Index: runtime/src/kmp_barrier.cpp
===================================================================
--- runtime/src/kmp_barrier.cpp
+++ runtime/src/kmp_barrier.cpp
@@ -1623,6 +1623,8 @@
           ompt_sync_region_barrier, ompt_scope_begin, my_parallel_data,
           my_task_data, codeptr);
     }
+    if (!KMP_MASTER_TID(ds_tid))
+      this_thr->th.ompt_thread_info.task_data = *OMPT_CUR_TASK_DATA(this_thr);
 #endif
     this_thr->th.ompt_thread_info.state = omp_state_wait_barrier_implicit;
   }
Index: runtime/src/kmp_runtime.cpp
===================================================================
--- runtime/src/kmp_runtime.cpp
+++ runtime/src/kmp_runtime.cpp
@@ -5669,7 +5669,6 @@
         __ompt_get_task_info_object(0)->frame.exit_frame = NULL;
 
         this_thr->th.ompt_thread_info.state = omp_state_overhead;
-        this_thr->th.ompt_thread_info.task_data = *OMPT_CUR_TASK_DATA(this_thr);
       }
 #endif
       /* join barrier after parallel region */
Index: runtime/test/ompt/synchronization/barrier/implicit_task_data.c
===================================================================
--- /dev/null
+++ runtime/test/ompt/synchronization/barrier/implicit_task_data.c
@@ -0,0 +1,183 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#define _BSD_SOURCE
+#include <stdio.h>
+#include <unistd.h>
+#include <inttypes.h>
+#include <omp.h>
+#include <ompt.h>
+
+static const char* ompt_thread_type_t_values[] = {
+  NULL,
+  "ompt_thread_initial",
+  "ompt_thread_worker",
+  "ompt_thread_other"
+};
+
+static ompt_get_unique_id_t ompt_get_unique_id;
+static ompt_get_thread_data_t ompt_get_thread_data;
+
+int main()
+{
+  #pragma omp parallel num_threads(4)
+  {
+    #pragma omp master
+    {
+      sleep(1);
+    }
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_sync_region'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_sync_region_wait'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // master thread implicit barrier at parallel end
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id=0, task_id=[[TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id=0, task_id=[[TASK_ID]], codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id=0, task_id=[[TASK_ID]], codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=0, task_id=[[TASK_ID]], codeptr_ra={{0x[0-f]*}}
+
+
+  // worker thread implicit barrier at parallel end
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id=0, task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id=0, task_id=[[TASK_ID]], codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id=0, task_id=[[TASK_ID]], codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=0, task_id=[[TASK_ID]], codeptr_ra=[[NULL]]
+
+  return 0;
+}
+
+static void
+on_ompt_callback_thread_begin(
+  ompt_thread_type_t thread_type,
+  ompt_data_t *thread_data)
+{
+  if(thread_data->ptr)
+    printf("%s\n", "0: thread_data initially not null");
+  thread_data->value = ompt_get_unique_id();
+  printf("%" PRIu64 ": ompt_event_thread_begin: thread_type=%s=%d, thread_id=%" PRIu64 "\n", ompt_get_thread_data()->value, ompt_thread_type_t_values[thread_type], thread_type, thread_data->value);
+}
+
+static void
+on_ompt_callback_sync_region(
+  ompt_sync_region_kind_t kind,
+  ompt_scope_endpoint_t endpoint,
+  ompt_data_t *parallel_data,
+  ompt_data_t *task_data,
+  const void *codeptr_ra)
+{
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      task_data->value = ompt_get_unique_id();
+      switch(kind)
+      {
+        case ompt_sync_region_barrier:
+          printf("%" PRIu64 ": ompt_event_barrier_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskwait:
+          printf("%" PRIu64 ": ompt_event_taskwait_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskgroup:
+          printf("%" PRIu64 ": ompt_event_taskgroup_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+          break;
+      }
+      break;
+    case ompt_scope_end:
+      switch(kind)
+      {
+        case ompt_sync_region_barrier:
+          printf("%" PRIu64 ": ompt_event_barrier_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskwait:
+          printf("%" PRIu64 ": ompt_event_taskwait_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskgroup:
+          printf("%" PRIu64 ": ompt_event_taskgroup_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, codeptr_ra);
+          break;
+      }
+      break;
+  }
+}
+
+static void
+on_ompt_callback_sync_region_wait(
+  ompt_sync_region_kind_t kind,
+  ompt_scope_endpoint_t endpoint,
+  ompt_data_t *parallel_data,
+  ompt_data_t *task_data,
+  const void *codeptr_ra)
+{
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      switch(kind)
+      {
+        case ompt_sync_region_barrier:
+          printf("%" PRIu64 ": ompt_event_wait_barrier_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskwait:
+          printf("%" PRIu64 ": ompt_event_wait_taskwait_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskgroup:
+          printf("%" PRIu64 ": ompt_event_wait_taskgroup_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+          break;
+      }
+      break;
+    case ompt_scope_end:
+      switch(kind)
+      {
+        case ompt_sync_region_barrier:
+          printf("%" PRIu64 ": ompt_event_wait_barrier_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskwait:
+          printf("%" PRIu64 ": ompt_event_wait_taskwait_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, codeptr_ra);
+          break;
+        case ompt_sync_region_taskgroup:
+          printf("%" PRIu64 ": ompt_event_wait_taskgroup_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, codeptr_ra);
+          break;
+      }
+      break;
+  }
+}
+
+#define register_callback_t(name, type)                       \
+do{                                                           \
+  type f_##name = &on_##name;                                 \
+  if (ompt_set_callback(name, (ompt_callback_t)f_##name) ==   \
+      ompt_set_never)                                         \
+    printf("0: Could not register callback '" #name "'\n");   \
+}while(0)
+
+#define register_callback(name) register_callback_t(name, name##_t)
+
+int ompt_initialize(
+  ompt_function_lookup_t lookup,
+  ompt_data_t *tool_data)
+{
+  ompt_set_callback_t ompt_set_callback;
+  ompt_set_callback = (ompt_set_callback_t) lookup("ompt_set_callback");
+  ompt_get_unique_id = (ompt_get_unique_id_t) lookup("ompt_get_unique_id");
+  ompt_get_thread_data = (ompt_get_thread_data_t) lookup("ompt_get_thread_data");
+  register_callback(ompt_callback_sync_region);
+  register_callback_t(ompt_callback_sync_region_wait, ompt_callback_sync_region_t);
+  register_callback(ompt_callback_thread_begin);
+  printf("0: NULL_POINTER=%p\n", (void*)NULL);
+  return 1; //success
+}
+
+void ompt_finalize(ompt_data_t *tool_data)
+{
+  printf("0: ompt_event_runtime_shutdown\n");
+}
+
+ompt_start_tool_result_t* ompt_start_tool(
+  unsigned int omp_version,
+  const char *runtime_version)
+{
+  static ompt_start_tool_result_t ompt_start_tool_result = {&ompt_initialize,&ompt_finalize, 0};
+  return &ompt_start_tool_result;
+}