diff options
Diffstat (limited to 'runtime/src/kmp_runtime.cpp')
| -rw-r--r-- | runtime/src/kmp_runtime.cpp | 236 |
1 files changed, 162 insertions, 74 deletions
diff --git a/runtime/src/kmp_runtime.cpp b/runtime/src/kmp_runtime.cpp index 7f6c149c792e..dd6e0ff70193 100644 --- a/runtime/src/kmp_runtime.cpp +++ b/runtime/src/kmp_runtime.cpp @@ -1190,8 +1190,8 @@ void __kmp_serialized_parallel(ident_t *loc, kmp_int32 global_tid) { ompt_callbacks.ompt_callback(ompt_callback_parallel_begin)( &(parent_task_info->task_data), &(parent_task_info->frame), - &ompt_parallel_data, team_size, ompt_parallel_invoker_program, - codeptr); + &ompt_parallel_data, team_size, + ompt_parallel_invoker_program | ompt_parallel_team, codeptr); } } #endif // OMPT_SUPPORT @@ -1481,9 +1481,13 @@ int __kmp_fork_call(ident_t *loc, int gtid, int team_size = master_set_numthreads ? master_set_numthreads : get__nproc_2(parent_team, master_tid); + int flags = OMPT_INVOKER(call_context) | + ((microtask == (microtask_t)__kmp_teams_master) + ? ompt_parallel_league + : ompt_parallel_team); ompt_callbacks.ompt_callback(ompt_callback_parallel_begin)( - parent_task_data, ompt_frame, &ompt_parallel_data, team_size, - OMPT_INVOKER(call_context), return_address); + parent_task_data, ompt_frame, &ompt_parallel_data, team_size, flags, + return_address); } master_th->th.ompt_thread_info.state = ompt_state_overhead; } @@ -1512,19 +1516,17 @@ int __kmp_fork_call(ident_t *loc, int gtid, // AC: we are in serialized parallel __kmpc_serialized_parallel(loc, gtid); KMP_DEBUG_ASSERT(parent_team->t.t_serialized > 1); - // AC: need this in order enquiry functions work - // correctly, will restore at join time - parent_team->t.t_serialized--; + #if OMPT_SUPPORT void *dummy; - void **exit_runtime_p; + void **exit_frame_p; ompt_lw_taskteam_t lw_taskteam; if (ompt_enabled.enabled) { __ompt_lw_taskteam_init(&lw_taskteam, master_th, gtid, &ompt_parallel_data, return_address); - exit_runtime_p = &(lw_taskteam.ompt_task_info.frame.exit_frame.ptr); + exit_frame_p = &(lw_taskteam.ompt_task_info.frame.exit_frame.ptr); __ompt_lw_taskteam_link(&lw_taskteam, master_th, 0); // don't use lw_taskteam after linking. content was swaped @@ -1532,19 +1534,23 @@ int __kmp_fork_call(ident_t *loc, int gtid, /* OMPT implicit task begin */ implicit_task_data = OMPT_CUR_TASK_DATA(master_th); if (ompt_enabled.ompt_callback_implicit_task) { - ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( - ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th), - implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial? OMPT_CUR_TASK_INFO(master_th) ->thread_num = __kmp_tid_from_gtid(gtid); + ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( + ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th), + implicit_task_data, 1, + OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); } /* OMPT state */ master_th->th.ompt_thread_info.state = ompt_state_work_parallel; } else { - exit_runtime_p = &dummy; + exit_frame_p = &dummy; } #endif + // AC: need to decrement t_serialized for enquiry functions to work + // correctly, will restore at join time + parent_team->t.t_serialized--; { KMP_TIME_PARTITIONED_BLOCK(OMP_parallel); @@ -1552,26 +1558,27 @@ int __kmp_fork_call(ident_t *loc, int gtid, __kmp_invoke_microtask(microtask, gtid, 0, argc, parent_team->t.t_argv #if OMPT_SUPPORT , - exit_runtime_p + exit_frame_p #endif ); } #if OMPT_SUPPORT - *exit_runtime_p = NULL; if (ompt_enabled.enabled) { + *exit_frame_p = NULL; OMPT_CUR_TASK_INFO(master_th)->frame.exit_frame = ompt_data_none; if (ompt_enabled.ompt_callback_implicit_task) { ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( ompt_scope_end, NULL, implicit_task_data, 1, - OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial? + OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); } + ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th); __ompt_lw_taskteam_unlink(master_th); - if (ompt_enabled.ompt_callback_parallel_end) { ompt_callbacks.ompt_callback(ompt_callback_parallel_end)( - OMPT_CUR_TEAM_DATA(master_th), OMPT_CUR_TASK_DATA(master_th), - OMPT_INVOKER(call_context), return_address); + &ompt_parallel_data, OMPT_CUR_TASK_DATA(master_th), + OMPT_INVOKER(call_context) | ompt_parallel_team, + return_address); } master_th->th.ompt_thread_info.state = ompt_state_overhead; } @@ -1586,6 +1593,15 @@ int __kmp_fork_call(ident_t *loc, int gtid, parent_team->t.t_level++; parent_team->t.t_def_allocator = master_th->th.th_def_allocator; // save +#if OMPT_SUPPORT + if (ompt_enabled.enabled) { + ompt_lw_taskteam_t lw_taskteam; + __ompt_lw_taskteam_init(&lw_taskteam, master_th, gtid, + &ompt_parallel_data, return_address); + __ompt_lw_taskteam_link(&lw_taskteam, master_th, 1, true); + } +#endif + /* Change number of threads in the team if requested */ if (master_set_numthreads) { // The parallel has num_threads clause if (master_set_numthreads < master_th->th.th_teams_size.nth) { @@ -1714,7 +1730,7 @@ int __kmp_fork_call(ident_t *loc, int gtid, #if OMPT_SUPPORT void *dummy; - void **exit_runtime_p; + void **exit_frame_p; ompt_task_info_t *task_info; ompt_lw_taskteam_t lw_taskteam; @@ -1727,19 +1743,21 @@ int __kmp_fork_call(ident_t *loc, int gtid, // don't use lw_taskteam after linking. content was swaped task_info = OMPT_CUR_TASK_INFO(master_th); - exit_runtime_p = &(task_info->frame.exit_frame.ptr); + exit_frame_p = &(task_info->frame.exit_frame.ptr); if (ompt_enabled.ompt_callback_implicit_task) { - ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( - ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th), - &(task_info->task_data), 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial? OMPT_CUR_TASK_INFO(master_th) ->thread_num = __kmp_tid_from_gtid(gtid); + ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( + ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th), + &(task_info->task_data), 1, + OMPT_CUR_TASK_INFO(master_th)->thread_num, + ompt_task_implicit); } /* OMPT state */ master_th->th.ompt_thread_info.state = ompt_state_work_parallel; } else { - exit_runtime_p = &dummy; + exit_frame_p = &dummy; } #endif @@ -1750,25 +1768,27 @@ int __kmp_fork_call(ident_t *loc, int gtid, parent_team->t.t_argv #if OMPT_SUPPORT , - exit_runtime_p + exit_frame_p #endif ); } #if OMPT_SUPPORT if (ompt_enabled.enabled) { - exit_runtime_p = NULL; + *exit_frame_p = NULL; if (ompt_enabled.ompt_callback_implicit_task) { ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( ompt_scope_end, NULL, &(task_info->task_data), 1, - OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial? + OMPT_CUR_TASK_INFO(master_th)->thread_num, + ompt_task_implicit); } - + ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th); __ompt_lw_taskteam_unlink(master_th); if (ompt_enabled.ompt_callback_parallel_end) { ompt_callbacks.ompt_callback(ompt_callback_parallel_end)( - OMPT_CUR_TEAM_DATA(master_th), parent_task_data, - OMPT_INVOKER(call_context), return_address); + &ompt_parallel_data, parent_task_data, + OMPT_INVOKER(call_context) | ompt_parallel_team, + return_address); } master_th->th.ompt_thread_info.state = ompt_state_overhead; } @@ -1800,6 +1820,23 @@ int __kmp_fork_call(ident_t *loc, int gtid, team->t.t_level--; // AC: call special invoker for outer "parallel" of teams construct invoker(gtid); +#if OMPT_SUPPORT + if (ompt_enabled.enabled) { + ompt_task_info_t *task_info = OMPT_CUR_TASK_INFO(master_th); + if (ompt_enabled.ompt_callback_implicit_task) { + ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( + ompt_scope_end, NULL, &(task_info->task_data), 0, + OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_initial); + } + if (ompt_enabled.ompt_callback_parallel_end) { + ompt_callbacks.ompt_callback(ompt_callback_parallel_end)( + &ompt_parallel_data, parent_task_data, + OMPT_INVOKER(call_context) | ompt_parallel_league, + return_address); + } + master_th->th.ompt_thread_info.state = ompt_state_overhead; + } +#endif } else { argv = args; for (i = argc - 1; i >= 0; --i) @@ -1813,7 +1850,7 @@ int __kmp_fork_call(ident_t *loc, int gtid, #if OMPT_SUPPORT void *dummy; - void **exit_runtime_p; + void **exit_frame_p; ompt_task_info_t *task_info; ompt_lw_taskteam_t lw_taskteam; @@ -1824,14 +1861,15 @@ int __kmp_fork_call(ident_t *loc, int gtid, __ompt_lw_taskteam_link(&lw_taskteam, master_th, 0); // don't use lw_taskteam after linking. content was swaped task_info = OMPT_CUR_TASK_INFO(master_th); - exit_runtime_p = &(task_info->frame.exit_frame.ptr); + exit_frame_p = &(task_info->frame.exit_frame.ptr); /* OMPT implicit task begin */ implicit_task_data = OMPT_CUR_TASK_DATA(master_th); if (ompt_enabled.ompt_callback_implicit_task) { ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th), - implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial? + implicit_task_data, 1, __kmp_tid_from_gtid(gtid), + ompt_task_implicit); OMPT_CUR_TASK_INFO(master_th) ->thread_num = __kmp_tid_from_gtid(gtid); } @@ -1839,7 +1877,7 @@ int __kmp_fork_call(ident_t *loc, int gtid, /* OMPT state */ master_th->th.ompt_thread_info.state = ompt_state_work_parallel; } else { - exit_runtime_p = &dummy; + exit_frame_p = &dummy; } #endif @@ -1849,18 +1887,19 @@ int __kmp_fork_call(ident_t *loc, int gtid, __kmp_invoke_microtask(microtask, gtid, 0, argc, args #if OMPT_SUPPORT , - exit_runtime_p + exit_frame_p #endif ); } #if OMPT_SUPPORT if (ompt_enabled.enabled) { - *exit_runtime_p = NULL; + *exit_frame_p = NULL; if (ompt_enabled.ompt_callback_implicit_task) { ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( ompt_scope_end, NULL, &(task_info->task_data), 1, - OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial? + OMPT_CUR_TASK_INFO(master_th)->thread_num, + ompt_task_implicit); } ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th); @@ -1868,7 +1907,8 @@ int __kmp_fork_call(ident_t *loc, int gtid, if (ompt_enabled.ompt_callback_parallel_end) { ompt_callbacks.ompt_callback(ompt_callback_parallel_end)( &ompt_parallel_data, parent_task_data, - OMPT_INVOKER(call_context), return_address); + OMPT_INVOKER(call_context) | ompt_parallel_team, + return_address); } master_th->th.ompt_thread_info.state = ompt_state_overhead; } @@ -2225,12 +2265,11 @@ static inline void __kmp_join_restore_state(kmp_info_t *thread, static inline void __kmp_join_ompt(int gtid, kmp_info_t *thread, kmp_team_t *team, ompt_data_t *parallel_data, - fork_context_e fork_context, void *codeptr) { + int flags, void *codeptr) { ompt_task_info_t *task_info = __ompt_get_task_info_object(0); if (ompt_enabled.ompt_callback_parallel_end) { ompt_callbacks.ompt_callback(ompt_callback_parallel_end)( - parallel_data, &(task_info->task_data), OMPT_INVOKER(fork_context), - codeptr); + parallel_data, &(task_info->task_data), flags, codeptr); } task_info->frame.enter_frame = ompt_data_none; @@ -2263,6 +2302,7 @@ void __kmp_join_call(ident_t *loc, int gtid master_th->th.th_ident = loc; #if OMPT_SUPPORT + void *team_microtask = (void *)team->t.t_pkfn; if (ompt_enabled.enabled) { master_th->th.ompt_thread_info.state = ompt_state_overhead; } @@ -2352,10 +2392,25 @@ void __kmp_join_call(ident_t *loc, int gtid if (master_th->th.th_teams_microtask && !exit_teams && team->t.t_pkfn != (microtask_t)__kmp_teams_master && team->t.t_level == master_th->th.th_teams_level + 1) { - // AC: We need to leave the team structure intact at the end of parallel - // inside the teams construct, so that at the next parallel same (hot) team - // works, only adjust nesting levels - +// AC: We need to leave the team structure intact at the end of parallel +// inside the teams construct, so that at the next parallel same (hot) team +// works, only adjust nesting levels +#if OMPT_SUPPORT + ompt_data_t ompt_parallel_data = ompt_data_none; + if (ompt_enabled.enabled) { + ompt_task_info_t *task_info = __ompt_get_task_info_object(0); + if (ompt_enabled.ompt_callback_implicit_task) { + int ompt_team_size = team->t.t_nproc; + ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( + ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size, + OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); + } + task_info->frame.exit_frame = ompt_data_none; + task_info->task_data = ompt_data_none; + ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th); + __ompt_lw_taskteam_unlink(master_th); + } +#endif /* Decrement our nested depth level */ team->t.t_level--; team->t.t_active_level--; @@ -2394,8 +2449,8 @@ void __kmp_join_call(ident_t *loc, int gtid #if OMPT_SUPPORT if (ompt_enabled.enabled) { - __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, fork_context, - codeptr); + __kmp_join_ompt(gtid, master_th, parent_team, &ompt_parallel_data, + OMPT_INVOKER(fork_context) | ompt_parallel_team, codeptr); } #endif @@ -2424,12 +2479,14 @@ void __kmp_join_call(ident_t *loc, int gtid if (ompt_enabled.enabled) { ompt_task_info_t *task_info = __ompt_get_task_info_object(0); if (ompt_enabled.ompt_callback_implicit_task) { - int ompt_team_size = team->t.t_nproc; + int flags = (team_microtask == (void *)__kmp_teams_master) + ? ompt_task_initial + : ompt_task_implicit; + int ompt_team_size = (flags == ompt_task_initial) ? 0 : team->t.t_nproc; ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size, - OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial? + OMPT_CUR_TASK_INFO(master_th)->thread_num, flags); } - task_info->frame.exit_frame = ompt_data_none; task_info->task_data = ompt_data_none; } @@ -2503,8 +2560,12 @@ void __kmp_join_call(ident_t *loc, int gtid __kmp_release_bootstrap_lock(&__kmp_forkjoin_lock); #if OMPT_SUPPORT + int flags = + OMPT_INVOKER(fork_context) | + ((team_microtask == (void *)__kmp_teams_master) ? ompt_parallel_league + : ompt_parallel_team); if (ompt_enabled.enabled) { - __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, fork_context, + __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, flags, codeptr); } #endif @@ -4432,7 +4493,7 @@ static void __kmp_initialize_team(kmp_team_t *team, int new_nproc, KF_TRACE(10, ("__kmp_initialize_team: exit: team=%p\n", team)); } -#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED +#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED /* Sets full mask for thread and returns old mask, no changes to structures. */ static void __kmp_set_thread_affinity_mask_full_tmp(kmp_affin_mask_t *old_mask) { @@ -4980,7 +5041,7 @@ __kmp_allocate_team(kmp_root_t *root, int new_nproc, int max_nproc, __kmp_partition_places(team); #endif } else { // team->t.t_nproc < new_nproc -#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED +#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED kmp_affin_mask_t *old_mask; if (KMP_AFFINITY_CAPABLE()) { KMP_CPU_ALLOC(old_mask); @@ -5029,7 +5090,7 @@ __kmp_allocate_team(kmp_root_t *root, int new_nproc, int max_nproc, __kmp_reinitialize_team(team, new_icvs, NULL); } -#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED +#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED /* Temporarily set full mask for master thread before creation of workers. The reason is that workers inherit the affinity from master, so if a lot of workers are created on the single core quickly, they @@ -5064,7 +5125,7 @@ __kmp_allocate_team(kmp_root_t *root, int new_nproc, int max_nproc, } } -#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED +#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED if (KMP_AFFINITY_CAPABLE()) { /* Restore initial master thread's affinity mask */ __kmp_set_system_affinity(old_mask, TRUE); @@ -5600,7 +5661,7 @@ void __kmp_free_thread(kmp_info_t *this_th) { void *__kmp_launch_thread(kmp_info_t *this_thr) { int gtid = this_thr->th.th_info.ds.ds_gtid; /* void *stack_data;*/ - kmp_team_t *(*volatile pteam); + kmp_team_t **volatile pteam; KMP_MB(); KA_TRACE(10, ("__kmp_launch_thread: T#%d start\n", gtid)); @@ -5618,18 +5679,15 @@ void *__kmp_launch_thread(kmp_info_t *this_thr) { this_thr->th.ompt_thread_info.state = ompt_state_overhead; this_thr->th.ompt_thread_info.wait_id = 0; this_thr->th.ompt_thread_info.idle_frame = OMPT_GET_FRAME_ADDRESS(0); + this_thr->th.ompt_thread_info.parallel_flags = 0; if (ompt_enabled.ompt_callback_thread_begin) { ompt_callbacks.ompt_callback(ompt_callback_thread_begin)( ompt_thread_worker, thread_data); } - } -#endif - -#if OMPT_SUPPORT - if (ompt_enabled.enabled) { this_thr->th.ompt_thread_info.state = ompt_state_idle; } #endif + /* This is the place where threads wait for work */ while (!TCR_4(__kmp_global.g.g_done)) { KMP_DEBUG_ASSERT(this_thr == __kmp_threads[gtid]); @@ -5647,7 +5705,7 @@ void *__kmp_launch_thread(kmp_info_t *this_thr) { } #endif - pteam = (kmp_team_t * (*))(&this_thr->th.th_team); + pteam = &this_thr->th.th_team; /* have we been allocated? */ if (TCR_SYNC_PTR(*pteam) && !TCR_4(__kmp_global.g.g_done)) { @@ -6956,16 +7014,16 @@ int __kmp_invoke_task_func(int gtid) { #if OMPT_SUPPORT void *dummy; - void **exit_runtime_p; + void **exit_frame_p; ompt_data_t *my_task_data; ompt_data_t *my_parallel_data; int ompt_team_size; if (ompt_enabled.enabled) { - exit_runtime_p = &( + exit_frame_p = &( team->t.t_implicit_task_taskdata[tid].ompt_task_info.frame.exit_frame.ptr); } else { - exit_runtime_p = &dummy; + exit_frame_p = &dummy; } my_task_data = @@ -6975,7 +7033,7 @@ int __kmp_invoke_task_func(int gtid) { ompt_team_size = team->t.t_nproc; ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( ompt_scope_begin, my_parallel_data, my_task_data, ompt_team_size, - __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial? + __kmp_tid_from_gtid(gtid), ompt_task_implicit); OMPT_CUR_TASK_INFO(this_thr)->thread_num = __kmp_tid_from_gtid(gtid); } #endif @@ -6994,11 +7052,12 @@ int __kmp_invoke_task_func(int gtid) { tid, (int)team->t.t_argc, (void **)team->t.t_argv #if OMPT_SUPPORT , - exit_runtime_p + exit_frame_p #endif ); #if OMPT_SUPPORT - *exit_runtime_p = NULL; + *exit_frame_p = NULL; + this_thr->th.ompt_thread_info.parallel_flags |= ompt_parallel_team; #endif #if KMP_STATS_ENABLED @@ -7077,7 +7136,22 @@ int __kmp_invoke_teams_master(int gtid) { (void *)__kmp_teams_master); #endif __kmp_run_before_invoked_task(gtid, 0, this_thr, team); +#if OMPT_SUPPORT + int tid = __kmp_tid_from_gtid(gtid); + ompt_data_t *task_data = + &team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data; + ompt_data_t *parallel_data = &team->t.ompt_team_info.parallel_data; + if (ompt_enabled.ompt_callback_implicit_task) { + ompt_callbacks.ompt_callback(ompt_callback_implicit_task)( + ompt_scope_begin, parallel_data, task_data, team->t.t_nproc, tid, + ompt_task_initial); + OMPT_CUR_TASK_INFO(this_thr)->thread_num = tid; + } +#endif __kmp_teams_master(gtid); +#if OMPT_SUPPORT + this_thr->th.ompt_thread_info.parallel_flags |= ompt_parallel_league; +#endif __kmp_run_after_invoked_task(gtid, 0, this_thr, team); return 1; } @@ -7118,19 +7192,32 @@ void __kmp_push_num_teams(ident_t *id, int gtid, int num_teams, thr->th.th_set_nproc = thr->th.th_teams_size.nteams = num_teams; // Remember the number of threads for inner parallel regions + if (!TCR_4(__kmp_init_middle)) + __kmp_middle_initialize(); // get internal globals calculated + KMP_DEBUG_ASSERT(__kmp_avail_proc); + KMP_DEBUG_ASSERT(__kmp_dflt_team_nth); if (num_threads == 0) { - if (!TCR_4(__kmp_init_middle)) - __kmp_middle_initialize(); // get __kmp_avail_proc calculated num_threads = __kmp_avail_proc / num_teams; + // adjust num_threads w/o warning as it is not user setting + // num_threads = min(num_threads, nthreads-var, thread-limit-var) + // no thread_limit clause specified - do not change thread-limit-var ICV + if (num_threads > __kmp_dflt_team_nth) { + num_threads = __kmp_dflt_team_nth; // honor nthreads-var ICV + } + if (num_threads > thr->th.th_current_task->td_icvs.thread_limit) { + num_threads = thr->th.th_current_task->td_icvs.thread_limit; + } // prevent team size to exceed thread-limit-var if (num_teams * num_threads > __kmp_teams_max_nth) { - // adjust num_threads w/o warning as it is not user setting num_threads = __kmp_teams_max_nth / num_teams; } } else { // This thread will be the master of the league masters // Store new thread limit; old limit is saved in th_cg_roots list thr->th.th_current_task->td_icvs.thread_limit = num_threads; - + // num_threads = min(num_threads, nthreads-var) + if (num_threads > __kmp_dflt_team_nth) { + num_threads = __kmp_dflt_team_nth; // honor nthreads-var ICV + } if (num_teams * num_threads > __kmp_teams_max_nth) { int new_threads = __kmp_teams_max_nth / num_teams; if (!__kmp_reserve_warn) { // user asked for too many threads @@ -8023,7 +8110,8 @@ __kmp_determine_reduction_method( int atomic_available = FAST_REDUCTION_ATOMIC_METHOD_GENERATED; -#if KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS64 +#if KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || \ + KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 #if KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD || \ KMP_OS_OPENBSD || KMP_OS_WINDOWS || KMP_OS_DARWIN || KMP_OS_HURD |