[OMPT] Reset task exit frame when execution is finished

[protze.joachim]

Summary

The exit address is set when execution of a task is started and should be reset as soon as the execution is finished.
Especially for the asm implementation of __kmp_invoke_microtask, resetting in this call would be painfull, so reset just after the invokation.

The testcase shows the effect of this patch:
Before, the implicit barriers at the end of an implicit task would see an exit address for the implicit task.

This barrier is a task scheduling point. Thus, any explicit task scheduled there would see an exit, but no reenter address for the implicit task.

[jlpeyton]

runtime/src/kmp_runtime.c

1549 ↗

(On Diff #67313)

Use NULL instead. And can this be moved down to the #if OMPT_SUPPORT section just 4 to 5 lines below?

1749 ↗

(On Diff #67313)

Same here.

1858 ↗

(On Diff #67313)

Same here.

6836 ↗

(On Diff #67313)

Use NULL instead.

[jlpeyton]

Adding Jonas and John to list of reviewers.

[protze.joachim]

Merged OMPT - blocks where appropriate, replaced =0 by =NULL.

[Hahnfeld]

LGTM from an OMPT perspective

[AndreyChurbanov]

LGTM

[jlpeyton]

LGTM

openmp/trunk/runtime/src/kmp_runtime.c

#endif
                 __kmp_invoke_microtask( microtask, gtid, 0, argc, parent_team->t.t_argv
 #if OMPT_SUPPORT
                                         , exit_runtime_p
 #endif
                                         );
             }
 
 #if OMPT_SUPPORT
+            *exit_runtime_p = NULL;
             if (ompt_enabled) {
 #if OMPT_TRACE
                 lw_taskteam.ompt_task_info.frame.exit_runtime_frame = NULL;
 
                 if (ompt_callbacks.ompt_callback(ompt_event_implicit_task_end)) {
                     ompt_callbacks.ompt_callback(ompt_event_implicit_task_end)(
                         ompt_parallel_id, ompt_task_id);
                 }
#endif
                     __kmp_invoke_microtask( microtask, gtid, 0, argc, parent_team->t.t_argv
 #if OMPT_SUPPORT
                         , exit_runtime_p
 #endif
                     );
                 }
 
 #if OMPT_SUPPORT
+                *exit_runtime_p = NULL;
                 if (ompt_enabled) {
                     lw_taskteam.ompt_task_info.frame.exit_runtime_frame = NULL;
 
 #if OMPT_TRACE
                     if (ompt_callbacks.ompt_callback(ompt_event_implicit_task_end)) {
                         ompt_callbacks.ompt_callback(ompt_event_implicit_task_end)(
                             ompt_parallel_id, ompt_task_id);
                     }
#endif
                     __kmp_invoke_microtask( microtask, gtid, 0, argc, args
 #if OMPT_SUPPORT
                         , exit_runtime_p
 #endif
                     );
                 }
 
 #if OMPT_SUPPORT
+                *exit_runtime_p = NULL;
                 if (ompt_enabled) {
 #if OMPT_TRACE
                     lw_taskteam.ompt_task_info.frame.exit_runtime_frame = NULL;
 
                     if (ompt_callbacks.ompt_callback(ompt_event_implicit_task_end)) {
                         ompt_callbacks.ompt_callback(ompt_event_implicit_task_end)(
                             my_parallel_id, my_task_id);
                     }
#endif
         KMP_TIME_PARTITIONED_BLOCK(OMP_parallel);
         KMP_SET_THREAD_STATE_BLOCK(IMPLICIT_TASK);
         rc = __kmp_invoke_microtask( (microtask_t) TCR_SYNC_PTR(team->t.t_pkfn),
                                      gtid, tid, (int) team->t.t_argc, (void **) team->t.t_argv
 #if OMPT_SUPPORT
                                      , exit_runtime_p
 #endif
                                      );
+#if OMPT_SUPPORT
+        *exit_runtime_p = NULL;
+#endif
     }
 
 #if USE_ITT_BUILD
     if ( __itt_stack_caller_create_ptr ) {
         __kmp_itt_stack_callee_leave( (__itt_caller)team->t.t_stack_id ); // inform ittnotify about leaving user's code
     }
 #endif /* USE_ITT_BUILD */
     __kmp_run_after_invoked_task( gtid, tid, this_thr, team );

openmp/trunk/runtime/test/ompt/parallel/nested.c

int main()
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[EXIT]], parent_task_frame.reenter=[[REENTER]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION:0x[0-f]+]], invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
   // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[NESTED_EXIT:0x[0-f]+]]
   // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_EXIT]], reenter_frame=[[NULL]]
   // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
   // THREADS: {{^}}[[MASTER_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[NESTED_REENTER:0x[0-f]+]]
   // THREADS-NOT: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
+  // explicit barrier
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_EXIT]], reenter_frame=[[NESTED_REENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // implicit barrier
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // THREADS-NOT: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
+  // implicit barrier
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
 
   // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
   // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]