[OMPT] Reset task exit frame when execution is finished
authorJonas Hahnfeld <Hahnfeld@itc.rwth-aachen.de>
Wed, 14 Sep 2016 13:59:19 +0000 (13:59 +0000)
committerJonas Hahnfeld <Hahnfeld@itc.rwth-aachen.de>
Wed, 14 Sep 2016 13:59:19 +0000 (13:59 +0000)
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.

Patch by Joachim Protze!

Differential Revision: https://reviews.llvm.org/D23307

llvm-svn: 281465

openmp/runtime/src/kmp_runtime.c
openmp/runtime/test/ompt/parallel/nested.c

index fb97c37..06b5d58 100644 (file)
@@ -1548,6 +1548,7 @@ __kmp_fork_call(
             }
 
 #if OMPT_SUPPORT
+            *exit_runtime_p = NULL;
             if (ompt_enabled) {
 #if OMPT_TRACE
                 lw_taskteam.ompt_task_info.frame.exit_runtime_frame = NULL;
@@ -1745,6 +1746,7 @@ __kmp_fork_call(
                 }
 
 #if OMPT_SUPPORT
+                *exit_runtime_p = NULL;
                 if (ompt_enabled) {
                     lw_taskteam.ompt_task_info.frame.exit_runtime_frame = NULL;
 
@@ -1851,6 +1853,7 @@ __kmp_fork_call(
                 }
 
 #if OMPT_SUPPORT
+                *exit_runtime_p = NULL;
                 if (ompt_enabled) {
 #if OMPT_TRACE
                     lw_taskteam.ompt_task_info.frame.exit_runtime_frame = NULL;
@@ -6838,6 +6841,9 @@ __kmp_invoke_task_func( int gtid )
                                      , exit_runtime_p
 #endif
                                      );
+#if OMPT_SUPPORT
+        *exit_runtime_p = NULL;
+#endif
     }
 
 #if USE_ITT_BUILD
index 63c5c4f..6847ab6 100644 (file)
@@ -64,12 +64,20 @@ int main()
   // 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]]