Add more conditions to check whether task waiting is necessary in kmp_omp_taskwait()

[vhscampos]

Summary

From:

• comment in file kmp_tasking.c line 1489
• file kmp_taskdeps.cpp line 558
• file kmp_taskdeps.cpp line 491

I consider that, not only 'team_serial', but also 'tasking_ser' and 'final' flags from kmp_taskdata_t should be used to determine whether it's necessary to read the volatile flag.

By the same logic used in wait_deps and task_with_deps, any of these three flags being set is a sufficient condition for the task to skip the waiting.

Please let me know your opinions. Thanks.

[AndreyChurbanov]

runtime/src/kmp_tasking.c

1483

The check "! taskdata->td_flags.tasking_ser" looks redundant here, because this condition is effectively the same as "__kmp_tasking_mode != tskm_immediate_exec" that has already been checked above at the line 1446.

The check of the td_flags.final flag does make sense.

[vhscampos]

Given that checking 'tasking_ser' is redundant, I removed it.

[AndreyChurbanov]

LGTM

[AndreyChurbanov]

committed r285678.

runtime/src/kmp_tasking.c

#endif /* USE_ITT_BUILD */
         taskdata->td_taskwait_thread   = gtid + 1;
 
 #if USE_ITT_BUILD
         void * itt_sync_obj = __kmp_itt_taskwait_object( gtid );
         if ( itt_sync_obj != NULL )
             __kmp_itt_taskwait_starting( gtid, itt_sync_obj );
 #endif /* USE_ITT_BUILD */
 
+        bool must_wait = ! taskdata->td_flags.team_serial && ! taskdata->td_flags.final;
+
 #if OMP_45_ENABLED
-        if ( ! taskdata->td_flags.team_serial || (thread->th.th_task_team != NULL && thread->th.th_task_team->tt.tt_found_proxy_tasks) )
-#else
-        if ( ! taskdata->td_flags.team_serial )
+        must_wait = must_wait || (thread->th.th_task_team != NULL && thread->th.th_task_team->tt.tt_found_proxy_tasks);
 #endif
+        if (must_wait)
         {
-            // GEH: if team serialized, avoid reading the volatile variable below.
             kmp_flag_32 flag(&(taskdata->td_incomplete_child_tasks), 0U);
             while ( TCR_4(taskdata -> td_incomplete_child_tasks) != 0 ) {
                 flag.execute_tasks(thread, gtid, FALSE, &thread_finished
                                    USE_ITT_BUILD_ARG(itt_sync_obj), __kmp_task_stealing_constraint );
             }
         }
 #if USE_ITT_BUILD
         if ( itt_sync_obj != NULL )