[OMPT] Fix parallel_data in implicit barrier-end

This is required to be NULL for implicit barriers at the end of a
parallel region. Noticed in review of D43191.

Differential Revision: https://reviews.llvm.org/D43308

git-svn-id: https://llvm.org/svn/llvm-project/openmp/trunk@325922 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 135 insertions, 98 deletions

diff --git a/runtime/src/kmp_barrier.cpp b/runtime/src/kmp_barrier.cpp
index cc125dc..ca107cd 100644
--- a/runtime/src/kmp_barrier.cpp
+++ b/runtime/src/kmp_barrier.cpp
@@ -1883,34 +1883,31 @@ void __kmp_fork_barrier(int gtid, int tid) {
   }
 
 #if OMPT_SUPPORT
-  if (ompt_enabled.enabled) {
-    if (this_thr->th.ompt_thread_info.state ==
-        omp_state_wait_barrier_implicit) {
-      int ds_tid = this_thr->th.th_info.ds.ds_tid;
-      ompt_data_t *tId = (team) ? OMPT_CUR_TASK_DATA(this_thr)
-                                : &(this_thr->th.ompt_thread_info.task_data);
-      this_thr->th.ompt_thread_info.state = omp_state_overhead;
+  if (ompt_enabled.enabled &&
+      this_thr->th.ompt_thread_info.state == omp_state_wait_barrier_implicit) {
+    int ds_tid = this_thr->th.th_info.ds.ds_tid;
+    ompt_data_t *task_data = (team)
+                                 ? OMPT_CUR_TASK_DATA(this_thr)
+                                 : &(this_thr->th.ompt_thread_info.task_data);
+    this_thr->th.ompt_thread_info.state = omp_state_overhead;
 #if OMPT_OPTIONAL
-      void *codeptr = NULL;
-      if (KMP_MASTER_TID(ds_tid) &&
-          (ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait) ||
-           ompt_callbacks.ompt_callback(ompt_callback_sync_region)))
-        codeptr = team->t.ompt_team_info.master_return_address;
-      if (ompt_enabled.ompt_callback_sync_region_wait) {
-        ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
-            ompt_sync_region_barrier, ompt_scope_end, NULL, tId, codeptr);
-      }
-      if (ompt_enabled.ompt_callback_sync_region) {
-        ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
-            ompt_sync_region_barrier, ompt_scope_end, NULL, tId, codeptr);
-      }
+    void *codeptr = NULL;
+    if (KMP_MASTER_TID(ds_tid) &&
+        (ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait) ||
+         ompt_callbacks.ompt_callback(ompt_callback_sync_region)))
+      codeptr = team->t.ompt_team_info.master_return_address;
+    if (ompt_enabled.ompt_callback_sync_region_wait) {
+      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
+          ompt_sync_region_barrier, ompt_scope_end, NULL, task_data, codeptr);
+    }
+    if (ompt_enabled.ompt_callback_sync_region) {
+      ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
+          ompt_sync_region_barrier, ompt_scope_end, NULL, task_data, codeptr);
+    }
 #endif
-      if (!KMP_MASTER_TID(ds_tid) && ompt_enabled.ompt_callback_implicit_task) {
-        ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
-            ompt_scope_end, NULL, tId, 0, ds_tid);
-      }
-      // return to idle state
-      this_thr->th.ompt_thread_info.state = omp_state_overhead;
+    if (!KMP_MASTER_TID(ds_tid) && ompt_enabled.ompt_callback_implicit_task) {
+      ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+          ompt_scope_end, NULL, task_data, 0, ds_tid);
     }
   }
 #endif
diff --git a/runtime/src/kmp_runtime.cpp b/runtime/src/kmp_runtime.cpp
index c05e973..f902f04 100644
--- a/runtime/src/kmp_runtime.cpp
+++ b/runtime/src/kmp_runtime.cpp
@@ -7164,10 +7164,10 @@ void __kmp_internal_join(ident_t *id, int gtid, kmp_team_t *team) {
 
   __kmp_join_barrier(gtid); /* wait for everyone */
 #if OMPT_SUPPORT
-  int ds_tid = this_thr->th.th_info.ds.ds_tid;
-  if (this_thr->th.ompt_thread_info.state == omp_state_wait_barrier_implicit) {
-    ompt_data_t *tId = OMPT_CUR_TASK_DATA(this_thr);
-    ompt_data_t *pId = OMPT_CUR_TEAM_DATA(this_thr);
+  if (ompt_enabled.enabled &&
+      this_thr->th.ompt_thread_info.state == omp_state_wait_barrier_implicit) {
+    int ds_tid = this_thr->th.th_info.ds.ds_tid;
+    ompt_data_t *task_data = OMPT_CUR_TASK_DATA(this_thr);
     this_thr->th.ompt_thread_info.state = omp_state_overhead;
 #if OMPT_OPTIONAL
     void *codeptr = NULL;
@@ -7178,19 +7178,17 @@ void __kmp_internal_join(ident_t *id, int gtid, kmp_team_t *team) {
 
     if (ompt_enabled.ompt_callback_sync_region_wait) {
       ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
-          ompt_sync_region_barrier, ompt_scope_end, pId, tId, codeptr);
+          ompt_sync_region_barrier, ompt_scope_end, NULL, task_data, codeptr);
     }
     if (ompt_enabled.ompt_callback_sync_region) {
       ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
-          ompt_sync_region_barrier, ompt_scope_end, pId, tId, codeptr);
+          ompt_sync_region_barrier, ompt_scope_end, NULL, task_data, codeptr);
     }
 #endif
     if (!KMP_MASTER_TID(ds_tid) && ompt_enabled.ompt_callback_implicit_task) {
       ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
-          ompt_scope_end, NULL, tId, 0, ds_tid);
+          ompt_scope_end, NULL, task_data, 0, ds_tid);
     }
-    // return to idle state
-    this_thr->th.ompt_thread_info.state = omp_state_overhead;
   }
 #endif
 
diff --git a/runtime/test/ompt/parallel/normal.c b/runtime/test/ompt/parallel/normal.c
index 015ff5c..2cc9ce1 100644
--- a/runtime/test/ompt/parallel/normal.c
+++ b/runtime/test/ompt/parallel/normal.c
@@ -1,11 +1,11 @@
 // RUN: %libomp-compile-and-run | FileCheck %s
-// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// RUN: %libomp-compile-and-run | %sort-threads \
+// RUN:                         | FileCheck --check-prefix=THREADS %s
 // REQUIRES: ompt
 #include "callback.h"
 
-int main()
-{
-  #pragma omp parallel num_threads(4)
+int main() {
+#pragma omp parallel num_threads(4)
   {
     print_ids(0);
     print_ids(1);
@@ -13,78 +13,120 @@ int main()
   print_fuzzy_address(1);
 
   // Check if libomp supports the callbacks for this test.
-  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_thread_begin'
-  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_thread_end'
-  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_parallel_begin'
-  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_parallel_end'
-  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_implicit_task'
-
+  // CHECK-NOT: {{^}}0: Could not register callback
 
   // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
-  
+
   // make sure initial data pointers are null
   // CHECK-NOT: 0: parallel_data initially not null
   // CHECK-NOT: 0: task_data initially not null
   // CHECK-NOT: 0: thread_data initially not null
 
-  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter={{0x[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
-
-  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // Only check callback names, arguments are verified in THREADS below.
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin
 
-  // Note that we cannot ensure that the worker threads have already called barrier_end and implicit_task_end before parallel_end!
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
 
-  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // Note that we cannot ensure that the worker threads have already called
+  // barrier_end and implicit_task_end before parallel_end!
 
-  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin
 
-  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin
 
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]], codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
-  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin
 
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end
 
   // THREADS: 0: NULL_POINTER=[[NULL:.*$]]
-  // THREADS: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_thread_begin: thread_type=ompt_thread_initial=1, thread_id=[[MASTER_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter={{0x[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker={{.*}}
-
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]]
-  // THREADS-NOT: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
-
-  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
-  // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
-
-  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
-  // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
-
-  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
-  // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_thread_begin
+  // THREADS-SAME: thread_type=ompt_thread_initial=1, thread_id=[[MASTER_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin
+  // THREADS-SAME: parent_task_id=[[PARENT_TASK_ID:[0-9]+]]
+  // THREADS-SAME: parent_task_frame.exit=[[NULL]]
+  // THREADS-SAME: parent_task_frame.reenter={{0x[0-f]+}}
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4
+  // THREADS-SAME: codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1
+  // THREADS-SAME: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]]
+  // THREADS-SAME: task_id=[[PARENT_TASK_ID]]
+  // THREADS-NOT: ompt_event_implicit_task_end
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS-SAME: codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end
+  // parallel_id is 0 because the region ended in the barrier!
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS-SAME: codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+
+  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin
+  // THREADS-SAME: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1
+  // THREADS-SAME: parallel_id=[[IMPLICIT_PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[PARENT_TASK_ID]]
+  // THREADS-NOT: ompt_event_implicit_task_end
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end
+  // parallel_id is 0 because the region ended in the barrier!
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+
+  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin
+  // THREADS-SAME: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1
+  // THREADS-SAME: parallel_id=[[IMPLICIT_PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[PARENT_TASK_ID]]
+  // THREADS-NOT: ompt_event_implicit_task_end
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end
+  // parallel_id is 0 because the region ended in the barrier!
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+
+  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin
+  // THREADS-SAME: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1
+  // THREADS-SAME: parallel_id=[[IMPLICIT_PARALLEL_ID]]
+  // THREADS-SAME: task_id=[[PARENT_TASK_ID]]
+  // THREADS-NOT: ompt_event_implicit_task_end
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin
+  // THREADS-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end
+  // parallel_id is 0 because the region ended in the barrier!
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
+  // THREADS-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
 
   return 0;
 }
diff --git a/runtime/test/ompt/tasks/serialized.c b/runtime/test/ompt/tasks/serialized.c
index dc2000f..12a0281 100644
--- a/runtime/test/ompt/tasks/serialized.c
+++ b/runtime/test/ompt/tasks/serialized.c
@@ -119,8 +119,8 @@ int main() {
   // CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // CHECK-SAME: exit_frame=[[NULL]], reenter_frame=[[NULL]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end
-  // FIXME: parallel_id should be 0 because the region ended in the barrier!
-  // CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // parallel_id is 0 because the region ended in the barrier!
+  // CHECK-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
   // CHECK-SAME: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]