint & kmp_lock_hint_adaptive)
1369 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
1372 if ((hint & omp_lock_hint_contended) && (hint & omp_lock_hint_uncontended))
1373 return __kmp_user_lock_seq;
1374 if ((hint & omp_lock_hint_speculative) &&
1375 (hint & omp_lock_hint_nonspeculative))
1376 return __kmp_user_lock_seq;
1379 if (hint & omp_lock_hint_contended)
1380 return lockseq_queuing;
1383 if ((hint & omp_lock_hint_uncontended) && !(hint & omp_lock_hint_speculative))
1387 if (hint & omp_lock_hint_speculative)
1388 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_spin) : __kmp_user_lock_seq;
1390 return __kmp_user_lock_seq;
1393 #if OMPT_SUPPORT && OMPT_OPTIONAL
1394 #if KMP_USE_DYNAMIC_LOCK
1395 static kmp_mutex_impl_t
1396 __ompt_get_mutex_impl_type(
void *user_lock, kmp_indirect_lock_t *ilock = 0) {
1398 switch (KMP_EXTRACT_D_TAG(user_lock)) {
1403 return kmp_mutex_impl_queuing;
1406 return kmp_mutex_impl_spin;
1409 case locktag_rtm_spin:
1410 return kmp_mutex_impl_speculative;
1413 return kmp_mutex_impl_none;
1415 ilock = KMP_LOOKUP_I_LOCK(user_lock);
1418 switch (ilock->type) {
1420 case locktag_adaptive:
1421 case locktag_rtm_queuing:
1422 return kmp_mutex_impl_speculative;
1424 case locktag_nested_tas:
1425 return kmp_mutex_impl_spin;
1427 case locktag_nested_futex:
1429 case locktag_ticket:
1430 case locktag_queuing:
1432 case locktag_nested_ticket:
1433 case locktag_nested_queuing:
1434 case locktag_nested_drdpa:
1435 return kmp_mutex_impl_queuing;
1437 return kmp_mutex_impl_none;
1442 static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
1443 switch (__kmp_user_lock_kind) {
1445 return kmp_mutex_impl_spin;
1452 return kmp_mutex_impl_queuing;
1455 case lk_rtm_queuing:
1458 return kmp_mutex_impl_speculative;
1461 return kmp_mutex_impl_none;
1481 kmp_critical_name *crit, uint32_t hint) {
1483 kmp_user_lock_p lck;
1484 #if OMPT_SUPPORT && OMPT_OPTIONAL
1485 ompt_state_t prev_state = ompt_state_undefined;
1486 ompt_thread_info_t ti;
1488 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(global_tid);
1490 codeptr = OMPT_GET_RETURN_ADDRESS(0);
1493 KC_TRACE(10, (
"__kmpc_critical: called T#%d\n", global_tid));
1494 __kmp_assert_valid_gtid(global_tid);
1496 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
1498 KMP_PUSH_PARTITIONED_TIMER(OMP_critical_wait);
1499 kmp_dyna_lockseq_t lockseq = __kmp_map_hint_to_lock(hint);
1501 if (KMP_IS_D_LOCK(lockseq)) {
1502 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
1503 KMP_GET_D_TAG(lockseq));
1505 __kmp_init_indirect_csptr(crit, loc, global_tid, KMP_GET_I_TAG(lockseq));
1511 if (KMP_EXTRACT_D_TAG(lk) != 0) {
1512 lck = (kmp_user_lock_p)lk;
1513 if (__kmp_env_consistency_check) {
1514 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1515 __kmp_map_hint_to_lock(hint));
1518 __kmp_itt_critical_acquiring(lck);
1520 #if OMPT_SUPPORT && OMPT_OPTIONAL
1521 if (ompt_enabled.enabled) {
1522 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1524 prev_state = ti.state;
1525 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1526 ti.state = ompt_state_wait_critical;
1529 if (ompt_enabled.ompt_callback_mutex_acquire) {
1530 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1531 ompt_mutex_critical, (
unsigned int)hint,
1532 __ompt_get_mutex_impl_type(crit), (ompt_wait_id_t)(uintptr_t)lck,
1537 #if KMP_USE_INLINED_TAS
1538 if (lockseq == lockseq_tas && !__kmp_env_consistency_check) {
1539 KMP_ACQUIRE_TAS_LOCK(lck, global_tid);
1541 #elif KMP_USE_INLINED_FUTEX
1542 if (lockseq == lockseq_futex && !__kmp_env_consistency_check) {
1543 KMP_ACQUIRE_FUTEX_LOCK(lck, global_tid);
1547 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
1550 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
1552 if (__kmp_env_consistency_check) {
1553 __kmp_push_sync(global_tid, ct_critical, loc, lck,
1554 __kmp_map_hint_to_lock(hint));
1557 __kmp_itt_critical_acquiring(lck);
1559 #if OMPT_SUPPORT && OMPT_OPTIONAL
1560 if (ompt_enabled.enabled) {
1561 ti = __kmp_threads[global_tid]->th.ompt_thread_info;
1563 prev_state = ti.state;
1564 ti.wait_id = (ompt_wait_id_t)(uintptr_t)lck;
1565 ti.state = ompt_state_wait_critical;
1568 if (ompt_enabled.ompt_callback_mutex_acquire) {
1569 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
1570 ompt_mutex_critical, (
unsigned int)hint,
1571 __ompt_get_mutex_impl_type(0, ilk), (ompt_wait_id_t)(uintptr_t)lck,
1576 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
1578 KMP_POP_PARTITIONED_TIMER();
1581 __kmp_itt_critical_acquired(lck);
1583 #if OMPT_SUPPORT && OMPT_OPTIONAL
1584 if (ompt_enabled.enabled) {
1586 ti.state = prev_state;
1590 if (ompt_enabled.ompt_callback_mutex_acquired) {
1591 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
1592 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
1597 KMP_PUSH_PARTITIONED_TIMER(OMP_critical);
1598 KA_TRACE(15, (
"__kmpc_critical: done T#%d\n", global_tid));
1613 kmp_critical_name *crit) {
1614 kmp_user_lock_p lck;
1616 KC_TRACE(10, (
"__kmpc_end_critical: called T#%d\n", global_tid));
1618 #if KMP_USE_DYNAMIC_LOCK
1619 int locktag = KMP_EXTRACT_D_TAG(crit);
1621 lck = (kmp_user_lock_p)crit;
1622 KMP_ASSERT(lck != NULL);
1623 if (__kmp_env_consistency_check) {
1624 __kmp_pop_sync(global_tid, ct_critical, loc);
1627 __kmp_itt_critical_releasing(lck);
1629 #if KMP_USE_INLINED_TAS
1630 if (locktag == locktag_tas && !__kmp_env_consistency_check) {
1631 KMP_RELEASE_TAS_LOCK(lck, global_tid);
1633 #elif KMP_USE_INLINED_FUTEX
1634 if (locktag == locktag_futex && !__kmp_env_consistency_check) {
1635 KMP_RELEASE_FUTEX_LOCK(lck, global_tid);
1639 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
1642 kmp_indirect_lock_t *ilk =
1643 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
1644 KMP_ASSERT(ilk != NULL);
1646 if (__kmp_env_consistency_check) {
1647 __kmp_pop_sync(global_tid, ct_critical, loc);
1650 __kmp_itt_critical_releasing(lck);
1652 KMP_I_LOCK_FUNC(ilk, unset)(lck, global_tid);
1657 if ((__kmp_user_lock_kind == lk_tas) &&
1658 (
sizeof(lck->tas.lk.poll) <= OMP_CRITICAL_SIZE)) {
1659 lck = (kmp_user_lock_p)crit;
1662 else if ((__kmp_user_lock_kind == lk_futex) &&
1663 (
sizeof(lck->futex.lk.poll) <= OMP_CRITICAL_SIZE)) {
1664 lck = (kmp_user_lock_p)crit;
1668 lck = (kmp_user_lock_p)TCR_PTR(*((kmp_user_lock_p *)crit));
1671 KMP_ASSERT(lck != NULL);
1673 if (__kmp_env_consistency_check)
1674 __kmp_pop_sync(global_tid, ct_critical, loc);
1677 __kmp_itt_critical_releasing(lck);
1681 __kmp_release_user_lock_with_checks(lck, global_tid);
1685 #if OMPT_SUPPORT && OMPT_OPTIONAL
1688 OMPT_STORE_RETURN_ADDRESS(global_tid);
1689 if (ompt_enabled.ompt_callback_mutex_released) {
1690 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
1691 ompt_mutex_critical, (ompt_wait_id_t)(uintptr_t)lck,
1692 OMPT_LOAD_RETURN_ADDRESS(0));
1696 KMP_POP_PARTITIONED_TIMER();
1697 KA_TRACE(15, (
"__kmpc_end_critical: done T#%d\n", global_tid));
1711 KC_TRACE(10, (
"__kmpc_barrier_master: called T#%d\n", global_tid));
1712 __kmp_assert_valid_gtid(global_tid);
1714 if (!TCR_4(__kmp_init_parallel))
1715 __kmp_parallel_initialize();
1717 __kmp_resume_if_soft_paused();
1719 if (__kmp_env_consistency_check)
1720 __kmp_check_barrier(global_tid, ct_barrier, loc);
1723 ompt_frame_t *ompt_frame;
1724 if (ompt_enabled.enabled) {
1725 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1726 if (ompt_frame->enter_frame.ptr == NULL)
1727 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1729 OMPT_STORE_RETURN_ADDRESS(global_tid);
1732 __kmp_threads[global_tid]->th.th_ident = loc;
1734 status = __kmp_barrier(bs_plain_barrier, global_tid, TRUE, 0, NULL, NULL);
1735 #if OMPT_SUPPORT && OMPT_OPTIONAL
1736 if (ompt_enabled.enabled) {
1737 ompt_frame->enter_frame = ompt_data_none;
1741 return (status != 0) ? 0 : 1;
1754 KC_TRACE(10, (
"__kmpc_end_barrier_master: called T#%d\n", global_tid));
1755 __kmp_assert_valid_gtid(global_tid);
1756 __kmp_end_split_barrier(bs_plain_barrier, global_tid);
1771 KC_TRACE(10, (
"__kmpc_barrier_master_nowait: called T#%d\n", global_tid));
1772 __kmp_assert_valid_gtid(global_tid);
1774 if (!TCR_4(__kmp_init_parallel))
1775 __kmp_parallel_initialize();
1777 __kmp_resume_if_soft_paused();
1779 if (__kmp_env_consistency_check) {
1781 KMP_WARNING(ConstructIdentInvalid);
1783 __kmp_check_barrier(global_tid, ct_barrier, loc);
1787 ompt_frame_t *ompt_frame;
1788 if (ompt_enabled.enabled) {
1789 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
1790 if (ompt_frame->enter_frame.ptr == NULL)
1791 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
1793 OMPT_STORE_RETURN_ADDRESS(global_tid);
1796 __kmp_threads[global_tid]->th.th_ident = loc;
1798 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
1799 #if OMPT_SUPPORT && OMPT_OPTIONAL
1800 if (ompt_enabled.enabled) {
1801 ompt_frame->enter_frame = ompt_data_none;
1807 if (__kmp_env_consistency_check) {
1813 __kmp_pop_sync(global_tid, ct_master, loc);
1833 __kmp_assert_valid_gtid(global_tid);
1834 kmp_int32 rc = __kmp_enter_single(global_tid, loc, TRUE);
1839 KMP_PUSH_PARTITIONED_TIMER(OMP_single);
1842 #if OMPT_SUPPORT && OMPT_OPTIONAL
1843 kmp_info_t *this_thr = __kmp_threads[global_tid];
1844 kmp_team_t *team = this_thr->th.th_team;
1845 int tid = __kmp_tid_from_gtid(global_tid);
1847 if (ompt_enabled.enabled) {
1849 if (ompt_enabled.ompt_callback_work) {
1850 ompt_callbacks.ompt_callback(ompt_callback_work)(
1851 ompt_work_single_executor, ompt_scope_begin,
1852 &(team->t.ompt_team_info.parallel_data),
1853 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1854 1, OMPT_GET_RETURN_ADDRESS(0));
1857 if (ompt_enabled.ompt_callback_work) {
1858 ompt_callbacks.ompt_callback(ompt_callback_work)(
1859 ompt_work_single_other, ompt_scope_begin,
1860 &(team->t.ompt_team_info.parallel_data),
1861 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data),
1862 1, OMPT_GET_RETURN_ADDRESS(0));
1863 ompt_callbacks.ompt_callback(ompt_callback_work)(
1864 ompt_work_single_other, ompt_scope_end,
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));
1886 __kmp_assert_valid_gtid(global_tid);
1887 __kmp_exit_single(global_tid);
1888 KMP_POP_PARTITIONED_TIMER();
1890 #if OMPT_SUPPORT && OMPT_OPTIONAL
1891 kmp_info_t *this_thr = __kmp_threads[global_tid];
1892 kmp_team_t *team = this_thr->th.th_team;
1893 int tid = __kmp_tid_from_gtid(global_tid);
1895 if (ompt_enabled.ompt_callback_work) {
1896 ompt_callbacks.ompt_callback(ompt_callback_work)(
1897 ompt_work_single_executor, ompt_scope_end,
1898 &(team->t.ompt_team_info.parallel_data),
1899 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
1900 OMPT_GET_RETURN_ADDRESS(0));
1913 KMP_POP_PARTITIONED_TIMER();
1914 KE_TRACE(10, (
"__kmpc_for_static_fini called T#%d\n", global_tid));
1916 #if OMPT_SUPPORT && OMPT_OPTIONAL
1917 if (ompt_enabled.ompt_callback_work) {
1918 ompt_work_t ompt_work_type = ompt_work_loop;
1919 ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
1920 ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
1924 ompt_work_type = ompt_work_loop;
1926 ompt_work_type = ompt_work_sections;
1928 ompt_work_type = ompt_work_distribute;
1933 KMP_DEBUG_ASSERT(ompt_work_type);
1935 ompt_callbacks.ompt_callback(ompt_callback_work)(
1936 ompt_work_type, ompt_scope_end, &(team_info->parallel_data),
1937 &(task_info->task_data), 0, OMPT_GET_RETURN_ADDRESS(0));
1940 if (__kmp_env_consistency_check)
1941 __kmp_pop_workshare(global_tid, ct_pdo, loc);
1947 void ompc_set_num_threads(
int arg) {
1949 __kmp_set_num_threads(arg, __kmp_entry_gtid());
1952 void ompc_set_dynamic(
int flag) {
1956 thread = __kmp_entry_thread();
1958 __kmp_save_internal_controls(thread);
1960 set__dynamic(thread, flag ?
true :
false);
1963 void ompc_set_nested(
int flag) {
1967 thread = __kmp_entry_thread();
1969 __kmp_save_internal_controls(thread);
1971 set__max_active_levels(thread, flag ? __kmp_dflt_max_active_levels : 1);
1974 void ompc_set_max_active_levels(
int max_active_levels) {
1979 __kmp_set_max_active_levels(__kmp_entry_gtid(), max_active_levels);
1982 void ompc_set_schedule(omp_sched_t kind,
int modifier) {
1984 __kmp_set_schedule(__kmp_entry_gtid(), (kmp_sched_t)kind, modifier);
1987 int ompc_get_ancestor_thread_num(
int level) {
1988 return __kmp_get_ancestor_thread_num(__kmp_entry_gtid(), level);
1991 int ompc_get_team_size(
int level) {
1992 return __kmp_get_team_size(__kmp_entry_gtid(), level);
1996 void KMP_EXPAND_NAME(ompc_set_affinity_format)(
char const *format) {
1997 if (!__kmp_init_serial) {
1998 __kmp_serial_initialize();
2000 __kmp_strncpy_truncate(__kmp_affinity_format, KMP_AFFINITY_FORMAT_SIZE,
2001 format, KMP_STRLEN(format) + 1);
2004 size_t KMP_EXPAND_NAME(ompc_get_affinity_format)(
char *buffer,
size_t size) {
2006 if (!__kmp_init_serial) {
2007 __kmp_serial_initialize();
2009 format_size = KMP_STRLEN(__kmp_affinity_format);
2010 if (buffer && size) {
2011 __kmp_strncpy_truncate(buffer, size, __kmp_affinity_format,
2017 void KMP_EXPAND_NAME(ompc_display_affinity)(
char const *format) {
2019 if (!TCR_4(__kmp_init_middle)) {
2020 __kmp_middle_initialize();
2022 __kmp_assign_root_init_mask();
2023 gtid = __kmp_get_gtid();
2024 __kmp_aux_display_affinity(gtid, format);
2027 size_t KMP_EXPAND_NAME(ompc_capture_affinity)(
char *buffer,
size_t buf_size,
2028 char const *format) {
2030 size_t num_required;
2031 kmp_str_buf_t capture_buf;
2032 if (!TCR_4(__kmp_init_middle)) {
2033 __kmp_middle_initialize();
2035 __kmp_assign_root_init_mask();
2036 gtid = __kmp_get_gtid();
2037 __kmp_str_buf_init(&capture_buf);
2038 num_required = __kmp_aux_capture_affinity(gtid, format, &capture_buf);
2039 if (buffer && buf_size) {
2040 __kmp_strncpy_truncate(buffer, buf_size, capture_buf.str,
2041 capture_buf.used + 1);
2043 __kmp_str_buf_free(&capture_buf);
2044 return num_required;
2047 void kmpc_set_stacksize(
int arg) {
2049 __kmp_aux_set_stacksize(arg);
2052 void kmpc_set_stacksize_s(
size_t arg) {
2054 __kmp_aux_set_stacksize(arg);
2057 void kmpc_set_blocktime(
int arg) {
2061 gtid = __kmp_entry_gtid();
2062 tid = __kmp_tid_from_gtid(gtid);
2063 thread = __kmp_thread_from_gtid(gtid);
2065 __kmp_aux_set_blocktime(arg, thread, tid);
2068 void kmpc_set_library(
int arg) {
2070 __kmp_user_set_library((
enum library_type)arg);
2073 void kmpc_set_defaults(
char const *str) {
2075 __kmp_aux_set_defaults(str, KMP_STRLEN(str));
2078 void kmpc_set_disp_num_buffers(
int arg) {
2081 if (__kmp_init_serial == FALSE && arg >= KMP_MIN_DISP_NUM_BUFF &&
2082 arg <= KMP_MAX_DISP_NUM_BUFF) {
2083 __kmp_dispatch_num_buffers = arg;
2087 int kmpc_set_affinity_mask_proc(
int proc,
void **mask) {
2088 #if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2091 if (!TCR_4(__kmp_init_middle)) {
2092 __kmp_middle_initialize();
2094 __kmp_assign_root_init_mask();
2095 return __kmp_aux_set_affinity_mask_proc(proc, mask);
2099 int kmpc_unset_affinity_mask_proc(
int proc,
void **mask) {
2100 #if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2103 if (!TCR_4(__kmp_init_middle)) {
2104 __kmp_middle_initialize();
2106 __kmp_assign_root_init_mask();
2107 return __kmp_aux_unset_affinity_mask_proc(proc, mask);
2111 int kmpc_get_affinity_mask_proc(
int proc,
void **mask) {
2112 #if defined(KMP_STUB) || !KMP_AFFINITY_SUPPORTED
2115 if (!TCR_4(__kmp_init_middle)) {
2116 __kmp_middle_initialize();
2118 __kmp_assign_root_init_mask();
2119 return __kmp_aux_get_affinity_mask_proc(proc, mask);
2169 void *cpy_data,
void (*cpy_func)(
void *,
void *),
2172 KC_TRACE(10, (
"__kmpc_copyprivate: called T#%d\n", gtid));
2173 __kmp_assert_valid_gtid(gtid);
2177 data_ptr = &__kmp_team_from_gtid(gtid)->t.t_copypriv_data;
2179 if (__kmp_env_consistency_check) {
2181 KMP_WARNING(ConstructIdentInvalid);
2188 *data_ptr = cpy_data;
2191 ompt_frame_t *ompt_frame;
2192 if (ompt_enabled.enabled) {
2193 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
2194 if (ompt_frame->enter_frame.ptr == NULL)
2195 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
2197 OMPT_STORE_RETURN_ADDRESS(gtid);
2201 __kmp_threads[gtid]->th.th_ident = loc;
2203 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2206 (*cpy_func)(cpy_data, *data_ptr);
2212 OMPT_STORE_RETURN_ADDRESS(gtid);
2215 __kmp_threads[gtid]->th.th_ident = loc;
2218 __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
2219 #if OMPT_SUPPORT && OMPT_OPTIONAL
2220 if (ompt_enabled.enabled) {
2221 ompt_frame->enter_frame = ompt_data_none;
2229 #define INIT_LOCK __kmp_init_user_lock_with_checks
2230 #define INIT_NESTED_LOCK __kmp_init_nested_user_lock_with_checks
2231 #define ACQUIRE_LOCK __kmp_acquire_user_lock_with_checks
2232 #define ACQUIRE_LOCK_TIMED __kmp_acquire_user_lock_with_checks_timed
2233 #define ACQUIRE_NESTED_LOCK __kmp_acquire_nested_user_lock_with_checks
2234 #define ACQUIRE_NESTED_LOCK_TIMED \
2235 __kmp_acquire_nested_user_lock_with_checks_timed
2236 #define RELEASE_LOCK __kmp_release_user_lock_with_checks
2237 #define RELEASE_NESTED_LOCK __kmp_release_nested_user_lock_with_checks
2238 #define TEST_LOCK __kmp_test_user_lock_with_checks
2239 #define TEST_NESTED_LOCK __kmp_test_nested_user_lock_with_checks
2240 #define DESTROY_LOCK __kmp_destroy_user_lock_with_checks
2241 #define DESTROY_NESTED_LOCK __kmp_destroy_nested_user_lock_with_checks
2246 #if KMP_USE_DYNAMIC_LOCK
2249 static __forceinline
void __kmp_init_lock_with_hint(
ident_t *loc,
void **lock,
2250 kmp_dyna_lockseq_t seq) {
2251 if (KMP_IS_D_LOCK(seq)) {
2252 KMP_INIT_D_LOCK(lock, seq);
2254 __kmp_itt_lock_creating((kmp_user_lock_p)lock, NULL);
2257 KMP_INIT_I_LOCK(lock, seq);
2259 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2260 __kmp_itt_lock_creating(ilk->lock, loc);
2266 static __forceinline
void
2267 __kmp_init_nest_lock_with_hint(
ident_t *loc,
void **lock,
2268 kmp_dyna_lockseq_t seq) {
2271 if (seq == lockseq_hle || seq == lockseq_rtm_queuing ||
2272 seq == lockseq_rtm_spin || seq == lockseq_adaptive)
2273 seq = __kmp_user_lock_seq;
2277 seq = lockseq_nested_tas;
2281 seq = lockseq_nested_futex;
2284 case lockseq_ticket:
2285 seq = lockseq_nested_ticket;
2287 case lockseq_queuing:
2288 seq = lockseq_nested_queuing;
2291 seq = lockseq_nested_drdpa;
2294 seq = lockseq_nested_queuing;
2296 KMP_INIT_I_LOCK(lock, seq);
2298 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(lock);
2299 __kmp_itt_lock_creating(ilk->lock, loc);
2304 void __kmpc_init_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
void **user_lock,
2306 KMP_DEBUG_ASSERT(__kmp_init_serial);
2307 if (__kmp_env_consistency_check && user_lock == NULL) {
2308 KMP_FATAL(LockIsUninitialized,
"omp_init_lock_with_hint");
2311 __kmp_init_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2313 #if OMPT_SUPPORT && OMPT_OPTIONAL
2315 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2317 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2318 if (ompt_enabled.ompt_callback_lock_init) {
2319 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2320 ompt_mutex_lock, (omp_lock_hint_t)hint,
2321 __ompt_get_mutex_impl_type(user_lock),
2322 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2328 void __kmpc_init_nest_lock_with_hint(
ident_t *loc, kmp_int32 gtid,
2329 void **user_lock, uintptr_t hint) {
2330 KMP_DEBUG_ASSERT(__kmp_init_serial);
2331 if (__kmp_env_consistency_check && user_lock == NULL) {
2332 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock_with_hint");
2335 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_map_hint_to_lock(hint));
2337 #if OMPT_SUPPORT && OMPT_OPTIONAL
2339 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2341 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2342 if (ompt_enabled.ompt_callback_lock_init) {
2343 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2344 ompt_mutex_nest_lock, (omp_lock_hint_t)hint,
2345 __ompt_get_mutex_impl_type(user_lock),
2346 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2354 void __kmpc_init_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2355 #if KMP_USE_DYNAMIC_LOCK
2357 KMP_DEBUG_ASSERT(__kmp_init_serial);
2358 if (__kmp_env_consistency_check && user_lock == NULL) {
2359 KMP_FATAL(LockIsUninitialized,
"omp_init_lock");
2361 __kmp_init_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2363 #if OMPT_SUPPORT && OMPT_OPTIONAL
2365 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2367 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2368 if (ompt_enabled.ompt_callback_lock_init) {
2369 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2370 ompt_mutex_lock, omp_lock_hint_none,
2371 __ompt_get_mutex_impl_type(user_lock),
2372 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2378 static char const *
const func =
"omp_init_lock";
2379 kmp_user_lock_p lck;
2380 KMP_DEBUG_ASSERT(__kmp_init_serial);
2382 if (__kmp_env_consistency_check) {
2383 if (user_lock == NULL) {
2384 KMP_FATAL(LockIsUninitialized, func);
2388 KMP_CHECK_USER_LOCK_INIT();
2390 if ((__kmp_user_lock_kind == lk_tas) &&
2391 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2392 lck = (kmp_user_lock_p)user_lock;
2395 else if ((__kmp_user_lock_kind == lk_futex) &&
2396 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2397 lck = (kmp_user_lock_p)user_lock;
2401 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2404 __kmp_set_user_lock_location(lck, loc);
2406 #if OMPT_SUPPORT && OMPT_OPTIONAL
2408 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2410 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2411 if (ompt_enabled.ompt_callback_lock_init) {
2412 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2413 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2414 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2419 __kmp_itt_lock_creating(lck);
2426 void __kmpc_init_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2427 #if KMP_USE_DYNAMIC_LOCK
2429 KMP_DEBUG_ASSERT(__kmp_init_serial);
2430 if (__kmp_env_consistency_check && user_lock == NULL) {
2431 KMP_FATAL(LockIsUninitialized,
"omp_init_nest_lock");
2433 __kmp_init_nest_lock_with_hint(loc, user_lock, __kmp_user_lock_seq);
2435 #if OMPT_SUPPORT && OMPT_OPTIONAL
2437 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2439 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2440 if (ompt_enabled.ompt_callback_lock_init) {
2441 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2442 ompt_mutex_nest_lock, omp_lock_hint_none,
2443 __ompt_get_mutex_impl_type(user_lock),
2444 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2450 static char const *
const func =
"omp_init_nest_lock";
2451 kmp_user_lock_p lck;
2452 KMP_DEBUG_ASSERT(__kmp_init_serial);
2454 if (__kmp_env_consistency_check) {
2455 if (user_lock == NULL) {
2456 KMP_FATAL(LockIsUninitialized, func);
2460 KMP_CHECK_USER_LOCK_INIT();
2462 if ((__kmp_user_lock_kind == lk_tas) &&
2463 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2464 OMP_NEST_LOCK_T_SIZE)) {
2465 lck = (kmp_user_lock_p)user_lock;
2468 else if ((__kmp_user_lock_kind == lk_futex) &&
2469 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2470 OMP_NEST_LOCK_T_SIZE)) {
2471 lck = (kmp_user_lock_p)user_lock;
2475 lck = __kmp_user_lock_allocate(user_lock, gtid, 0);
2478 INIT_NESTED_LOCK(lck);
2479 __kmp_set_user_lock_location(lck, loc);
2481 #if OMPT_SUPPORT && OMPT_OPTIONAL
2483 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2485 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2486 if (ompt_enabled.ompt_callback_lock_init) {
2487 ompt_callbacks.ompt_callback(ompt_callback_lock_init)(
2488 ompt_mutex_nest_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2489 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2494 __kmp_itt_lock_creating(lck);
2500 void __kmpc_destroy_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2501 #if KMP_USE_DYNAMIC_LOCK
2504 kmp_user_lock_p lck;
2505 if (KMP_EXTRACT_D_TAG(user_lock) == 0) {
2506 lck = ((kmp_indirect_lock_t *)KMP_LOOKUP_I_LOCK(user_lock))->lock;
2508 lck = (kmp_user_lock_p)user_lock;
2510 __kmp_itt_lock_destroyed(lck);
2512 #if OMPT_SUPPORT && OMPT_OPTIONAL
2514 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2516 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2517 if (ompt_enabled.ompt_callback_lock_destroy) {
2518 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2519 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2522 KMP_D_LOCK_FUNC(user_lock, destroy)((kmp_dyna_lock_t *)user_lock);
2524 kmp_user_lock_p lck;
2526 if ((__kmp_user_lock_kind == lk_tas) &&
2527 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2528 lck = (kmp_user_lock_p)user_lock;
2531 else if ((__kmp_user_lock_kind == lk_futex) &&
2532 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2533 lck = (kmp_user_lock_p)user_lock;
2537 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_lock");
2540 #if OMPT_SUPPORT && OMPT_OPTIONAL
2542 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2544 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2545 if (ompt_enabled.ompt_callback_lock_destroy) {
2546 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2547 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2552 __kmp_itt_lock_destroyed(lck);
2556 if ((__kmp_user_lock_kind == lk_tas) &&
2557 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2561 else if ((__kmp_user_lock_kind == lk_futex) &&
2562 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2567 __kmp_user_lock_free(user_lock, gtid, lck);
2573 void __kmpc_destroy_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2574 #if KMP_USE_DYNAMIC_LOCK
2577 kmp_indirect_lock_t *ilk = KMP_LOOKUP_I_LOCK(user_lock);
2578 __kmp_itt_lock_destroyed(ilk->lock);
2580 #if OMPT_SUPPORT && OMPT_OPTIONAL
2582 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2584 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2585 if (ompt_enabled.ompt_callback_lock_destroy) {
2586 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2587 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2590 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) +
sizeof(lck->tas.lk.depth_locked) <=
2598 OMP_NEST_LOCK_T_SIZE)) {
2599 lck = (kmp_user_lock_p)user_lock;
2602 else if ((__kmp_user_lock_kind == lk_futex) &&
2603 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2604 OMP_NEST_LOCK_T_SIZE)) {
2605 lck = (kmp_user_lock_p)user_lock;
2609 lck = __kmp_lookup_user_lock(user_lock,
"omp_destroy_nest_lock");
2612 #if OMPT_SUPPORT && OMPT_OPTIONAL
2614 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2616 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2617 if (ompt_enabled.ompt_callback_lock_destroy) {
2618 ompt_callbacks.ompt_callback(ompt_callback_lock_destroy)(
2619 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2624 __kmp_itt_lock_destroyed(lck);
2627 DESTROY_NESTED_LOCK(lck);
2629 if ((__kmp_user_lock_kind == lk_tas) &&
2630 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2631 OMP_NEST_LOCK_T_SIZE)) {
2635 else if ((__kmp_user_lock_kind == lk_futex) &&
2636 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2637 OMP_NEST_LOCK_T_SIZE)) {
2642 __kmp_user_lock_free(user_lock, gtid, lck);
2647 void __kmpc_set_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2649 #if KMP_USE_DYNAMIC_LOCK
2650 int tag = KMP_EXTRACT_D_TAG(user_lock);
2652 __kmp_itt_lock_acquiring(
2656 #if OMPT_SUPPORT && OMPT_OPTIONAL
2658 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2660 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2661 if (ompt_enabled.ompt_callback_mutex_acquire) {
2662 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2663 ompt_mutex_lock, omp_lock_hint_none,
2664 __ompt_get_mutex_impl_type(user_lock),
2665 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2668 #if KMP_USE_INLINED_TAS
2669 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2670 KMP_ACQUIRE_TAS_LOCK(user_lock, gtid);
2672 #elif KMP_USE_INLINED_FUTEX
2673 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2674 KMP_ACQUIRE_FUTEX_LOCK(user_lock, gtid);
2678 __kmp_direct_set[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2681 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2683 #if OMPT_SUPPORT && OMPT_OPTIONAL
2684 if (ompt_enabled.ompt_callback_mutex_acquired) {
2685 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2686 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2692 kmp_user_lock_p lck;
2694 if ((__kmp_user_lock_kind == lk_tas) &&
2695 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2696 lck = (kmp_user_lock_p)user_lock;
2699 else if ((__kmp_user_lock_kind == lk_futex) &&
2700 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2701 lck = (kmp_user_lock_p)user_lock;
2705 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_lock");
2709 __kmp_itt_lock_acquiring(lck);
2711 #if OMPT_SUPPORT && OMPT_OPTIONAL
2713 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2715 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2716 if (ompt_enabled.ompt_callback_mutex_acquire) {
2717 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2718 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
2719 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2723 ACQUIRE_LOCK(lck, gtid);
2726 __kmp_itt_lock_acquired(lck);
2729 #if OMPT_SUPPORT && OMPT_OPTIONAL
2730 if (ompt_enabled.ompt_callback_mutex_acquired) {
2731 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2732 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2739 void __kmpc_set_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2740 #if KMP_USE_DYNAMIC_LOCK
2743 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
2745 #if OMPT_SUPPORT && OMPT_OPTIONAL
2747 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2749 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2750 if (ompt_enabled.enabled) {
2751 if (ompt_enabled.ompt_callback_mutex_acquire) {
2752 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2753 ompt_mutex_nest_lock, omp_lock_hint_none,
2754 __ompt_get_mutex_impl_type(user_lock),
2755 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2759 int acquire_status =
2760 KMP_D_LOCK_FUNC(user_lock, set)((kmp_dyna_lock_t *)user_lock, gtid);
2761 (void)acquire_status;
2763 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
2766 #if OMPT_SUPPORT && OMPT_OPTIONAL
2767 if (ompt_enabled.enabled) {
2768 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2769 if (ompt_enabled.ompt_callback_mutex_acquired) {
2771 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2772 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
2776 if (ompt_enabled.ompt_callback_nest_lock) {
2778 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2779 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2787 kmp_user_lock_p lck;
2789 if ((__kmp_user_lock_kind == lk_tas) &&
2790 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2791 OMP_NEST_LOCK_T_SIZE)) {
2792 lck = (kmp_user_lock_p)user_lock;
2795 else if ((__kmp_user_lock_kind == lk_futex) &&
2796 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
2797 OMP_NEST_LOCK_T_SIZE)) {
2798 lck = (kmp_user_lock_p)user_lock;
2802 lck = __kmp_lookup_user_lock(user_lock,
"omp_set_nest_lock");
2806 __kmp_itt_lock_acquiring(lck);
2808 #if OMPT_SUPPORT && OMPT_OPTIONAL
2810 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2812 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2813 if (ompt_enabled.enabled) {
2814 if (ompt_enabled.ompt_callback_mutex_acquire) {
2815 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
2816 ompt_mutex_nest_lock, omp_lock_hint_none,
2817 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
2823 ACQUIRE_NESTED_LOCK(lck, gtid, &acquire_status);
2826 __kmp_itt_lock_acquired(lck);
2829 #if OMPT_SUPPORT && OMPT_OPTIONAL
2830 if (ompt_enabled.enabled) {
2831 if (acquire_status == KMP_LOCK_ACQUIRED_FIRST) {
2832 if (ompt_enabled.ompt_callback_mutex_acquired) {
2834 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
2835 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2838 if (ompt_enabled.ompt_callback_nest_lock) {
2840 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2841 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2850 void __kmpc_unset_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2851 #if KMP_USE_DYNAMIC_LOCK
2853 int tag = KMP_EXTRACT_D_TAG(user_lock);
2855 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2857 #if KMP_USE_INLINED_TAS
2858 if (tag == locktag_tas && !__kmp_env_consistency_check) {
2859 KMP_RELEASE_TAS_LOCK(user_lock, gtid);
2861 #elif KMP_USE_INLINED_FUTEX
2862 if (tag == locktag_futex && !__kmp_env_consistency_check) {
2863 KMP_RELEASE_FUTEX_LOCK(user_lock, gtid);
2867 __kmp_direct_unset[tag]((kmp_dyna_lock_t *)user_lock, gtid);
2870 #if OMPT_SUPPORT && OMPT_OPTIONAL
2872 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2874 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2875 if (ompt_enabled.ompt_callback_mutex_released) {
2876 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2877 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2883 kmp_user_lock_p lck;
2888 if ((__kmp_user_lock_kind == lk_tas) &&
2889 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
2890 #if KMP_OS_LINUX && \
2891 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
2894 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2896 TCW_4(((kmp_user_lock_p)user_lock)->tas.lk.poll, 0);
2899 #if OMPT_SUPPORT && OMPT_OPTIONAL
2901 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2903 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2904 if (ompt_enabled.ompt_callback_mutex_released) {
2905 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2906 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2912 lck = (kmp_user_lock_p)user_lock;
2916 else if ((__kmp_user_lock_kind == lk_futex) &&
2917 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
2918 lck = (kmp_user_lock_p)user_lock;
2922 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_lock");
2926 __kmp_itt_lock_releasing(lck);
2929 RELEASE_LOCK(lck, gtid);
2931 #if OMPT_SUPPORT && OMPT_OPTIONAL
2933 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2935 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2936 if (ompt_enabled.ompt_callback_mutex_released) {
2937 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2938 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
2946 void __kmpc_unset_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
2947 #if KMP_USE_DYNAMIC_LOCK
2950 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2952 int release_status =
2953 KMP_D_LOCK_FUNC(user_lock, unset)((kmp_dyna_lock_t *)user_lock, gtid);
2954 (void)release_status;
2956 #if OMPT_SUPPORT && OMPT_OPTIONAL
2958 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
2960 codeptr = OMPT_GET_RETURN_ADDRESS(0);
2961 if (ompt_enabled.enabled) {
2962 if (release_status == KMP_LOCK_RELEASED) {
2963 if (ompt_enabled.ompt_callback_mutex_released) {
2965 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
2966 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
2969 }
else if (ompt_enabled.ompt_callback_nest_lock) {
2971 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
2972 ompt_scope_end, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
2979 kmp_user_lock_p lck;
2983 if ((__kmp_user_lock_kind == lk_tas) &&
2984 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
2985 OMP_NEST_LOCK_T_SIZE)) {
2986 #if KMP_OS_LINUX && \
2987 (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64)
2989 kmp_tas_lock_t *tl = (kmp_tas_lock_t *)user_lock;
2991 __kmp_itt_lock_releasing((kmp_user_lock_p)user_lock);
2994 #if OMPT_SUPPORT && OMPT_OPTIONAL
2995 int release_status = KMP_LOCK_STILL_HELD;
2998 if (--(tl->lk.depth_locked) == 0) {
2999 TCW_4(tl->lk.poll, 0);
3000 #if OMPT_SUPPORT && OMPT_OPTIONAL
3001 release_status = KMP_LOCK_RELEASED;
3006 #if OMPT_SUPPORT && OMPT_OPTIONAL
3008 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3010 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3011 if (ompt_enabled.enabled) {
3012 if (release_status == KMP_LOCK_RELEASED) {
3013 if (ompt_enabled.ompt_callback_mutex_released) {
3015 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3016 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3018 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3020 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3021 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3028 lck = (kmp_user_lock_p)user_lock;
3032 else if ((__kmp_user_lock_kind == lk_futex) &&
3033 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3034 OMP_NEST_LOCK_T_SIZE)) {
3035 lck = (kmp_user_lock_p)user_lock;
3039 lck = __kmp_lookup_user_lock(user_lock,
"omp_unset_nest_lock");
3043 __kmp_itt_lock_releasing(lck);
3047 release_status = RELEASE_NESTED_LOCK(lck, gtid);
3048 #if OMPT_SUPPORT && OMPT_OPTIONAL
3050 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3052 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3053 if (ompt_enabled.enabled) {
3054 if (release_status == KMP_LOCK_RELEASED) {
3055 if (ompt_enabled.ompt_callback_mutex_released) {
3057 ompt_callbacks.ompt_callback(ompt_callback_mutex_released)(
3058 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3060 }
else if (ompt_enabled.ompt_callback_nest_lock) {
3062 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3063 ompt_mutex_scope_end, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3072 int __kmpc_test_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3075 #if KMP_USE_DYNAMIC_LOCK
3077 int tag = KMP_EXTRACT_D_TAG(user_lock);
3079 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3081 #if OMPT_SUPPORT && OMPT_OPTIONAL
3083 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3085 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3086 if (ompt_enabled.ompt_callback_mutex_acquire) {
3087 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3088 ompt_mutex_lock, omp_lock_hint_none,
3089 __ompt_get_mutex_impl_type(user_lock),
3090 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3093 #if KMP_USE_INLINED_TAS
3094 if (tag == locktag_tas && !__kmp_env_consistency_check) {
3095 KMP_TEST_TAS_LOCK(user_lock, gtid, rc);
3097 #elif KMP_USE_INLINED_FUTEX
3098 if (tag == locktag_futex && !__kmp_env_consistency_check) {
3099 KMP_TEST_FUTEX_LOCK(user_lock, gtid, rc);
3103 rc = __kmp_direct_test[tag]((kmp_dyna_lock_t *)user_lock, gtid);
3107 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3109 #if OMPT_SUPPORT && OMPT_OPTIONAL
3110 if (ompt_enabled.ompt_callback_mutex_acquired) {
3111 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3112 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3118 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3125 kmp_user_lock_p lck;
3128 if ((__kmp_user_lock_kind == lk_tas) &&
3129 (
sizeof(lck->tas.lk.poll) <= OMP_LOCK_T_SIZE)) {
3130 lck = (kmp_user_lock_p)user_lock;
3133 else if ((__kmp_user_lock_kind == lk_futex) &&
3134 (
sizeof(lck->futex.lk.poll) <= OMP_LOCK_T_SIZE)) {
3135 lck = (kmp_user_lock_p)user_lock;
3139 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_lock");
3143 __kmp_itt_lock_acquiring(lck);
3145 #if OMPT_SUPPORT && OMPT_OPTIONAL
3147 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3149 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3150 if (ompt_enabled.ompt_callback_mutex_acquire) {
3151 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3152 ompt_mutex_lock, omp_lock_hint_none, __ompt_get_mutex_impl_type(),
3153 (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3157 rc = TEST_LOCK(lck, gtid);
3160 __kmp_itt_lock_acquired(lck);
3162 __kmp_itt_lock_cancelled(lck);
3165 #if OMPT_SUPPORT && OMPT_OPTIONAL
3166 if (rc && ompt_enabled.ompt_callback_mutex_acquired) {
3167 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3168 ompt_mutex_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3172 return (rc ? FTN_TRUE : FTN_FALSE);
3180 int __kmpc_test_nest_lock(
ident_t *loc, kmp_int32 gtid,
void **user_lock) {
3181 #if KMP_USE_DYNAMIC_LOCK
3184 __kmp_itt_lock_acquiring((kmp_user_lock_p)user_lock);
3186 #if OMPT_SUPPORT && OMPT_OPTIONAL
3188 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3190 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3191 if (ompt_enabled.ompt_callback_mutex_acquire) {
3192 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3193 ompt_mutex_nest_lock, omp_lock_hint_none,
3194 __ompt_get_mutex_impl_type(user_lock),
3195 (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3198 rc = KMP_D_LOCK_FUNC(user_lock, test)((kmp_dyna_lock_t *)user_lock, gtid);
3201 __kmp_itt_lock_acquired((kmp_user_lock_p)user_lock);
3203 __kmp_itt_lock_cancelled((kmp_user_lock_p)user_lock);
3206 #if OMPT_SUPPORT && OMPT_OPTIONAL
3207 if (ompt_enabled.enabled && rc) {
3209 if (ompt_enabled.ompt_callback_mutex_acquired) {
3211 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3212 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)user_lock,
3216 if (ompt_enabled.ompt_callback_nest_lock) {
3218 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3219 ompt_scope_begin, (ompt_wait_id_t)(uintptr_t)user_lock, codeptr);
3228 kmp_user_lock_p lck;
3231 if ((__kmp_user_lock_kind == lk_tas) &&
3232 (
sizeof(lck->tas.lk.poll) +
sizeof(lck->tas.lk.depth_locked) <=
3233 OMP_NEST_LOCK_T_SIZE)) {
3234 lck = (kmp_user_lock_p)user_lock;
3237 else if ((__kmp_user_lock_kind == lk_futex) &&
3238 (
sizeof(lck->futex.lk.poll) +
sizeof(lck->futex.lk.depth_locked) <=
3239 OMP_NEST_LOCK_T_SIZE)) {
3240 lck = (kmp_user_lock_p)user_lock;
3244 lck = __kmp_lookup_user_lock(user_lock,
"omp_test_nest_lock");
3248 __kmp_itt_lock_acquiring(lck);
3251 #if OMPT_SUPPORT && OMPT_OPTIONAL
3253 void *codeptr = OMPT_LOAD_RETURN_ADDRESS(gtid);
3255 codeptr = OMPT_GET_RETURN_ADDRESS(0);
3256 if (ompt_enabled.enabled) &&
3257 ompt_enabled.ompt_callback_mutex_acquire) {
3258 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquire)(
3259 ompt_mutex_nest_lock, omp_lock_hint_none,
3260 __ompt_get_mutex_impl_type(), (ompt_wait_id_t)(uintptr_t)lck,
3265 rc = TEST_NESTED_LOCK(lck, gtid);
3268 __kmp_itt_lock_acquired(lck);
3270 __kmp_itt_lock_cancelled(lck);
3273 #if OMPT_SUPPORT && OMPT_OPTIONAL
3274 if (ompt_enabled.enabled && rc) {
3276 if (ompt_enabled.ompt_callback_mutex_acquired) {
3278 ompt_callbacks.ompt_callback(ompt_callback_mutex_acquired)(
3279 ompt_mutex_nest_lock, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3282 if (ompt_enabled.ompt_callback_nest_lock) {
3284 ompt_callbacks.ompt_callback(ompt_callback_nest_lock)(
3285 ompt_mutex_scope_begin, (ompt_wait_id_t)(uintptr_t)lck, codeptr);
3304 #define __KMP_SET_REDUCTION_METHOD(gtid, rmethod) \
3305 ((__kmp_threads[(gtid)]->th.th_local.packed_reduction_method) = (rmethod))
3307 #define __KMP_GET_REDUCTION_METHOD(gtid) \
3308 (__kmp_threads[(gtid)]->th.th_local.packed_reduction_method)
3314 static __forceinline
void
3315 __kmp_enter_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3316 kmp_critical_name *crit) {
3322 kmp_user_lock_p lck;
3324 #if KMP_USE_DYNAMIC_LOCK
3326 kmp_dyna_lock_t *lk = (kmp_dyna_lock_t *)crit;
3329 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3330 KMP_COMPARE_AND_STORE_ACQ32((
volatile kmp_int32 *)crit, 0,
3331 KMP_GET_D_TAG(__kmp_user_lock_seq));
3333 __kmp_init_indirect_csptr(crit, loc, global_tid,
3334 KMP_GET_I_TAG(__kmp_user_lock_seq));
3340 if (KMP_EXTRACT_D_TAG(lk) != 0) {
3341 lck = (kmp_user_lock_p)lk;
3342 KMP_DEBUG_ASSERT(lck != NULL);
3343 if (__kmp_env_consistency_check) {
3344 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3346 KMP_D_LOCK_FUNC(lk, set)(lk, global_tid);
3348 kmp_indirect_lock_t *ilk = *((kmp_indirect_lock_t **)lk);
3350 KMP_DEBUG_ASSERT(lck != NULL);
3351 if (__kmp_env_consistency_check) {
3352 __kmp_push_sync(global_tid, ct_critical, loc, lck, __kmp_user_lock_seq);
3354 KMP_I_LOCK_FUNC(ilk, set)(lck, global_tid);
3362 if (__kmp_base_user_lock_size <= INTEL_CRITICAL_SIZE) {
3363 lck = (kmp_user_lock_p)crit;
3365 lck = __kmp_get_critical_section_ptr(crit, loc, global_tid);
3367 KMP_DEBUG_ASSERT(lck != NULL);
3369 if (__kmp_env_consistency_check)
3370 __kmp_push_sync(global_tid, ct_critical, loc, lck);
3372 __kmp_acquire_user_lock_with_checks(lck, global_tid);
3378 static __forceinline
void
3379 __kmp_end_critical_section_reduce_block(
ident_t *loc, kmp_int32 global_tid,
3380 kmp_critical_name *crit) {
3382 kmp_user_lock_p lck;
3384 #if KMP_USE_DYNAMIC_LOCK
3386 if (KMP_IS_D_LOCK(__kmp_user_lock_seq)) {
3387 lck = (kmp_user_lock_p)crit;
3388 if (__kmp_env_consistency_check)
3389 __kmp_pop_sync(global_tid, ct_critical, loc);
3390 KMP_D_LOCK_FUNC(lck, unset)((kmp_dyna_lock_t *)lck, global_tid);
3392 kmp_indirect_lock_t *ilk =
3393 (kmp_indirect_lock_t *)TCR_PTR(*((kmp_indirect_lock_t **)crit));
3394 if (__kmp_env_consistency_check)
3395 __kmp_pop_sync(global_tid, ct_critical, loc);
3396 KMP_I_LOCK_FUNC(ilk, unset)(ilk->lock, global_tid);
3404 if (__kmp_base_user_lock_size > 32) {
3405 lck = *((kmp_user_lock_p *)crit);
3406 KMP_ASSERT(lck != NULL);
3408 lck = (kmp_user_lock_p)crit;
3411 if (__kmp_env_consistency_check)
3412 __kmp_pop_sync(global_tid, ct_critical, loc);
3414 __kmp_release_user_lock_with_checks(lck, global_tid);
3419 static __forceinline
int
3420 __kmp_swap_teams_for_teams_reduction(kmp_info_t *th, kmp_team_t **team_p,
3425 if (th->th.th_teams_microtask) {
3426 *team_p = team = th->th.th_team;
3427 if (team->t.t_level == th->th.th_teams_level) {
3429 KMP_DEBUG_ASSERT(!th->th.th_info.ds.ds_tid);
3431 th->th.th_info.ds.ds_tid = team->t.t_master_tid;
3432 th->th.th_team = team->t.t_parent;
3433 th->th.th_team_nproc = th->th.th_team->t.t_nproc;
3434 th->th.th_task_team = th->th.th_team->t.t_task_team[0];
3435 *task_state = th->th.th_task_state;
3436 th->th.th_task_state = 0;
3444 static __forceinline
void
3445 __kmp_restore_swapped_teams(kmp_info_t *th, kmp_team_t *team,
int task_state) {
3447 th->th.th_info.ds.ds_tid = 0;
3448 th->th.th_team = team;
3449 th->th.th_team_nproc = team->t.t_nproc;
3450 th->th.th_task_team = team->t.t_task_team[task_state];
3451 __kmp_type_convert(task_state, &(th->th.th_task_state));
3472 size_t reduce_size,
void *reduce_data,
3473 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3474 kmp_critical_name *lck) {
3478 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3481 int teams_swapped = 0, task_state;
3482 KA_TRACE(10, (
"__kmpc_reduce_nowait() enter: called T#%d\n", global_tid));
3483 __kmp_assert_valid_gtid(global_tid);
3491 if (!TCR_4(__kmp_init_parallel))
3492 __kmp_parallel_initialize();
3494 __kmp_resume_if_soft_paused();
3497 #if KMP_USE_DYNAMIC_LOCK
3498 if (__kmp_env_consistency_check)
3499 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3501 if (__kmp_env_consistency_check)
3502 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3505 th = __kmp_thread_from_gtid(global_tid);
3506 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3524 packed_reduction_method = __kmp_determine_reduction_method(
3525 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3526 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3528 OMPT_REDUCTION_DECL(th, global_tid);
3529 if (packed_reduction_method == critical_reduce_block) {
3531 OMPT_REDUCTION_BEGIN;
3533 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3536 }
else if (packed_reduction_method == empty_reduce_block) {
3538 OMPT_REDUCTION_BEGIN;
3544 }
else if (packed_reduction_method == atomic_reduce_block) {
3554 if (__kmp_env_consistency_check)
3555 __kmp_pop_sync(global_tid, ct_reduce, loc);
3557 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3558 tree_reduce_block)) {
3578 ompt_frame_t *ompt_frame;
3579 if (ompt_enabled.enabled) {
3580 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3581 if (ompt_frame->enter_frame.ptr == NULL)
3582 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3584 OMPT_STORE_RETURN_ADDRESS(global_tid);
3587 __kmp_threads[global_tid]->th.th_ident = loc;
3590 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3591 global_tid, FALSE, reduce_size, reduce_data, reduce_func);
3592 retval = (retval != 0) ? (0) : (1);
3593 #if OMPT_SUPPORT && OMPT_OPTIONAL
3594 if (ompt_enabled.enabled) {
3595 ompt_frame->enter_frame = ompt_data_none;
3601 if (__kmp_env_consistency_check) {
3603 __kmp_pop_sync(global_tid, ct_reduce, loc);
3612 if (teams_swapped) {
3613 __kmp_restore_swapped_teams(th, team, task_state);
3617 (
"__kmpc_reduce_nowait() exit: called T#%d: method %08x, returns %08x\n",
3618 global_tid, packed_reduction_method, retval));
3632 kmp_critical_name *lck) {
3634 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3636 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() enter: called T#%d\n", global_tid));
3637 __kmp_assert_valid_gtid(global_tid);
3639 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3641 OMPT_REDUCTION_DECL(__kmp_thread_from_gtid(global_tid), global_tid);
3643 if (packed_reduction_method == critical_reduce_block) {
3645 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3648 }
else if (packed_reduction_method == empty_reduce_block) {
3655 }
else if (packed_reduction_method == atomic_reduce_block) {
3662 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3663 tree_reduce_block)) {
3674 if (__kmp_env_consistency_check)
3675 __kmp_pop_sync(global_tid, ct_reduce, loc);
3677 KA_TRACE(10, (
"__kmpc_end_reduce_nowait() exit: called T#%d: method %08x\n",
3678 global_tid, packed_reduction_method));
3701 size_t reduce_size,
void *reduce_data,
3702 void (*reduce_func)(
void *lhs_data,
void *rhs_data),
3703 kmp_critical_name *lck) {
3706 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3709 int teams_swapped = 0, task_state;
3711 KA_TRACE(10, (
"__kmpc_reduce() enter: called T#%d\n", global_tid));
3712 __kmp_assert_valid_gtid(global_tid);
3720 if (!TCR_4(__kmp_init_parallel))
3721 __kmp_parallel_initialize();
3723 __kmp_resume_if_soft_paused();
3726 #if KMP_USE_DYNAMIC_LOCK
3727 if (__kmp_env_consistency_check)
3728 __kmp_push_sync(global_tid, ct_reduce, loc, NULL, 0);
3730 if (__kmp_env_consistency_check)
3731 __kmp_push_sync(global_tid, ct_reduce, loc, NULL);
3734 th = __kmp_thread_from_gtid(global_tid);
3735 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3737 packed_reduction_method = __kmp_determine_reduction_method(
3738 loc, global_tid, num_vars, reduce_size, reduce_data, reduce_func, lck);
3739 __KMP_SET_REDUCTION_METHOD(global_tid, packed_reduction_method);
3741 OMPT_REDUCTION_DECL(th, global_tid);
3743 if (packed_reduction_method == critical_reduce_block) {
3745 OMPT_REDUCTION_BEGIN;
3746 __kmp_enter_critical_section_reduce_block(loc, global_tid, lck);
3749 }
else if (packed_reduction_method == empty_reduce_block) {
3751 OMPT_REDUCTION_BEGIN;
3756 }
else if (packed_reduction_method == atomic_reduce_block) {
3760 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3761 tree_reduce_block)) {
3767 ompt_frame_t *ompt_frame;
3768 if (ompt_enabled.enabled) {
3769 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3770 if (ompt_frame->enter_frame.ptr == NULL)
3771 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3773 OMPT_STORE_RETURN_ADDRESS(global_tid);
3776 __kmp_threads[global_tid]->th.th_ident =
3780 __kmp_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3781 global_tid, TRUE, reduce_size, reduce_data, reduce_func);
3782 retval = (retval != 0) ? (0) : (1);
3783 #if OMPT_SUPPORT && OMPT_OPTIONAL
3784 if (ompt_enabled.enabled) {
3785 ompt_frame->enter_frame = ompt_data_none;
3791 if (__kmp_env_consistency_check) {
3793 __kmp_pop_sync(global_tid, ct_reduce, loc);
3802 if (teams_swapped) {
3803 __kmp_restore_swapped_teams(th, team, task_state);
3807 (
"__kmpc_reduce() exit: called T#%d: method %08x, returns %08x\n",
3808 global_tid, packed_reduction_method, retval));
3823 kmp_critical_name *lck) {
3825 PACKED_REDUCTION_METHOD_T packed_reduction_method;
3828 int teams_swapped = 0, task_state;
3830 KA_TRACE(10, (
"__kmpc_end_reduce() enter: called T#%d\n", global_tid));
3831 __kmp_assert_valid_gtid(global_tid);
3833 th = __kmp_thread_from_gtid(global_tid);
3834 teams_swapped = __kmp_swap_teams_for_teams_reduction(th, &team, &task_state);
3836 packed_reduction_method = __KMP_GET_REDUCTION_METHOD(global_tid);
3840 OMPT_REDUCTION_DECL(th, global_tid);
3842 if (packed_reduction_method == critical_reduce_block) {
3843 __kmp_end_critical_section_reduce_block(loc, global_tid, lck);
3849 ompt_frame_t *ompt_frame;
3850 if (ompt_enabled.enabled) {
3851 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3852 if (ompt_frame->enter_frame.ptr == NULL)
3853 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3855 OMPT_STORE_RETURN_ADDRESS(global_tid);
3858 __kmp_threads[global_tid]->th.th_ident = loc;
3860 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3861 #if OMPT_SUPPORT && OMPT_OPTIONAL
3862 if (ompt_enabled.enabled) {
3863 ompt_frame->enter_frame = ompt_data_none;
3867 }
else if (packed_reduction_method == empty_reduce_block) {
3875 ompt_frame_t *ompt_frame;
3876 if (ompt_enabled.enabled) {
3877 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3878 if (ompt_frame->enter_frame.ptr == NULL)
3879 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3881 OMPT_STORE_RETURN_ADDRESS(global_tid);
3884 __kmp_threads[global_tid]->th.th_ident = loc;
3886 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3887 #if OMPT_SUPPORT && OMPT_OPTIONAL
3888 if (ompt_enabled.enabled) {
3889 ompt_frame->enter_frame = ompt_data_none;
3893 }
else if (packed_reduction_method == atomic_reduce_block) {
3896 ompt_frame_t *ompt_frame;
3897 if (ompt_enabled.enabled) {
3898 __ompt_get_task_info_internal(0, NULL, NULL, &ompt_frame, NULL, NULL);
3899 if (ompt_frame->enter_frame.ptr == NULL)
3900 ompt_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
3902 OMPT_STORE_RETURN_ADDRESS(global_tid);
3906 __kmp_threads[global_tid]->th.th_ident = loc;
3908 __kmp_barrier(bs_plain_barrier, global_tid, FALSE, 0, NULL, NULL);
3909 #if OMPT_SUPPORT && OMPT_OPTIONAL
3910 if (ompt_enabled.enabled) {
3911 ompt_frame->enter_frame = ompt_data_none;
3915 }
else if (TEST_REDUCTION_METHOD(packed_reduction_method,
3916 tree_reduce_block)) {
3919 __kmp_end_split_barrier(UNPACK_REDUCTION_BARRIER(packed_reduction_method),
3927 if (teams_swapped) {
3928 __kmp_restore_swapped_teams(th, team, task_state);
3931 if (__kmp_env_consistency_check)
3932 __kmp_pop_sync(global_tid, ct_reduce, loc);
3934 KA_TRACE(10, (
"__kmpc_end_reduce() exit: called T#%d: method %08x\n",
3935 global_tid, packed_reduction_method));
3940 #undef __KMP_GET_REDUCTION_METHOD
3941 #undef __KMP_SET_REDUCTION_METHOD
3945 kmp_uint64 __kmpc_get_taskid() {
3950 gtid = __kmp_get_gtid();
3954 thread = __kmp_thread_from_gtid(gtid);
3955 return thread->th.th_current_task->td_task_id;
3959 kmp_uint64 __kmpc_get_parent_taskid() {
3963 kmp_taskdata_t *parent_task;
3965 gtid = __kmp_get_gtid();
3969 thread = __kmp_thread_from_gtid(gtid);
3970 parent_task = thread->th.th_current_task->td_parent;
3971 return (parent_task == NULL ? 0 : parent_task->td_task_id);
3987 const struct kmp_dim *dims) {
3988 __kmp_assert_valid_gtid(gtid);
3990 kmp_int64 last, trace_count;
3991 kmp_info_t *th = __kmp_threads[gtid];
3992 kmp_team_t *team = th->th.th_team;
3994 kmp_disp_t *pr_buf = th->th.th_dispatch;
3995 dispatch_shared_info_t *sh_buf;
3999 (
"__kmpc_doacross_init() enter: called T#%d, num dims %d, active %d\n",
4000 gtid, num_dims, !team->t.t_serialized));
4001 KMP_DEBUG_ASSERT(dims != NULL);
4002 KMP_DEBUG_ASSERT(num_dims > 0);
4004 if (team->t.t_serialized) {
4005 KA_TRACE(20, (
"__kmpc_doacross_init() exit: serialized team\n"));
4008 KMP_DEBUG_ASSERT(team->t.t_nproc > 1);
4009 idx = pr_buf->th_doacross_buf_idx++;
4011 sh_buf = &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4014 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info == NULL);
4015 pr_buf->th_doacross_info = (kmp_int64 *)__kmp_thread_malloc(
4016 th,
sizeof(kmp_int64) * (4 * num_dims + 1));
4017 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4018 pr_buf->th_doacross_info[0] =
4019 (kmp_int64)num_dims;
4022 pr_buf->th_doacross_info[1] = (kmp_int64)&sh_buf->doacross_num_done;
4023 pr_buf->th_doacross_info[2] = dims[0].lo;
4024 pr_buf->th_doacross_info[3] = dims[0].up;
4025 pr_buf->th_doacross_info[4] = dims[0].st;
4027 for (j = 1; j < num_dims; ++j) {
4030 if (dims[j].st == 1) {
4032 range_length = dims[j].up - dims[j].lo + 1;
4034 if (dims[j].st > 0) {
4035 KMP_DEBUG_ASSERT(dims[j].up > dims[j].lo);
4036 range_length = (kmp_uint64)(dims[j].up - dims[j].lo) / dims[j].st + 1;
4038 KMP_DEBUG_ASSERT(dims[j].lo > dims[j].up);
4040 (kmp_uint64)(dims[j].lo - dims[j].up) / (-dims[j].st) + 1;
4043 pr_buf->th_doacross_info[last++] = range_length;
4044 pr_buf->th_doacross_info[last++] = dims[j].lo;
4045 pr_buf->th_doacross_info[last++] = dims[j].up;
4046 pr_buf->th_doacross_info[last++] = dims[j].st;
4051 if (dims[0].st == 1) {
4052 trace_count = dims[0].up - dims[0].lo + 1;
4053 }
else if (dims[0].st > 0) {
4054 KMP_DEBUG_ASSERT(dims[0].up > dims[0].lo);
4055 trace_count = (kmp_uint64)(dims[0].up - dims[0].lo) / dims[0].st + 1;
4057 KMP_DEBUG_ASSERT(dims[0].lo > dims[0].up);
4058 trace_count = (kmp_uint64)(dims[0].lo - dims[0].up) / (-dims[0].st) + 1;
4060 for (j = 1; j < num_dims; ++j) {
4061 trace_count *= pr_buf->th_doacross_info[4 * j + 1];
4063 KMP_DEBUG_ASSERT(trace_count > 0);
4067 if (idx != sh_buf->doacross_buf_idx) {
4069 __kmp_wait_4((
volatile kmp_uint32 *)&sh_buf->doacross_buf_idx, idx,
4076 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET32(
4077 (
volatile kmp_int32 *)&sh_buf->doacross_flags, NULL, 1);
4079 flags = (kmp_uint32 *)KMP_COMPARE_AND_STORE_RET64(
4080 (
volatile kmp_int64 *)&sh_buf->doacross_flags, NULL, 1LL);
4082 if (flags == NULL) {
4085 (size_t)trace_count / 8 + 8;
4086 flags = (kmp_uint32 *)__kmp_thread_calloc(th, size, 1);
4088 sh_buf->doacross_flags = flags;
4089 }
else if (flags == (kmp_uint32 *)1) {
4092 while (*(
volatile kmp_int32 *)&sh_buf->doacross_flags == 1)
4094 while (*(
volatile kmp_int64 *)&sh_buf->doacross_flags == 1LL)
4101 KMP_DEBUG_ASSERT(sh_buf->doacross_flags > (kmp_uint32 *)1);
4102 pr_buf->th_doacross_flags =
4103 sh_buf->doacross_flags;
4105 KA_TRACE(20, (
"__kmpc_doacross_init() exit: T#%d\n", gtid));
4108 void __kmpc_doacross_wait(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4109 __kmp_assert_valid_gtid(gtid);
4113 kmp_int64 iter_number;
4114 kmp_info_t *th = __kmp_threads[gtid];
4115 kmp_team_t *team = th->th.th_team;
4117 kmp_int64 lo, up, st;
4119 KA_TRACE(20, (
"__kmpc_doacross_wait() enter: called T#%d\n", gtid));
4120 if (team->t.t_serialized) {
4121 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: serialized team\n"));
4126 pr_buf = th->th.th_dispatch;
4127 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4128 num_dims = (size_t)pr_buf->th_doacross_info[0];
4129 lo = pr_buf->th_doacross_info[2];
4130 up = pr_buf->th_doacross_info[3];
4131 st = pr_buf->th_doacross_info[4];
4132 #
if OMPT_SUPPORT && OMPT_OPTIONAL
4133 ompt_dependence_t deps[num_dims];
4136 if (vec[0] < lo || vec[0] > up) {
4137 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4138 "bounds [%lld,%lld]\n",
4139 gtid, vec[0], lo, up));
4142 iter_number = vec[0] - lo;
4143 }
else if (st > 0) {
4144 if (vec[0] < lo || vec[0] > up) {
4145 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4146 "bounds [%lld,%lld]\n",
4147 gtid, vec[0], lo, up));
4150 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4152 if (vec[0] > lo || vec[0] < up) {
4153 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4154 "bounds [%lld,%lld]\n",
4155 gtid, vec[0], lo, up));
4158 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4160 #if OMPT_SUPPORT && OMPT_OPTIONAL
4161 deps[0].variable.value = iter_number;
4162 deps[0].dependence_type = ompt_dependence_type_sink;
4164 for (i = 1; i < num_dims; ++i) {
4167 ln = pr_buf->th_doacross_info[j + 1];
4168 lo = pr_buf->th_doacross_info[j + 2];
4169 up = pr_buf->th_doacross_info[j + 3];
4170 st = pr_buf->th_doacross_info[j + 4];
4172 if (vec[i] < lo || vec[i] > up) {
4173 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4174 "bounds [%lld,%lld]\n",
4175 gtid, vec[i], lo, up));
4179 }
else if (st > 0) {
4180 if (vec[i] < lo || vec[i] > up) {
4181 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4182 "bounds [%lld,%lld]\n",
4183 gtid, vec[i], lo, up));
4186 iter = (kmp_uint64)(vec[i] - lo) / st;
4188 if (vec[i] > lo || vec[i] < up) {
4189 KA_TRACE(20, (
"__kmpc_doacross_wait() exit: T#%d iter %lld is out of "
4190 "bounds [%lld,%lld]\n",
4191 gtid, vec[i], lo, up));
4194 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4196 iter_number = iter + ln * iter_number;
4197 #if OMPT_SUPPORT && OMPT_OPTIONAL
4198 deps[i].variable.value = iter;
4199 deps[i].dependence_type = ompt_dependence_type_sink;
4202 shft = iter_number % 32;
4205 while ((flag & pr_buf->th_doacross_flags[iter_number]) == 0) {
4209 #if OMPT_SUPPORT && OMPT_OPTIONAL
4210 if (ompt_enabled.ompt_callback_dependences) {
4211 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4212 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4216 (
"__kmpc_doacross_wait() exit: T#%d wait for iter %lld completed\n",
4217 gtid, (iter_number << 5) + shft));
4220 void __kmpc_doacross_post(
ident_t *loc,
int gtid,
const kmp_int64 *vec) {
4221 __kmp_assert_valid_gtid(gtid);
4225 kmp_int64 iter_number;
4226 kmp_info_t *th = __kmp_threads[gtid];
4227 kmp_team_t *team = th->th.th_team;
4231 KA_TRACE(20, (
"__kmpc_doacross_post() enter: called T#%d\n", gtid));
4232 if (team->t.t_serialized) {
4233 KA_TRACE(20, (
"__kmpc_doacross_post() exit: serialized team\n"));
4239 pr_buf = th->th.th_dispatch;
4240 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info != NULL);
4241 num_dims = (size_t)pr_buf->th_doacross_info[0];
4242 lo = pr_buf->th_doacross_info[2];
4243 st = pr_buf->th_doacross_info[4];
4244 #
if OMPT_SUPPORT && OMPT_OPTIONAL
4245 ompt_dependence_t deps[num_dims];
4248 iter_number = vec[0] - lo;
4249 }
else if (st > 0) {
4250 iter_number = (kmp_uint64)(vec[0] - lo) / st;
4252 iter_number = (kmp_uint64)(lo - vec[0]) / (-st);
4254 #if OMPT_SUPPORT && OMPT_OPTIONAL
4255 deps[0].variable.value = iter_number;
4256 deps[0].dependence_type = ompt_dependence_type_source;
4258 for (i = 1; i < num_dims; ++i) {
4261 ln = pr_buf->th_doacross_info[j + 1];
4262 lo = pr_buf->th_doacross_info[j + 2];
4263 st = pr_buf->th_doacross_info[j + 4];
4266 }
else if (st > 0) {
4267 iter = (kmp_uint64)(vec[i] - lo) / st;
4269 iter = (kmp_uint64)(lo - vec[i]) / (-st);
4271 iter_number = iter + ln * iter_number;
4272 #if OMPT_SUPPORT && OMPT_OPTIONAL
4273 deps[i].variable.value = iter;
4274 deps[i].dependence_type = ompt_dependence_type_source;
4277 #if OMPT_SUPPORT && OMPT_OPTIONAL
4278 if (ompt_enabled.ompt_callback_dependences) {
4279 ompt_callbacks.ompt_callback(ompt_callback_dependences)(
4280 &(OMPT_CUR_TASK_INFO(th)->task_data), deps, (kmp_uint32)num_dims);
4283 shft = iter_number % 32;
4287 if ((flag & pr_buf->th_doacross_flags[iter_number]) == 0)
4288 KMP_TEST_THEN_OR32(&pr_buf->th_doacross_flags[iter_number], flag);
4289 KA_TRACE(20, (
"__kmpc_doacross_post() exit: T#%d iter %lld posted\n", gtid,
4290 (iter_number << 5) + shft));
4293 void __kmpc_doacross_fini(
ident_t *loc,
int gtid) {
4294 __kmp_assert_valid_gtid(gtid);
4296 kmp_info_t *th = __kmp_threads[gtid];
4297 kmp_team_t *team = th->th.th_team;
4298 kmp_disp_t *pr_buf = th->th.th_dispatch;
4300 KA_TRACE(20, (
"__kmpc_doacross_fini() enter: called T#%d\n", gtid));
4301 if (team->t.t_serialized) {
4302 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: serialized team %p\n", team));
4306 KMP_TEST_THEN_INC32((kmp_uintptr_t)(pr_buf->th_doacross_info[1])) + 1;
4307 if (num_done == th->th.th_team_nproc) {
4309 int idx = pr_buf->th_doacross_buf_idx - 1;
4310 dispatch_shared_info_t *sh_buf =
4311 &team->t.t_disp_buffer[idx % __kmp_dispatch_num_buffers];
4312 KMP_DEBUG_ASSERT(pr_buf->th_doacross_info[1] ==
4313 (kmp_int64)&sh_buf->doacross_num_done);
4314 KMP_DEBUG_ASSERT(num_done == sh_buf->doacross_num_done);
4315 KMP_DEBUG_ASSERT(idx == sh_buf->doacross_buf_idx);
4316 __kmp_thread_free(th, CCAST(kmp_uint32 *, sh_buf->doacross_flags));
4317 sh_buf->doacross_flags = NULL;
4318 sh_buf->doacross_num_done = 0;
4319 sh_buf->doacross_buf_idx +=
4320 __kmp_dispatch_num_buffers;
4323 pr_buf->th_doacross_flags = NULL;
4324 __kmp_thread_free(th, (
void *)pr_buf->th_doacross_info);
4325 pr_buf->th_doacross_info = NULL;
4326 KA_TRACE(20, (
"__kmpc_doacross_fini() exit: T#%d\n", gtid));
4330 void *omp_alloc(
size_t size, omp_allocator_handle_t allocator) {
4331 return __kmp_alloc(__kmp_entry_gtid(), 0, size, allocator);
4334 void *omp_aligned_alloc(
size_t align,
size_t size,
4335 omp_allocator_handle_t allocator) {
4336 return __kmp_alloc(__kmp_entry_gtid(), align, size, allocator);
4339 void *omp_calloc(
size_t nmemb,
size_t size, omp_allocator_handle_t allocator) {
4340 return __kmp_calloc(__kmp_entry_gtid(), 0, nmemb, size, allocator);
4343 void *omp_aligned_calloc(
size_t align,
size_t nmemb,
size_t size,
4344 omp_allocator_handle_t allocator) {
4345 return __kmp_calloc(__kmp_entry_gtid(), align, nmemb, size, allocator);
4348 void *omp_realloc(
void *ptr,
size_t size, omp_allocator_handle_t allocator,
4349 omp_allocator_handle_t free_allocator) {
4350 return __kmp_realloc(__kmp_entry_gtid(), ptr, size, allocator,
4354 void omp_free(
void *ptr, omp_allocator_handle_t allocator) {
4355 ___kmpc_free(__kmp_entry_gtid(), ptr, allocator);
4359 int __kmpc_get_target_offload(
void) {
4360 if (!__kmp_init_serial) {
4361 __kmp_serial_initialize();
4363 return __kmp_target_offload;
4366 int __kmpc_pause_resource(kmp_pause_status_t level) {
4367 if (!__kmp_init_serial) {
4370 return __kmp_pause_resource(level);
4373 void __kmpc_error(
ident_t *loc,
int severity,
const char *message) {
4374 if (!__kmp_init_serial)
4375 __kmp_serial_initialize();
4377 KMP_ASSERT(severity == severity_warning || severity == severity_fatal);
4380 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_error) {
4381 ompt_callbacks.ompt_callback(ompt_callback_error)(
4382 (ompt_severity_t)severity, message, KMP_STRLEN(message),
4383 OMPT_GET_RETURN_ADDRESS(0));
4389 kmp_str_loc_t str_loc = __kmp_str_loc_init(loc->
psource,
false);
4391 __kmp_str_format(
"%s:%s:%s", str_loc.file, str_loc.line, str_loc.col);
4392 __kmp_str_loc_free(&str_loc);
4394 src_loc = __kmp_str_format(
"unknown");
4397 if (severity == severity_warning)
4398 KMP_WARNING(UserDirectedWarning, src_loc, message);
4400 KMP_FATAL(UserDirectedError, src_loc, message);
4402 __kmp_str_free(&src_loc);
4406 void __kmpc_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4408 #if OMPT_SUPPORT && OMPT_OPTIONAL
4409 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4410 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4411 int tid = __kmp_tid_from_gtid(gtid);
4412 ompt_callbacks.ompt_callback(ompt_callback_work)(
4413 ompt_work_scope, ompt_scope_begin,
4414 &(team->t.ompt_team_info.parallel_data),
4415 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4416 OMPT_GET_RETURN_ADDRESS(0));
4422 void __kmpc_end_scope(
ident_t *loc, kmp_int32 gtid,
void *reserved) {
4424 #if OMPT_SUPPORT && OMPT_OPTIONAL
4425 if (ompt_enabled.enabled && ompt_enabled.ompt_callback_work) {
4426 kmp_team_t *team = __kmp_threads[gtid]->th.th_team;
4427 int tid = __kmp_tid_from_gtid(gtid);
4428 ompt_callbacks.ompt_callback(ompt_callback_work)(
4429 ompt_work_scope, ompt_scope_end,
4430 &(team->t.ompt_team_info.parallel_data),
4431 &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data), 1,
4432 OMPT_GET_RETURN_ADDRESS(0));
4437 #ifdef KMP_USE_VERSION_SYMBOLS
4446 #ifdef omp_set_affinity_format
4447 #undef omp_set_affinity_format
4449 #ifdef omp_get_affinity_format
4450 #undef omp_get_affinity_format
4452 #ifdef omp_display_affinity
4453 #undef omp_display_affinity
4455 #ifdef omp_capture_affinity
4456 #undef omp_capture_affinity
4458 KMP_VERSION_OMPC_SYMBOL(ompc_set_affinity_format, omp_set_affinity_format, 50,
4460 KMP_VERSION_OMPC_SYMBOL(ompc_get_affinity_format, omp_get_affinity_format, 50,
4462 KMP_VERSION_OMPC_SYMBOL(ompc_display_affinity, omp_display_affinity, 50,
4464 KMP_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)
kmp_int32 __kmpc_global_num_threads(ident_t *loc)
kmp_int32 __kmpc_global_thread_num(ident_t *loc)
kmp_int32 __kmpc_in_parallel(ident_t *loc)
kmp_int32 __kmpc_bound_thread_num(ident_t *loc)
kmp_int32 __kmpc_bound_num_threads(ident_t *loc)
void __kmpc_end_ordered(ident_t *loc, kmp_int32 gtid)
void __kmpc_end_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)
void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_masked(ident_t *loc, kmp_int32 global_tid)
kmp_int32 __kmpc_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_critical_with_hint(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit, uint32_t hint)
kmp_int32 __kmpc_single(ident_t *loc, kmp_int32 global_tid)
void __kmpc_doacross_init(ident_t *loc, int gtid, int num_dims, const struct kmp_dim *dims)
void __kmpc_end_master(ident_t *loc, kmp_int32 global_tid)
void __kmpc_end_single(ident_t *loc, kmp_int32 global_tid)
void __kmpc_ordered(ident_t *loc, kmp_int32 gtid)
kmp_int32 __kmpc_masked(ident_t *loc, kmp_int32 global_tid, kmp_int32 filter)
void __kmpc_critical(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *crit)