Update implementation of OMPT to the specification OpenMP 5.0 Preview 1 (TR4).

The code is tested to work with latest clang, GNU and Intel compiler. The implementation
is optimized for low overhead when no tool is attached shifting the cost to execution with
tool attached.

This patch does not implement OMPT for libomptarget.

Patch by Simon Convent and Joachim Protze

Differential Revision: https://reviews.llvm.org/D38185

llvm-svn: 317085

56 files changed, 3610 insertions, 396 deletions

diff --git a/openmp/runtime/test/ompt/callback.h b/openmp/runtime/test/ompt/callback.h
index d2d8966a8b9..9b1945663ed 100644..100755
--- a/openmp/runtime/test/ompt/callback.h
+++ b/openmp/runtime/test/ompt/callback.h
@@ -1,119 +1,660 @@
+#define _BSD_SOURCE
 #include <stdio.h>
 #include <inttypes.h>
+#include <omp.h>
 #include <ompt.h>
+#include "ompt-signal.h"
 
-static ompt_get_task_id_t ompt_get_task_id;
-static ompt_get_task_frame_t ompt_get_task_frame;
-static ompt_get_thread_id_t ompt_get_thread_id;
-static ompt_get_parallel_id_t ompt_get_parallel_id;
+static const char* ompt_thread_type_t_values[] = {
+  NULL,
+  "ompt_thread_initial",
+  "ompt_thread_worker",
+  "ompt_thread_other"
+};
+
+static const char* ompt_task_status_t_values[] = {
+  NULL,
+  "ompt_task_complete",
+  "ompt_task_yield",
+  "ompt_task_cancel",
+  "ompt_task_others"
+};
+static const char* ompt_cancel_flag_t_values[] = {
+  "ompt_cancel_parallel",
+  "ompt_cancel_sections",
+  "ompt_cancel_do",
+  "ompt_cancel_taskgroup",
+  "ompt_cancel_activated",
+  "ompt_cancel_detected",
+  "ompt_cancel_discarded_task"
+};
+
+static ompt_set_callback_t ompt_set_callback;
+static ompt_get_task_info_t ompt_get_task_info;
+static ompt_get_thread_data_t ompt_get_thread_data;
+static ompt_get_parallel_info_t ompt_get_parallel_info;
+static ompt_get_unique_id_t ompt_get_unique_id;
+static ompt_get_num_places_t ompt_get_num_places;
+static ompt_get_place_proc_ids_t ompt_get_place_proc_ids;
+static ompt_get_place_num_t ompt_get_place_num;
+static ompt_get_partition_place_nums_t ompt_get_partition_place_nums;
+static ompt_get_proc_id_t ompt_get_proc_id;
+static ompt_enumerate_states_t ompt_enumerate_states;
+static ompt_enumerate_mutex_impls_t ompt_enumerate_mutex_impls;
 
 static void print_ids(int level)
 {
-  ompt_frame_t* frame = ompt_get_task_frame(level);
-  printf("%" PRIu64 ": level %d: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", exit_frame=%p, reenter_frame=%p\n", ompt_get_thread_id(), level, ompt_get_parallel_id(level), ompt_get_task_id(level), frame->exit_runtime_frame, frame->reenter_runtime_frame);
+  ompt_frame_t* frame ;
+  ompt_data_t* parallel_data;
+  ompt_data_t* task_data;
+  int exists_task = ompt_get_task_info(level, NULL, &task_data, &frame, &parallel_data, NULL);
+  if (frame)
+  {
+    printf("%" PRIu64 ": task level %d: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", exit_frame=%p, reenter_frame=%p\n", ompt_get_thread_data()->value, level, exists_task ? parallel_data->value : 0, exists_task ? task_data->value : 0, frame->exit_runtime_frame, frame->reenter_runtime_frame);
+  }
+  else
+    printf("%" PRIu64 ": task level %d: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", frame=%p\n", ompt_get_thread_data()->value, level, exists_task ? parallel_data->value : 0, exists_task ? task_data->value : 0, frame);
 }
 
 #define print_frame(level)\
 do {\
-  printf("%" PRIu64 ": __builtin_frame_address(%d)=%p\n", ompt_get_thread_id(), level, __builtin_frame_address(level));\
+  printf("%" PRIu64 ": __builtin_frame_address(%d)=%p\n", ompt_get_thread_data()->value, level, __builtin_frame_address(level));\
 } while(0)
 
+#define print_current_address(id)\
+{}              /* Empty block between "#pragma omp ..." and __asm__ statement as a workaround for icc bug */ \
+__asm__("nop"); /* provide an instruction as jump target (compiler would insert an instruction if label is target of a jmp ) */ \
+ompt_label_##id:\
+    printf("%" PRIu64 ": current_address=%p or %p\n", ompt_get_thread_data()->value, (char*)(&& ompt_label_##id)-1, (char*)(&& ompt_label_##id)-4) 
+    /* "&& label" returns the address of the label (GNU extension); works with gcc, clang, icc */
+    /* for void-type runtime function, the label is after the nop (-1), for functions with return value, there is a mov instruction before the label (-4) */
+
+#define print_fuzzy_address(id)\
+{}              /* Empty block between "#pragma omp ..." and __asm__ statement as a workaround for icc bug */ \
+__asm__("nop"); /* provide an instruction as jump target (compiler would insert an instruction if label is target of a jmp ) */ \
+ompt_label_##id:\
+    printf("%" PRIu64 ": fuzzy_address=0x%lx or 0x%lx\n", ompt_get_thread_data()->value, ((uint64_t)(char*)(&& ompt_label_##id))/256-1, ((uint64_t)(char*)(&& ompt_label_##id))/256) 
+    /* "&& label" returns the address of the label (GNU extension); works with gcc, clang, icc */
+    /* for void-type runtime function, the label is after the nop (-1), for functions with return value, there is a mov instruction before the label (-4) */
+
+static void format_task_type(int type, char* buffer)
+{
+  char* progress = buffer;
+  if(type & ompt_task_initial) progress += sprintf(progress, "ompt_task_initial");
+  if(type & ompt_task_implicit) progress += sprintf(progress, "ompt_task_implicit");
+  if(type & ompt_task_explicit) progress += sprintf(progress, "ompt_task_explicit");
+  if(type & ompt_task_target) progress += sprintf(progress, "ompt_task_target");
+  if(type & ompt_task_undeferred) progress += sprintf(progress, "|ompt_task_undeferred");
+  if(type & ompt_task_untied) progress += sprintf(progress, "|ompt_task_untied");
+  if(type & ompt_task_final) progress += sprintf(progress, "|ompt_task_final");
+  if(type & ompt_task_mergeable) progress += sprintf(progress, "|ompt_task_mergeable");
+  if(type & ompt_task_merged) progress += sprintf(progress, "|ompt_task_merged");
+}
+
+static void
+on_ompt_callback_mutex_acquire(
+  ompt_mutex_kind_t kind,
+  unsigned int hint,
+  unsigned int impl,
+  ompt_wait_id_t wait_id,
+  const void *codeptr_ra)
+{
+  switch(kind)
+  {
+    case ompt_mutex_lock:
+      printf("%" PRIu64 ": ompt_event_wait_lock: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    case ompt_mutex_nest_lock:
+      printf("%" PRIu64 ": ompt_event_wait_nest_lock: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    case ompt_mutex_critical:
+      printf("%" PRIu64 ": ompt_event_wait_critical: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    case ompt_mutex_atomic:
+      printf("%" PRIu64 ": ompt_event_wait_atomic: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    case ompt_mutex_ordered:
+      printf("%" PRIu64 ": ompt_event_wait_ordered: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    default:
+      break;
+  }
+}
+
+static void
+on_ompt_callback_mutex_acquired(
+  ompt_mutex_kind_t kind,
+  ompt_wait_id_t wait_id,
+  const void *codeptr_ra)
+{
+  switch(kind)
+  {
+    case ompt_mutex_lock:
+      printf("%" PRIu64 ": ompt_event_acquired_lock: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_nest_lock:
+      printf("%" PRIu64 ": ompt_event_acquired_nest_lock_first: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_critical:
+      printf("%" PRIu64 ": ompt_event_acquired_critical: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_atomic:
+      printf("%" PRIu64 ": ompt_event_acquired_atomic: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_ordered:
+      printf("%" PRIu64 ": ompt_event_acquired_ordered: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    default:
+      break;
+  }
+}
+
+static void
+on_ompt_callback_mutex_released(
+  ompt_mutex_kind_t kind,
+  ompt_wait_id_t wait_id,
+  const void *codeptr_ra)
+{
+  switch(kind)
+  {
+    case ompt_mutex_lock:
+      printf("%" PRIu64 ": ompt_event_release_lock: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_nest_lock:
+      printf("%" PRIu64 ": ompt_event_release_nest_lock_last: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_critical:
+      printf("%" PRIu64 ": ompt_event_release_critical: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_atomic:
+      printf("%" PRIu64 ": ompt_event_release_atomic: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_ordered:
+      printf("%" PRIu64 ": ompt_event_release_ordered: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    default:
+      break;
+  }
+}
+
+static void
+on_ompt_callback_nest_lock(
+    ompt_scope_endpoint_t endpoint,
+    ompt_wait_id_t wait_id,
+    const void *codeptr_ra)
+{
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      printf("%" PRIu64 ": ompt_event_acquired_nest_lock_next: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_scope_end:
+      printf("%" PRIu64 ": ompt_event_release_nest_lock_prev: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+  }
+}
+
+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:
+      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);
+          print_ids(0);
+          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;
+  }
+}
+
+static void
+on_ompt_callback_flush(
+    ompt_data_t *thread_data,
+    const void *codeptr_ra)
+{
+  printf("%" PRIu64 ": ompt_event_flush: codeptr_ra=%p\n", thread_data->value, codeptr_ra);
+}
+
+static void
+on_ompt_callback_cancel(
+    ompt_data_t *task_data,
+    int flags,
+    const void *codeptr_ra)
+{
+  const char* first_flag_value;
+  const char* second_flag_value;
+  if(flags & ompt_cancel_parallel)
+    first_flag_value = ompt_cancel_flag_t_values[0];
+  else if(flags & ompt_cancel_sections)
+    first_flag_value = ompt_cancel_flag_t_values[1];
+  else if(flags & ompt_cancel_do)
+    first_flag_value = ompt_cancel_flag_t_values[2];
+  else if(flags & ompt_cancel_taskgroup)
+    first_flag_value = ompt_cancel_flag_t_values[3];
+
+  if(flags & ompt_cancel_activated)
+    second_flag_value = ompt_cancel_flag_t_values[4];
+  else if(flags & ompt_cancel_detected)
+    second_flag_value = ompt_cancel_flag_t_values[5];
+  else if(flags & ompt_cancel_discarded_task)
+    second_flag_value = ompt_cancel_flag_t_values[6];
+    
+  printf("%" PRIu64 ": ompt_event_cancel: task_data=%" PRIu64 ", flags=%s|%s=%" PRIu32 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, task_data->value, first_flag_value, second_flag_value, flags,  codeptr_ra);
+}
 
 static void
-on_ompt_event_barrier_begin(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t task_id)
+on_ompt_callback_idle(
+  ompt_scope_endpoint_t endpoint)
 {
-  printf("%" PRIu64 ": ompt_event_barrier_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
-  print_ids(0);
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      printf("%" PRIu64 ": ompt_event_idle_begin:\n", ompt_get_thread_data()->value);
+      break;
+    case ompt_scope_end:
+      printf("%" PRIu64 ": ompt_event_idle_end:\n", ompt_get_thread_data()->value);
+      break;
+  }
 }
 
 static void
-on_ompt_event_barrier_end(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t task_id)
+on_ompt_callback_implicit_task(
+    ompt_scope_endpoint_t endpoint,
+    ompt_data_t *parallel_data,
+    ompt_data_t *task_data,
+    unsigned int team_size,
+    unsigned int thread_num)
 {
-  printf("%" PRIu64 ": ompt_event_barrier_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      if(task_data->ptr)
+        printf("%s\n", "0: task_data initially not null");
+      task_data->value = ompt_get_unique_id();
+      printf("%" PRIu64 ": ompt_event_implicit_task_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", team_size=%" PRIu32 ", thread_num=%" PRIu32 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, team_size, thread_num);
+      break;
+    case ompt_scope_end:
+      printf("%" PRIu64 ": ompt_event_implicit_task_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", team_size=%" PRIu32 ", thread_num=%" PRIu32 "\n", ompt_get_thread_data()->value, (parallel_data)?parallel_data->value:0, task_data->value, team_size, thread_num);
+      break;
+  }
 }
 
 static void
-on_ompt_event_implicit_task_begin(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t task_id)
+on_ompt_callback_lock_init(
+  ompt_mutex_kind_t kind,
+  unsigned int hint,
+  unsigned int impl,
+  ompt_wait_id_t wait_id,
+  const void *codeptr_ra)
 {
-  printf("%" PRIu64 ": ompt_event_implicit_task_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+  switch(kind)
+  {
+    case ompt_mutex_lock:
+      printf("%" PRIu64 ": ompt_event_init_lock: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    case ompt_mutex_nest_lock:
+      printf("%" PRIu64 ": ompt_event_init_nest_lock: wait_id=%" PRIu64 ", hint=%" PRIu32 ", impl=%" PRIu32 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, hint, impl, codeptr_ra);
+      break;
+    default:
+      break;
+  }
 }
 
 static void
-on_ompt_event_implicit_task_end(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t task_id)
+on_ompt_callback_lock_destroy(
+  ompt_mutex_kind_t kind,
+  ompt_wait_id_t wait_id,
+  const void *codeptr_ra)
 {
-  printf("%" PRIu64 ": ompt_event_implicit_task_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+  switch(kind)
+  {
+    case ompt_mutex_lock:
+      printf("%" PRIu64 ": ompt_event_destroy_lock: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    case ompt_mutex_nest_lock:
+      printf("%" PRIu64 ": ompt_event_destroy_nest_lock: wait_id=%" PRIu64 ", codeptr_ra=%p \n", ompt_get_thread_data()->value, wait_id, codeptr_ra);
+      break;
+    default:
+      break;
+  }
 }
 
 static void
-on_ompt_event_loop_begin(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t parent_task_id,
-  void *workshare_function)
+on_ompt_callback_work(
+  ompt_work_type_t wstype,
+  ompt_scope_endpoint_t endpoint,
+  ompt_data_t *parallel_data,
+  ompt_data_t *task_data,
+  uint64_t count,
+  const void *codeptr_ra)
 {
-  printf("%" PRIu64 ": ompt_event_loop_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", workshare_function=%p\n", ompt_get_thread_id(), parallel_id, parent_task_id, workshare_function);
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      switch(wstype)
+      {
+        case ompt_work_loop:
+          printf("%" PRIu64 ": ompt_event_loop_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_sections:
+          printf("%" PRIu64 ": ompt_event_sections_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_single_executor:
+          printf("%" PRIu64 ": ompt_event_single_in_block_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_single_other:
+          printf("%" PRIu64 ": ompt_event_single_others_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_workshare:
+          //impl
+          break;
+        case ompt_work_distribute:
+          printf("%" PRIu64 ": ompt_event_distribute_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_taskloop:
+          //impl
+          printf("%" PRIu64 ": ompt_event_taskloop_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+      }
+      break;
+    case ompt_scope_end:
+      switch(wstype)
+      {
+        case ompt_work_loop:
+          printf("%" PRIu64 ": ompt_event_loop_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_sections:
+          printf("%" PRIu64 ": ompt_event_sections_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_single_executor:
+          printf("%" PRIu64 ": ompt_event_single_in_block_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_single_other:
+          printf("%" PRIu64 ": ompt_event_single_others_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_workshare:
+          //impl
+          break;
+        case ompt_work_distribute:
+          printf("%" PRIu64 ": ompt_event_distribute_end: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+        case ompt_work_taskloop:
+          //impl
+          printf("%" PRIu64 ": ompt_event_taskloop_end: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra, count);
+          break;
+      }
+      break;
+  }
 }
 
 static void
-on_ompt_event_loop_end(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t task_id)
+on_ompt_callback_master(
+  ompt_scope_endpoint_t endpoint,
+  ompt_data_t *parallel_data,
+  ompt_data_t *task_data,
+  const void *codeptr_ra)
 {
-  printf("%" PRIu64 ": ompt_event_loop_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+  switch(endpoint)
+  {
+    case ompt_scope_begin:
+      printf("%" PRIu64 ": ompt_event_master_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_scope_end:
+      printf("%" PRIu64 ": ompt_event_master_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, codeptr_ra);
+      break;
+  }
 }
 
 static void
-on_ompt_event_parallel_begin(
-  ompt_task_id_t parent_task_id,
-  ompt_frame_t *parent_task_frame,
-  ompt_parallel_id_t parallel_id,
+on_ompt_callback_parallel_begin(
+  ompt_data_t *parent_task_data,
+  const ompt_frame_t *parent_task_frame,
+  ompt_data_t* parallel_data,
   uint32_t requested_team_size,
-  void *parallel_function,
-  ompt_invoker_t invoker)
+  ompt_invoker_t invoker,
+  const void *codeptr_ra)
 {
-  printf("%" PRIu64 ": ompt_event_parallel_begin: parent_task_id=%" PRIu64 ", parent_task_frame.exit=%p, parent_task_frame.reenter=%p, parallel_id=%" PRIu64 ", requested_team_size=%" PRIu32 ", parallel_function=%p, invoker=%d\n", ompt_get_thread_id(), parent_task_id, parent_task_frame->exit_runtime_frame, parent_task_frame->reenter_runtime_frame, parallel_id, requested_team_size, parallel_function, invoker);
+  if(parallel_data->ptr)
+    printf("%s\n", "0: parallel_data initially not null");
+  parallel_data->value = ompt_get_unique_id();
+  printf("%" PRIu64 ": ompt_event_parallel_begin: parent_task_id=%" PRIu64 ", parent_task_frame.exit=%p, parent_task_frame.reenter=%p, parallel_id=%" PRIu64 ", requested_team_size=%" PRIu32 ", codeptr_ra=%p, invoker=%d\n", ompt_get_thread_data()->value, parent_task_data->value, parent_task_frame->exit_runtime_frame, parent_task_frame->reenter_runtime_frame, parallel_data->value, requested_team_size, codeptr_ra, invoker);
 }
 
 static void
-on_ompt_event_parallel_end(
-  ompt_parallel_id_t parallel_id,
-  ompt_task_id_t task_id,
-  ompt_invoker_t invoker)
+on_ompt_callback_parallel_end(
+  ompt_data_t *parallel_data,
+  ompt_data_t *task_data,
+  ompt_invoker_t invoker,
+  const void *codeptr_ra)
 {
-  printf("%" PRIu64 ": ompt_event_parallel_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", invoker=%d\n", ompt_get_thread_id(), parallel_id, task_id, invoker);
+  printf("%" PRIu64 ": ompt_event_parallel_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", invoker=%d, codeptr_ra=%p\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, invoker, codeptr_ra);
 }
 
+static void
+on_ompt_callback_task_create(
+    ompt_data_t *parent_task_data,     /* id of parent task            */
+    const ompt_frame_t *parent_frame,  /* frame data for parent task   */
+    ompt_data_t* new_task_data,        /* id of created task           */
+    int type,
+    int has_dependences,
+    const void *codeptr_ra)               /* pointer to outlined function */
+{
+  if(new_task_data->ptr)
+    printf("%s\n", "0: new_task_data initially not null");
+  new_task_data->value = ompt_get_unique_id();
+  char buffer[2048];
+
+  format_task_type(type, buffer);
 
-void ompt_initialize(
+  //there is no paralllel_begin callback for implicit parallel region
+  //thus it is initialized in initial task
+  if(type & ompt_task_initial)
+  {
+    ompt_data_t *parallel_data;
+    ompt_get_parallel_info(0, &parallel_data, NULL);
+    if(parallel_data->ptr)
+      printf("%s\n", "0: parallel_data initially not null");
+    parallel_data->value = ompt_get_unique_id();
+  }
+
+  printf("%" PRIu64 ": ompt_event_task_create: parent_task_id=%" PRIu64 ", parent_task_frame.exit=%p, parent_task_frame.reenter=%p, new_task_id=%" PRIu64 ", codeptr_ra=%p, task_type=%s=%d, has_dependences=%s\n", ompt_get_thread_data()->value, parent_task_data ? parent_task_data->value : 0, parent_frame ? parent_frame->exit_runtime_frame : NULL, parent_frame ? parent_frame->reenter_runtime_frame : NULL, new_task_data->value, codeptr_ra, buffer, type, has_dependences ? "yes" : "no");
+}
+
+static void
+on_ompt_callback_task_schedule(
+    ompt_data_t *first_task_data,
+    ompt_task_status_t prior_task_status,
+    ompt_data_t *second_task_data)
+{
+  printf("%" PRIu64 ": ompt_event_task_schedule: first_task_id=%" PRIu64 ", second_task_id=%" PRIu64 ", prior_task_status=%s=%d\n", ompt_get_thread_data()->value, first_task_data->value, second_task_data->value, ompt_task_status_t_values[prior_task_status], prior_task_status);
+  if(prior_task_status == ompt_task_complete)
+  {
+    printf("%" PRIu64 ": ompt_event_task_end: task_id=%" PRIu64 "\n", ompt_get_thread_data()->value, first_task_data->value);
+  }
+}
+
+static void
+on_ompt_callback_task_dependences(
+  ompt_data_t *task_data,
+  const ompt_task_dependence_t *deps,
+  int ndeps)
+{
+  printf("%" PRIu64 ": ompt_event_task_dependences: task_id=%" PRIu64 ", deps=%p, ndeps=%d\n", ompt_get_thread_data()->value, task_data->value, (void *)deps, ndeps);
+}
+
+static void
+on_ompt_callback_task_dependence(
+  ompt_data_t *first_task_data,
+  ompt_data_t *second_task_data)
+{
+  printf("%" PRIu64 ": ompt_event_task_dependence_pair: first_task_id=%" PRIu64 ", second_task_id=%" PRIu64 "\n", ompt_get_thread_data()->value, first_task_data->value, second_task_data->value);
+}
+
+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_thread_end(
+  ompt_data_t *thread_data)
+{
+  printf("%" PRIu64 ": ompt_event_thread_end: thread_id=%" PRIu64 "\n", ompt_get_thread_data()->value, thread_data->value);
+}
+
+static int
+on_ompt_callback_control_tool(
+  uint64_t command,
+  uint64_t modifier,
+  void *arg,
+  const void *codeptr_ra)
+{
+  ompt_frame_t* omptTaskFrame;
+  ompt_get_task_info(0, NULL, (ompt_data_t**) NULL, &omptTaskFrame, NULL, NULL);
+  printf("%" PRIu64 ": ompt_event_control_tool: command=%" PRIu64 ", modifier=%" PRIu64 ", arg=%p, codeptr_ra=%p, current_task_frame.exit=%p, current_task_frame.reenter=%p \n", ompt_get_thread_data()->value, command, modifier, arg, codeptr_ra, omptTaskFrame->exit_runtime_frame, omptTaskFrame->reenter_runtime_frame);
+  return 0; //success
+}
+
+#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,
-  const char *runtime_version,
-  unsigned int ompt_version)
+  ompt_fns_t* fns)
 {
-  ompt_set_callback_t ompt_set_callback = (ompt_set_callback_t) lookup("ompt_set_callback");
-  ompt_get_task_id = (ompt_get_task_id_t) lookup("ompt_get_task_id");
-  ompt_get_task_frame = (ompt_get_task_frame_t) lookup("ompt_get_task_frame");
-  ompt_get_thread_id = (ompt_get_thread_id_t) lookup("ompt_get_thread_id");
-  ompt_get_parallel_id = (ompt_get_parallel_id_t) lookup("ompt_get_parallel_id");
+  ompt_set_callback = (ompt_set_callback_t) lookup("ompt_set_callback");
+  ompt_get_task_info = (ompt_get_task_info_t) lookup("ompt_get_task_info");
+  ompt_get_thread_data = (ompt_get_thread_data_t) lookup("ompt_get_thread_data");
+  ompt_get_parallel_info = (ompt_get_parallel_info_t) lookup("ompt_get_parallel_info");
+  ompt_get_unique_id = (ompt_get_unique_id_t) lookup("ompt_get_unique_id");
 
-  ompt_set_callback(ompt_event_barrier_begin, (ompt_callback_t) &on_ompt_event_barrier_begin);
-  ompt_set_callback(ompt_event_barrier_end, (ompt_callback_t) &on_ompt_event_barrier_end);
-  ompt_set_callback(ompt_event_implicit_task_begin, (ompt_callback_t) &on_ompt_event_implicit_task_begin);
-  ompt_set_callback(ompt_event_implicit_task_end, (ompt_callback_t) &on_ompt_event_implicit_task_end);
-  ompt_set_callback(ompt_event_loop_begin, (ompt_callback_t) &on_ompt_event_loop_begin);
-  ompt_set_callback(ompt_event_loop_end, (ompt_callback_t) &on_ompt_event_loop_end);
-  ompt_set_callback(ompt_event_parallel_begin, (ompt_callback_t) &on_ompt_event_parallel_begin);
-  ompt_set_callback(ompt_event_parallel_end, (ompt_callback_t) &on_ompt_event_parallel_end);
-  printf("0: NULL_POINTER=%p\n", NULL);
+  ompt_get_num_places = (ompt_get_num_places_t) lookup("ompt_get_num_places");
+  ompt_get_place_proc_ids = (ompt_get_place_proc_ids_t) lookup("ompt_get_place_proc_ids");
+  ompt_get_place_num = (ompt_get_place_num_t) lookup("ompt_get_place_num");
+  ompt_get_partition_place_nums = (ompt_get_partition_place_nums_t) lookup("ompt_get_partition_place_nums");
+  ompt_get_proc_id = (ompt_get_proc_id_t) lookup("ompt_get_proc_id");
+  ompt_enumerate_states = (ompt_enumerate_states_t) lookup("ompt_enumerate_states");
+  ompt_enumerate_mutex_impls = (ompt_enumerate_mutex_impls_t) lookup("ompt_enumerate_mutex_impls");
+
+  register_callback(ompt_callback_mutex_acquire);
+  register_callback_t(ompt_callback_mutex_acquired, ompt_callback_mutex_t);
+  register_callback_t(ompt_callback_mutex_released, ompt_callback_mutex_t);
+  register_callback(ompt_callback_nest_lock);
+  register_callback(ompt_callback_sync_region);
+  register_callback_t(ompt_callback_sync_region_wait, ompt_callback_sync_region_t);
+  register_callback(ompt_callback_control_tool);
+  register_callback(ompt_callback_flush);
+  register_callback(ompt_callback_cancel);
+  register_callback(ompt_callback_idle);
+  register_callback(ompt_callback_implicit_task);
+  register_callback_t(ompt_callback_lock_init, ompt_callback_mutex_acquire_t);
+  register_callback_t(ompt_callback_lock_destroy, ompt_callback_mutex_t);
+  register_callback(ompt_callback_work);
+  register_callback(ompt_callback_master);
+  register_callback(ompt_callback_parallel_begin);
+  register_callback(ompt_callback_parallel_end);
+  register_callback(ompt_callback_task_create);
+  register_callback(ompt_callback_task_schedule);
+  register_callback(ompt_callback_task_dependences);
+  register_callback(ompt_callback_task_dependence);
+  register_callback(ompt_callback_thread_begin);
+  register_callback(ompt_callback_thread_end);
+  printf("0: NULL_POINTER=%p\n", (void*)NULL);
+  return 1; //success
+}
+
+void ompt_finalize(ompt_fns_t* fns)
+{
+  printf("0: ompt_event_runtime_shutdown\n");
 }
 
-ompt_initialize_t ompt_tool()
+ompt_fns_t* ompt_start_tool(
+  unsigned int omp_version,
+  const char *runtime_version)
 {
-  return &ompt_initialize;
+  static ompt_fns_t ompt_fns = {&ompt_initialize,&ompt_finalize};
+  return &ompt_fns;
 }
diff --git a/openmp/runtime/test/ompt/cancel/cancel_parallel.c b/openmp/runtime/test/ompt/cancel/cancel_parallel.c
new file mode 100644
index 00000000000..a37480107e7
--- /dev/null
+++ b/openmp/runtime/test/ompt/cancel/cancel_parallel.c
@@ -0,0 +1,42 @@
+// RUN: %libomp-compile && env OMP_CANCELLATION=true %libomp-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// Current GOMP interface implementation does not support cancellation
+// XFAIL: gcc
+
+#include "callback.h"
+#include "omp.h"
+
+int main()
+{
+  #pragma omp parallel num_threads(2)
+  {
+    if(omp_get_thread_num() == 0)
+    {
+      printf("%" PRIu64 ": fuzzy_address=0x%lx or 0x%lx\n", ompt_get_thread_data()->value, ((uint64_t)(char*)(&& ompt_label_1))/256-1, ((uint64_t)(char*)(&& ompt_label_1))/256);
+      #pragma omp cancel parallel
+      print_fuzzy_address(1); //does not actually print the address but provides a label
+    }
+    else
+    {
+      delay(100);
+      printf("%" PRIu64 ": fuzzy_address=0x%lx or 0x%lx\n", ompt_get_thread_data()->value, ((uint64_t)(char*)(&& ompt_label_2))/256-1, ((uint64_t)(char*)(&& ompt_label_2))/256);
+      #pragma omp cancellation point parallel
+      print_fuzzy_address(2); //does not actually print the address but provides a label
+    }
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_cancel'
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[NULL]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[NULL]], task_type=ompt_task_initial=1, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_parallel|ompt_cancel_activated=17, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: fuzzy_address={{.*}}[[OTHER_RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_parallel|ompt_cancel_detected=33, codeptr_ra=[[OTHER_RETURN_ADDRESS]]{{[0-f][0-f]}}
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/cancel/cancel_taskgroup.c b/openmp/runtime/test/ompt/cancel/cancel_taskgroup.c
new file mode 100644
index 00000000000..28558f85fb3
--- /dev/null
+++ b/openmp/runtime/test/ompt/cancel/cancel_taskgroup.c
@@ -0,0 +1,88 @@
+// RUN:  %libomp-compile && env OMP_CANCELLATION=true %libomp-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// Current GOMP interface implementation does not support cancellation
+// XFAIL: gcc
+
+#include "callback.h"
+#include <unistd.h>  
+#include <stdio.h>
+
+int main()
+{
+  int condition=0;
+  #pragma omp parallel num_threads(2)
+  {}
+
+  print_frame(0);
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {
+      #pragma omp taskgroup
+      {
+        #pragma omp task shared(condition)
+        {
+          printf("start execute task 1\n");
+          OMPT_SIGNAL(condition);
+          OMPT_WAIT(condition,2);
+          #pragma omp cancellation point taskgroup
+          printf("end execute task 1\n");
+        }
+        #pragma omp task shared(condition)
+        {
+          printf("start execute task 2\n");
+          OMPT_SIGNAL(condition);
+          OMPT_WAIT(condition,2);
+          #pragma omp cancellation point taskgroup
+          printf("end execute task 2\n");
+        }
+      #pragma omp task shared(condition)
+        {
+          printf("start execute task 3\n");
+          OMPT_SIGNAL(condition);
+          OMPT_WAIT(condition,2);
+          #pragma omp cancellation point taskgroup
+          printf("end execute task 3\n");
+        }
+      #pragma omp task if(0) shared(condition)
+        {
+          printf("start execute task 4\n");
+          OMPT_WAIT(condition,1);
+          #pragma omp cancel taskgroup
+          printf("end execute task 4\n");
+        }
+        OMPT_SIGNAL(condition);
+      }
+    }
+    #pragma omp barrier
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_master'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_cancel'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_thread_begin'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_master_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID]], parent_task_frame.exit={{0x[0-f]*}}, parent_task_frame.reenter={{0x[0-f]*}}, new_task_id=[[FIRST_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}, task_type=ompt_task_explicit=4, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID]], parent_task_frame.exit={{0x[0-f]*}}, parent_task_frame.reenter={{0x[0-f]*}}, new_task_id=[[SECOND_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}, task_type=ompt_task_explicit=4, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID]], parent_task_frame.exit={{0x[0-f]*}}, parent_task_frame.reenter={{0x[0-f]*}}, new_task_id=[[THIRD_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}, task_type=ompt_task_explicit=4, has_dependences=no
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID]], parent_task_frame.exit={{0x[0-f]*}}, parent_task_frame.reenter={{0x[0-f]*}}, new_task_id=[[CANCEL_TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]*}}, task_type=ompt_task_explicit|ompt_task_undeferred=134217732, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[PARENT_TASK_ID]], second_task_id=[[CANCEL_TASK_ID]], prior_task_status=ompt_task_others=4
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_cancel: task_data=[[CANCEL_TASK_ID]], flags=ompt_cancel_taskgroup|ompt_cancel_activated=24, codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[CANCEL_TASK_ID]], second_task_id=[[PARENT_TASK_ID]], prior_task_status=ompt_task_cancel=3
+
+  // CHECK-DAG: {{^}}{{[0-9]+}}: ompt_event_cancel: task_data={{[0-9]+}}, flags=ompt_cancel_taskgroup|ompt_cancel_discarded_task=72, codeptr_ra=[[NULL]]
+  // CHECK-DAG: {{^}}{{[0-9]+}}: ompt_event_cancel: task_data={{[0-9]+}}, flags=ompt_cancel_taskgroup|ompt_cancel_discarded_task=72, codeptr_ra=[[NULL]]
+  
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_thread_begin: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_cancel: task_data={{[0-9]+}}, flags=ompt_cancel_taskgroup|ompt_cancel_detected=40, codeptr_ra={{0x[0-f]*}}
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/cancel/cancel_worksharing.c b/openmp/runtime/test/ompt/cancel/cancel_worksharing.c
new file mode 100644
index 00000000000..41073362a3a
--- /dev/null
+++ b/openmp/runtime/test/ompt/cancel/cancel_worksharing.c
@@ -0,0 +1,68 @@
+// RUN: %libomp-compile && env OMP_CANCELLATION=true %libomp-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// Current GOMP interface implementation does not support cancellation
+// XFAIL: gcc
+
+
+#include "callback.h"
+#include <unistd.h>
+
+int main()
+{
+  int condition=0;
+  #pragma omp parallel num_threads(2)
+  {
+    int x = 0;
+    int i;
+    #pragma omp for
+    for(i = 0; i < 2; i++)
+    {
+      if(i == 0)
+      {
+        x++;
+        OMPT_SIGNAL(condition);
+        #pragma omp cancel for
+      }
+      else
+      {
+        x++;
+        OMPT_WAIT(condition,1);
+        delay(10000);
+        #pragma omp cancellation point for
+      }
+    }
+  }
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp sections
+    {
+      #pragma omp section
+      {
+        OMPT_SIGNAL(condition);
+        #pragma omp cancel sections
+      }
+      #pragma omp section
+      {
+        OMPT_WAIT(condition,2);
+        delay(10000);
+        #pragma omp cancellation point sections
+      }
+    }
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_cancel'
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[NULL]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[NULL]], task_type=ompt_task_initial=1, has_dependences=no
+  
+  // cancel for and sections
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_do|ompt_cancel_activated=20, codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_sections|ompt_cancel_activated=18, codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[OTHER_THREAD_ID:[0-9]+]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_do|ompt_cancel_detected=36, codeptr_ra={{0x[0-f]*}}
+  // CHECK: {{^}}[[OTHER_THREAD_ID:[0-9]+]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_sections|ompt_cancel_detected=34, codeptr_ra={{0x[0-f]*}}
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/loadtool/tool_available.c b/openmp/runtime/test/ompt/loadtool/tool_available.c
new file mode 100644
index 00000000000..0205193176e
--- /dev/null
+++ b/openmp/runtime/test/ompt/loadtool/tool_available.c
@@ -0,0 +1,59 @@
+// RUN: %libomp-compile -DCODE && %libomp-compile -DTOOL -o%T/tool.so -shared -fPIC && env OMP_TOOL_LIBRARIES=%T/tool.so %libomp-run | FileCheck %s
+// REQUIRES: ompt
+
+/*
+ *  This file contains code for an OMPT shared library tool to be 
+ *  loaded and the code for the OpenMP executable. 
+ *  -DTOOL enables the code for the tool during compilation
+ *  -DCODE enables the code for the executable during compilation
+ *  The RUN line compiles the two binaries and then tries to load
+ *  the tool using the OMP_TOOL_LIBRARIES environmental variable.
+ */
+
+#ifdef CODE
+#include "omp.h"
+
+int main()
+{
+  #pragma omp parallel num_threads(2)
+  {
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 
+  
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}0: ompt_event_runtime_shutdown
+
+  return 0;
+}
+
+#endif /* CODE */
+
+#ifdef TOOL
+
+#include <stdio.h>
+#include <ompt.h>
+
+int ompt_initialize(
+  ompt_function_lookup_t lookup,
+  ompt_fns_t* fns)
+{
+  printf("0: NULL_POINTER=%p\n", (void*)NULL);
+  return 1; //success
+}
+
+void ompt_finalize(ompt_fns_t* fns)
+{
+  printf("%d: ompt_event_runtime_shutdown\n", omp_get_thread_num());
+}
+
+ompt_fns_t* ompt_start_tool(
+  unsigned int omp_version,
+  const char *runtime_version)
+{
+  static ompt_fns_t ompt_fns = {&ompt_initialize,&ompt_finalize};
+  return &ompt_fns;
+}
+#endif /* TOOL */
diff --git a/openmp/runtime/test/ompt/misc/api_calls.c b/openmp/runtime/test/ompt/misc/api_calls.c
new file mode 100644
index 00000000000..f7f1e16aa6d
--- /dev/null
+++ b/openmp/runtime/test/ompt/misc/api_calls.c
@@ -0,0 +1,66 @@
+// RUN: %libomp-compile && env OMP_PLACES=cores %libomp-run | FileCheck %s
+// REQUIRES: ompt, linux
+#include "callback.h"
+#include <omp.h>
+#define __USE_GNU
+#include <sched.h>
+#undef __USE_GNU
+
+void print_list(char* function_name, int list[])
+{
+  printf("%" PRIu64 ": %s(0)=(%d", ompt_get_thread_data()->value, function_name, list[0]);
+  int i;
+  for(i = 1; i < omp_get_place_num_procs(0); i++)
+  {
+    printf(",%d", list[i]);
+  }
+  printf(")\n");
+}
+
+int main()
+{
+  #pragma omp parallel num_threads(1)
+  {
+  	printf("%" PRIu64 ": omp_get_num_places()=%d\n", ompt_get_thread_data()->value, omp_get_num_places());
+  	printf("%" PRIu64 ": ompt_get_num_places()=%d\n", ompt_get_thread_data()->value, ompt_get_num_places());
+
+    int omp_ids[omp_get_place_num_procs(0)];
+    omp_get_place_proc_ids(0, omp_ids);
+    print_list("omp_get_place_proc_ids" ,omp_ids);
+    int ompt_ids[omp_get_place_num_procs(0)];
+    ompt_get_place_proc_ids(0, omp_get_place_num_procs(0), ompt_ids);
+    print_list("ompt_get_place_proc_ids", ompt_ids);
+
+  	printf("%" PRIu64 ": omp_get_place_num()=%d\n", ompt_get_thread_data()->value, omp_get_place_num());
+  	printf("%" PRIu64 ": ompt_get_place_num()=%d\n", ompt_get_thread_data()->value, ompt_get_place_num());
+
+    int omp_nums[omp_get_partition_num_places()];
+    omp_get_partition_place_nums(omp_nums);
+    print_list("omp_get_partition_place_nums" ,omp_nums);
+    int ompt_nums[omp_get_partition_num_places()];
+    ompt_get_partition_place_nums(omp_get_partition_num_places(), ompt_nums);
+    print_list("ompt_get_partition_place_nums", ompt_nums);
+
+    printf("%" PRIu64 ": sched_getcpu()=%d\n", ompt_get_thread_data()->value, sched_getcpu());
+    printf("%" PRIu64 ": ompt_get_proc_id()=%d\n", ompt_get_thread_data()->value, ompt_get_proc_id());
+  }
+
+  // Check if libomp supports the callbacks for this test.
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: omp_get_num_places()=[[NUM_PLACES:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_get_num_places()=[[NUM_PLACES]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: omp_get_place_proc_ids(0)=([[PROC_IDS:[0-9\,]+]])
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_get_place_proc_ids(0)=([[PROC_IDS]])
+
+  // CHECK: {{^}}[[MASTER_ID]]: omp_get_place_num()=[[PLACE_NUM:[-]?[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_get_place_num()=[[PLACE_NUM]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: sched_getcpu()=[[CPU_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_get_proc_id()=[[CPU_ID]]
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/misc/control_tool.c b/openmp/runtime/test/ompt/misc/control_tool.c
new file mode 100644
index 00000000000..7fc4f023a36
--- /dev/null
+++ b/openmp/runtime/test/ompt/misc/control_tool.c
@@ -0,0 +1,27 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  #pragma omp parallel num_threads(1)
+  {
+    print_frame(1);
+    print_frame(0);
+    omp_control_tool(omp_control_tool_flush, 1, NULL);
+    print_current_address(0);
+  }
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_control_tool'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address(1)=[[EXIT_FRAME:0x[0-f]*]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER_FRAME:0x[0-f]*]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_control_tool: command=3, modifier=1, arg=[[NULL]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]*]], current_task_frame.exit=[[EXIT_FRAME]], current_task_frame.reenter=[[REENTER_FRAME]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/misc/control_tool_no_ompt_support.c b/openmp/runtime/test/ompt/misc/control_tool_no_ompt_support.c
new file mode 100644
index 00000000000..ee64da04f0d
--- /dev/null
+++ b/openmp/runtime/test/ompt/misc/control_tool_no_ompt_support.c
@@ -0,0 +1,12 @@
+// RUN: %libomp-compile-and-run
+#include <omp.h>
+
+int main()
+{
+  #pragma omp parallel num_threads(1)
+  {
+    omp_control_tool(omp_control_tool_flush, 1, NULL);
+  }
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/misc/idle.c b/openmp/runtime/test/ompt/misc/idle.c
new file mode 100644
index 00000000000..7413c32fe12
--- /dev/null
+++ b/openmp/runtime/test/ompt/misc/idle.c
@@ -0,0 +1,32 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+  #pragma omp parallel num_threads(3)
+  {
+    #pragma omp atomic
+    x++;
+  }
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp atomic
+    x++;
+  }
+
+
+  printf("x=%d\n", x);
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_idle'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_idle_begin:
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_idle_end:
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/ompt-signal.h b/openmp/runtime/test/ompt/ompt-signal.h
new file mode 100644
index 00000000000..b5c28cf5dee
--- /dev/null
+++ b/openmp/runtime/test/ompt/ompt-signal.h
@@ -0,0 +1,31 @@
+#if defined(WIN32) || defined(_WIN32)
+#include <windows.h>
+#define delay() Sleep(1);
+#else
+#include <unistd.h>
+#define delay(t) usleep(t);
+#endif
+
+// These functions are used to provide a signal-wait mechanism to enforce expected scheduling for the test cases.
+// Conditional variable (s) needs to be shared! Initialize to 0
+
+#define OMPT_SIGNAL(s) ompt_signal(&s)
+//inline 
+void ompt_signal(int* s) 
+{                
+  #pragma omp atomic
+  (*s)++;
+}
+                
+#define OMPT_WAIT(s,v) ompt_wait(&s,v)
+// wait for s >= v
+//inline 
+void ompt_wait(int *s, int v)
+{
+  int wait=0;
+  do{
+    delay(10);
+    #pragma omp atomic read
+	  wait = (*s);
+  }while(wait<v);
+}
diff --git a/openmp/runtime/test/ompt/parallel/dynamic_enough_threads.c b/openmp/runtime/test/ompt/parallel/dynamic_enough_threads.c
new file mode 100644
index 00000000000..4c340ba7b89
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/dynamic_enough_threads.c
@@ -0,0 +1,43 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+
+int main()
+{
+  omp_set_dynamic(1);
+
+  #pragma omp parallel num_threads(4)
+  {
+    print_ids(0);
+    print_ids(1);
+  }
+  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'
+
+  //team-size of 1-4 is expected
+
+  // 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: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]], team_size={{[1-4]}}
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/dynamic_not_enough_threads.c b/openmp/runtime/test/ompt/parallel/dynamic_not_enough_threads.c
new file mode 100644
index 00000000000..f3a6e170bea
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/dynamic_not_enough_threads.c
@@ -0,0 +1,43 @@
+// RUN: %libomp-compile && env OMP_THREAD_LIMIT=2 %libomp-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+
+int main()
+{
+  omp_set_dynamic(1);
+
+  #pragma omp parallel num_threads(4)
+  {
+    print_ids(0);
+    print_ids(1);
+  }
+  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'
+
+  //team-size of 1-4 is expected
+
+  // 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: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]], team_size={{[1-4]}}
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/max_active_levels_serialized.c b/openmp/runtime/test/ompt/parallel/max_active_levels_serialized.c
new file mode 100644
index 00000000000..6b45354acd8
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/max_active_levels_serialized.c
@@ -0,0 +1,72 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  omp_set_nested(1);
+  omp_set_max_active_levels(1);
+
+  #pragma omp parallel num_threads(2)
+  {
+    print_ids(0);
+    print_ids(1);
+    #pragma omp parallel num_threads(2)
+    {
+      print_ids(0);
+      print_ids(1);
+      print_ids(2);
+    }
+    print_fuzzy_address(1);
+  }
+  print_fuzzy_address(2);
+
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+  // THREADS: 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
+
+  // THREADS: {{^}}[[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=2, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+  
+  // 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: {{^}}[[MASTER_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=2, codeptr_ra=[[NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][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]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: 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: {{^}}[[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=2, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/nested.c b/openmp/runtime/test/ompt/parallel/nested.c
index 1c7b68d1e9c..d128e8021c6 100644
--- a/openmp/runtime/test/ompt/parallel/nested.c
+++ b/openmp/runtime/test/ompt/parallel/nested.c
@@ -3,9 +3,11 @@
 // REQUIRES: ompt
 #include "callback.h"
 #include <omp.h>
+#include <unistd.h>
 
 int main()
 {
+  int condition=0;
   omp_set_nested(1);
   print_frame(0);
 
@@ -15,6 +17,10 @@ int main()
     print_ids(0);
     print_ids(1);
     print_frame(0);
+
+    //get all implicit task events before starting nested:
+    #pragma omp barrier
+    
     #pragma omp parallel num_threads(4)
     {
       print_frame(1);
@@ -22,17 +28,38 @@ int main()
       print_ids(1);
       print_ids(2);
       print_frame(0);
+      OMPT_SIGNAL(condition);
+      OMPT_WAIT(condition,16);
       #pragma omp barrier
+      print_fuzzy_address(1);
       print_ids(0);
     }
+    print_fuzzy_address(2);
     print_ids(0);
   }
+  print_fuzzy_address(3);
+
+
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
 
   // CHECK: 0: NULL_POINTER=[[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, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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!
 
@@ -46,219 +73,224 @@ int main()
   // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
 
 
   // THREADS: {{^}}0: NULL_POINTER=[[NULL:.*$]]
   // THREADS: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address(0)=[[MAIN_REENTER:0x[0-f]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], 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]+]]
 
   // nested parallel masters
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
   // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
   // THREADS: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
-  // 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_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, codeptr_ra=[[NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][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]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_EXIT]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], 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_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], codeptr_ra=[[BARRIER_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // THREADS: {{^}}[[MASTER_ID]]: task 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]]
-  // 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]]: fuzzy_address={{.*}}[[BARRIER_RETURN_ADDRESS]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_EXIT]], reenter_frame=[[NULL]]
   // 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: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // THREADS: {{^}}[[MASTER_ID]]: task 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={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]], codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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]], codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
   // THREADS-NOT: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
   // 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: {{^}}[[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]]: task 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={{[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: {{^}}[[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]}}
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
 
   // 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]]: 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: {{^}}[[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, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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]]: 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: {{^}}[[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, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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]]: 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: {{^}}[[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, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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]]
 
   // nested parallel worker threads
   // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: 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=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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]]
 
   return 0;
 }
diff --git a/openmp/runtime/test/ompt/parallel/nested_lwt.c b/openmp/runtime/test/ompt/parallel/nested_lwt.c
index 6128816ac12..b3327c2de66 100644
--- a/openmp/runtime/test/ompt/parallel/nested_lwt.c
+++ b/openmp/runtime/test/ompt/parallel/nested_lwt.c
@@ -3,35 +3,59 @@
 // REQUIRES: ompt
 #include "callback.h"
 #include <omp.h>
+#include <unistd.h>
 
 int main()
 {
   omp_set_nested(1);
+  int condition;
 
   #pragma omp parallel num_threads(4)
   {
     print_ids(0);
     print_ids(1);
+    //get all implicit task events before starting nested:
+    #pragma omp barrier
     #pragma omp parallel num_threads(1)
     {
       print_ids(0);
       print_ids(1);
       print_ids(2);
+      //get all implicit task events before starting nested:
+      #pragma omp barrier
       #pragma omp parallel num_threads(4)
       {
         print_ids(0);
         print_ids(1);
         print_ids(2);
         print_ids(3);
+      OMPT_SIGNAL(condition);
+      OMPT_WAIT(condition,16);
       }
+      print_fuzzy_address(1);
     }
+    print_fuzzy_address(2);
   }
+  print_fuzzy_address(3);
+
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
 
   // CHECK: 0: NULL_POINTER=[[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, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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=0x{{[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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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!
 
@@ -48,251 +72,261 @@ int main()
 
 
   // THREADS: 0: NULL_POINTER=[[NULL:.*$]]
-  // THREADS: {{^}}[[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=[[TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+  // THREADS: {{^}}[[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=[[TASK_FRAME_ENTER: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]+]]
 
   // nested parallel masters
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
 
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, parallel_function=[[NESTED_PARALLEL_FUNCTION:0x[0-f]+]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, codeptr_ra=[[NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][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]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
 
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_NESTED_PARALLEL_FUNCTION:0x[0-f]+]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[NESTED_NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[NESTED_NESTED_RETURN_ADDRESS]]
   // THREADS-NOT: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
   // 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]]
-  // 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: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
 
   // 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]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[NESTED_NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_NESTED_RETURN_ADDRESS]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[NESTED_NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_NESTED_RETURN_ADDRESS]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=1, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit=[[NESTED_NESTED_TASK_FRAME_EXIT]], parent_task_frame.reenter=[[NESTED_NESTED_TASK_FRAME_ENTER:0x[0-f]+]], parallel_id=[[NESTED_NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=[[NESTED_NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_NESTED_TASK_FRAME_EXIT:0x[0-f]+]], reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame=[[NESTED_NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NESTED_TASK_FRAME_EXIT]], reenter_frame=[[NESTED_TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_NESTED_RETURN_ADDRESS]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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]]
 
   // nested parallel worker threads
   // THREADS: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
   // can't reliably tell which parallel region is the parent...
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
-  // THREADS: {{^}}[[THREAD_ID]]: level 3: parallel_id=0, task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}
+  // THREADS: {{^}}[[THREAD_ID]]: task level 3: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[TASK_FRAME_ENTER]]
   // THREADS-NOT: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_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]]
 
 
   return 0;
diff --git a/openmp/runtime/test/ompt/parallel/nested_serialized.c b/openmp/runtime/test/ompt/parallel/nested_serialized.c
index b43323f89f5..e55eb2e8b33 100644
--- a/openmp/runtime/test/ompt/parallel/nested_serialized.c
+++ b/openmp/runtime/test/ompt/parallel/nested_serialized.c
@@ -18,13 +18,29 @@ int main()
       print_ids(1);
       print_ids(2);
     }
+    print_fuzzy_address(1);
   }
+  print_fuzzy_address(2);
+
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
 
   // CHECK: 0: NULL_POINTER=[[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, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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=0x{{[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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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!
 
@@ -41,67 +57,71 @@ int main()
 
 
   // THREADS: 0: NULL_POINTER=[[NULL:.*$]]
-  // THREADS: {{^}}[[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, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+  // THREADS: {{^}}[[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]+]]
 
   // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_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:0x[0-f]+]], invoker=[[PARALLEL_INVOKER]]
+  // 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: {{^}}[[MASTER_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, codeptr_ra=[[NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][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]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
   // 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]]
-  // 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: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
 
   // 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]]: 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: {{^}}[[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, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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]]: 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: {{^}}[[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, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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]]: 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: {{^}}[[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, codeptr_ra=[[NESTED_RETURN_ADDRESS]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: level 2: parallel_id=0, task_id=[[PARENT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[NESTED_IMPLICIT_TASK_ID]]
   // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
   // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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]]
 
   return 0;
 }
diff --git a/openmp/runtime/test/ompt/parallel/no_thread_num_clause.c b/openmp/runtime/test/ompt/parallel/no_thread_num_clause.c
new file mode 100644
index 00000000000..251037edcd2
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/no_thread_num_clause.c
@@ -0,0 +1,95 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// REQUIRES: ompt
+#include "callback.h"
+
+int main()
+{
+  omp_set_num_threads(4);
+  #pragma omp parallel
+  {
+    print_ids(0);
+    print_ids(1);
+  }
+  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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: parallel_data initially not null
+  // CHECK-NOT: 0: task_data initially not null
+  // CHECK-NOT: 0: thread_data initially not null
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter={{0x[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // Note that we cannot ensure that the worker threads have already called barrier_end and implicit_task_end before parallel_end!
+
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+
+
+  // THREADS: 0: NULL_POINTER=[[NULL:.*$]]
+  // THREADS: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_thread_begin: thread_type=ompt_thread_initial=1, thread_id=[[MASTER_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=0, parent_task_frame.exit=(nil), parent_task_frame.reenter=(nil), new_task_id=281474976710658, codeptr_ra=(nil), task_type=ompt_task_initial=1, has_dependences=no
+  // 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={{[0-9]+}}
+
+  // 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]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  // 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]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/normal.c b/openmp/runtime/test/ompt/parallel/normal.c
index 36f57e7a1cb..015ff5c9a26 100644
--- a/openmp/runtime/test/ompt/parallel/normal.c
+++ b/openmp/runtime/test/ompt/parallel/normal.c
@@ -10,12 +10,27 @@ int main()
     print_ids(0);
     print_ids(1);
   }
+  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: 0: NULL_POINTER=[[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, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+  
+  // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, 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!
 
@@ -28,43 +43,48 @@ int main()
   // 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: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // 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]]
 
 
   // THREADS: 0: NULL_POINTER=[[NULL:.*$]]
-  // THREADS: {{^}}[[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, parallel_function=0x{{[0-f]+}}, invoker={{.*}}
+  // 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]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[MASTER_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]]
+  // 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]]
-  // 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: {{^}}[[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_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:[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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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:[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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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_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:[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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: 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]]
 
   return 0;
 }
diff --git a/openmp/runtime/test/ompt/parallel/not_enough_threads.c b/openmp/runtime/test/ompt/parallel/not_enough_threads.c
new file mode 100644
index 00000000000..cdc6b2f77e3
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/not_enough_threads.c
@@ -0,0 +1,76 @@
+// RUN: %libomp-compile && env OMP_THREAD_LIMIT=2 %libomp-run | FileCheck %s
+// RUN: %libomp-compile && env OMP_THREAD_LIMIT=2 %libomp-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// REQUIRES: ompt
+#include "callback.h"
+
+int main()
+{
+  #pragma omp parallel num_threads(4)
+  {
+    print_ids(0);
+    print_ids(1);
+  }
+  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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: parallel_data initially not null
+  // CHECK-NOT: 0: task_data initially not null
+  // CHECK-NOT: 0: thread_data initially not null
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter={{0x[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // Note that we cannot ensure that the worker threads have already called barrier_end and implicit_task_end before parallel_end!
+
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK-DAG: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+
+
+  // THREADS: 0: NULL_POINTER=[[NULL:.*$]]
+  // 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={{[0-9]+}}
+
+  // 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]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]]
+  // THREADS: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  // 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]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/parallel_if0.c b/openmp/runtime/test/ompt/parallel/parallel_if0.c
new file mode 100644
index 00000000000..8827448b3fb
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/parallel_if0.c
@@ -0,0 +1,75 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+
+int main()
+{
+//  print_frame(0);
+  #pragma omp parallel if(0)
+  {
+//    print_frame(1);
+    print_ids(0);
+    print_ids(1);
+//    print_frame(0);
+    #pragma omp parallel if(0)
+    {
+//      print_frame(1);
+      print_ids(0);
+      print_ids(1);
+      print_ids(2);
+//      print_frame(0);
+      #pragma omp task
+      {
+//        print_frame(1);
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+        print_ids(3);
+      }
+    }
+    print_fuzzy_address(1);
+  }
+  print_fuzzy_address(2);
+
+  // Check if libomp supports the callbacks for this test.
+  // 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_event_implicit_task_begin'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_event_implicit_task_end'
+
+  // 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=1, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_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=1, codeptr_ra=[[NESTED_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame={{0x[0-f]+}}
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id=[[EXPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[NESTED_IMPLICIT_TASK_ID]], second_task_id=[[EXPLICIT_TASK_ID]], prior_task_status=ompt_task_others=4
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[EXPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[EXPLICIT_TASK_ID]], second_task_id=[[NESTED_IMPLICIT_TASK_ID]], prior_task_status=ompt_task_complete=1
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_end: task_id=[[EXPLICIT_TASK_ID]]
+
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=0, task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[NESTED_RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/serialized.c b/openmp/runtime/test/ompt/parallel/serialized.c
index beb64da45f5..55bf98fee7c 100644
--- a/openmp/runtime/test/ompt/parallel/serialized.c
+++ b/openmp/runtime/test/ompt/parallel/serialized.c
@@ -4,21 +4,73 @@
 
 int main()
 {
+//  print_frame(0);
   #pragma omp parallel num_threads(1)
   {
+//    print_frame(1);
     print_ids(0);
     print_ids(1);
+//    print_frame(0);
+    #pragma omp parallel num_threads(1)
+    {
+//      print_frame(1);
+      print_ids(0);
+      print_ids(1);
+      print_ids(2);
+//      print_frame(0);
+      #pragma omp task
+      {
+//        print_frame(1);
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+        print_ids(3);
+      }
+    }
+    print_fuzzy_address(1);
   }
+  print_fuzzy_address(2);
+
+  // Check if libomp supports the callbacks for this test.
+  // 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_event_implicit_task_begin'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_event_implicit_task_end'
 
   // CHECK: 0: NULL_POINTER=[[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=1, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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=1, codeptr_ra=[[OUTER_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
 
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // CHECK: {{^}}[[MASTER_ID]]: level 1: parallel_id=0, task_id=[[PARENT_TASK_ID]]
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_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=1, codeptr_ra=[[INNER_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame={{0x[0-f]+}}
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[NESTED_IMPLICIT_TASK_ID]], parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id=[[EXPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[NESTED_IMPLICIT_TASK_ID]], second_task_id=[[EXPLICIT_TASK_ID]], prior_task_status=ompt_task_others=4
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[EXPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame={{0x[0-f]+}}, reenter_frame={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[EXPLICIT_TASK_ID]], second_task_id=[[NESTED_IMPLICIT_TASK_ID]], prior_task_status=ompt_task_complete=1
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_end: task_id=[[EXPLICIT_TASK_ID]]
+
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=0, task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]], codeptr_ra=[[INNER_RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[INNER_RETURN_ADDRESS]]
 
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=0, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], invoker=[[PARALLEL_INVOKER]], codeptr_ra=[[OUTER_RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[OUTER_RETURN_ADDRESS]]
 
   return 0;
 }
diff --git a/openmp/runtime/test/ompt/synchronization/barrier/explicit.c b/openmp/runtime/test/ompt/synchronization/barrier/explicit.c
new file mode 100644
index 00000000000..4def7f83931
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/barrier/explicit.c
@@ -0,0 +1,57 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp atomic
+    x++;
+
+    #pragma omp barrier
+    print_current_address();
+
+    #pragma omp atomic
+    x++;
+  }
+
+
+  // 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 explicit barrier 
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  // master thread implicit barrier at parallel end
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+
+
+  // worker thread explicit barrier 
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[THREAD_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  // worker thread implicit barrier at parallel end
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/barrier/for_loop.c b/openmp/runtime/test/ompt/synchronization/barrier/for_loop.c
new file mode 100644
index 00000000000..0e3abc59086
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/barrier/for_loop.c
@@ -0,0 +1,55 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int y[] = {0,1,2,3};
+
+  #pragma omp parallel num_threads(2)
+  {
+    //implicit barrier at end of for loop
+    int i;
+    #pragma omp for
+    for (i = 0; i < 4; i++)
+    {
+      y[i]++;
+    }
+    print_current_address();
+  }
+
+
+  // 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 loop end 
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+  // master thread implicit barrier at parallel end
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+
+  // worker thread explicit barrier 
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+  // worker thread implicit barrier after parallel
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/barrier/for_simd.c b/openmp/runtime/test/ompt/synchronization/barrier/for_simd.c
new file mode 100644
index 00000000000..376a435c04e
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/barrier/for_simd.c
@@ -0,0 +1,32 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int y[] = {0,1,2,3};
+
+  int i;
+  #pragma omp for simd
+  for (i = 0; i < 4; i++)
+  {
+    y[i]++;
+  }
+
+
+  // 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:.*$]]
+
+  // master thread implicit barrier at simd loop end 
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/barrier/parallel_region.c b/openmp/runtime/test/ompt/synchronization/barrier/parallel_region.c
new file mode 100644
index 00000000000..ea0a23f1ed5
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/barrier/parallel_region.c
@@ -0,0 +1,40 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+
+  //implicit barrier at end of a parallel region
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp atomic
+    x++;
+  }
+  print_fuzzy_address();
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_sync_region'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_sync_region_wait'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // master thread implicit barrier at parallel end
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+
+  // worker thread implicit barrier at parallel end
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/barrier/sections.c b/openmp/runtime/test/ompt/synchronization/barrier/sections.c
new file mode 100644
index 00000000000..4e1dfdd301c
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/barrier/sections.c
@@ -0,0 +1,63 @@
+// RUN: %libomp-compile-and-run | %sort-threads  | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+
+  #pragma omp parallel num_threads(2)
+  {
+    //implicit barrier after sections with nowait but with lastprivates
+    //implicit barrier at end of sections
+    #pragma omp sections
+    {
+      #pragma omp section 
+      {
+        #pragma omp atomic
+        x++;
+      }
+      
+      #pragma omp section 
+      {
+        #pragma omp atomic
+        x++;
+      }
+    }
+  }
+
+
+  // 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 sections end
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+  // master thread implicit barrier at parallel end
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+
+  // worker thread implicit barrier at sections end
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+  // worker thread implicit barrier at parallel end
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/barrier/single.c b/openmp/runtime/test/ompt/synchronization/barrier/single.c
new file mode 100644
index 00000000000..4b928ff16ef
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/barrier/single.c
@@ -0,0 +1,60 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+
+  #pragma omp parallel num_threads(2)
+  {
+    //implicit barrier at end of single
+    #pragma omp single
+    {
+      x++;
+    }
+    print_fuzzy_address();
+    //critical section to avoid merge of two barriers into one
+    #pragma omp critical
+    {
+      x++;
+    }
+  }
+
+
+  // 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 single end
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  // master thread implicit barrier at parallel end
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+
+
+  // worker thread implicit barrier at single end
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  // worker thread implicit barrier at parallel end
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_wait_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[NULL]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/critical.c b/openmp/runtime/test/ompt/synchronization/critical.c
new file mode 100644
index 00000000000..7787ede9ac6
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/critical.c
@@ -0,0 +1,31 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  #pragma omp critical
+  {
+    print_current_address(1);
+    print_ids(0);
+  }
+  print_current_address(2);
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_wait_critical: wait_id=[[WAIT_ID:[0-9]+]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_critical: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_critical: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/flush.c b/openmp/runtime/test/ompt/synchronization/flush.c
new file mode 100644
index 00000000000..875871d1f43
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/flush.c
@@ -0,0 +1,32 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// GCC generates code that does not call the runtime for the flush construct
+// XFAIL: gcc
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  #pragma omp parallel num_threads(2)
+  {
+    int tid = omp_get_thread_num();
+    
+    #pragma omp flush
+    print_current_address(1);
+  }
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_flush'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_flush: codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: current_address=[[RETURN_ADDRESS]]
+  //
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_flush: codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: current_address=[[RETURN_ADDRESS]]
+
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/lock.c b/openmp/runtime/test/ompt/synchronization/lock.c
new file mode 100644
index 00000000000..b546d1e8d7d
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/lock.c
@@ -0,0 +1,44 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  //need to use an OpenMP construct so that OMPT will be initalized
+  #pragma omp parallel num_threads(1)
+    print_ids(0);
+
+  omp_lock_t lock;
+  printf("%" PRIu64 ": &lock: %lli\n", ompt_get_thread_data()->value, (long long) &lock);
+  omp_init_lock(&lock);
+  print_current_address(1);
+  omp_set_lock(&lock);
+  print_current_address(2);
+  omp_unset_lock(&lock);
+  print_current_address(3);
+  omp_destroy_lock(&lock);
+  print_current_address(4);
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: &lock: [[WAIT_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_init_lock: wait_id=[[WAIT_ID]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_lock: wait_id=[[WAIT_ID]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_destroy_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+ 
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/master.c b/openmp/runtime/test/ompt/synchronization/master.c
new file mode 100644
index 00000000000..9e1e7c28e3b
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/master.c
@@ -0,0 +1,36 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+// GCC generates code that does not call the runtime for the master construct
+// XFAIL: gcc
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {
+      print_fuzzy_address(1);
+      x++;
+    }
+    print_current_address(2);
+  }
+
+  printf("%" PRIu64 ": x=%d\n", ompt_get_thread_data()->value, x);
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_master'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_master_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_master_end: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], codeptr_ra=[[RETURN_ADDRESS_END:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: current_address=[[RETURN_ADDRESS_END]]
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/nest_lock.c b/openmp/runtime/test/ompt/synchronization/nest_lock.c
new file mode 100644
index 00000000000..39cd4cc704a
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/nest_lock.c
@@ -0,0 +1,52 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  //need to use an OpenMP construct so that OMPT will be initalized
+  #pragma omp parallel num_threads(1)
+    print_ids(0);
+
+  omp_nest_lock_t nest_lock;
+  printf("%" PRIu64 ": &nest_lock: %lli\n", ompt_get_thread_data()->value, (long long) &nest_lock);
+  omp_init_nest_lock(&nest_lock);
+  print_current_address(1);
+  omp_set_nest_lock(&nest_lock);
+  print_current_address(2);
+  omp_set_nest_lock(&nest_lock);
+  print_current_address(3);
+  omp_unset_nest_lock(&nest_lock);
+  print_current_address(4);
+  omp_unset_nest_lock(&nest_lock);
+  print_current_address(5);
+  omp_destroy_nest_lock(&nest_lock);
+  print_current_address(6);
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_init_nest_lock: wait_id=[[WAIT_ID:[0-9]+]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_first: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_next: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_prev: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_last: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_destroy_nest_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/ordered.c b/openmp/runtime/test/ompt/synchronization/ordered.c
new file mode 100644
index 00000000000..1d498b9e122
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/ordered.c
@@ -0,0 +1,31 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  #pragma omp ordered
+  {
+    print_current_address(1);
+    print_ids(0);
+  }
+  print_current_address(2);
+  
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_wait_ordered: wait_id=[[WAIT_ID:[0-9]+]], hint={{[0-9]+}}, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_ordered: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_ordered: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/taskgroup.c b/openmp/runtime/test/ompt/synchronization/taskgroup.c
new file mode 100644
index 00000000000..a85bf205991
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/taskgroup.c
@@ -0,0 +1,48 @@
+// RUN:  %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <unistd.h>  
+#include <stdio.h>
+
+int main()
+{
+  int condition=0;
+  int x=0;
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {
+      #pragma omp taskgroup
+      {
+        print_current_address(1);
+        #pragma omp task
+        {
+          #pragma omp atomic
+          x++;
+        }
+      }
+      print_current_address(2);
+    }
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_master'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_cancel'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_thread_begin'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_taskgroup_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_taskgroup_begin: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_taskgroup_end: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_taskgroup_end: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/taskwait.c b/openmp/runtime/test/ompt/synchronization/taskwait.c
new file mode 100644
index 00000000000..3e319ff8229
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/taskwait.c
@@ -0,0 +1,35 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {
+      #pragma omp task
+      {
+        x++;
+      }
+      #pragma omp taskwait
+      print_current_address(1);
+    }
+  }
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_sync_region'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_sync_region_wait'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_taskwait_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_taskwait_begin: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_taskwait_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: ompt_event_taskwait_end: parallel_id={{[0-9]+}}, task_id={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS]]
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/test_lock.c b/openmp/runtime/test/ompt/synchronization/test_lock.c
new file mode 100644
index 00000000000..bae46416a79
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/test_lock.c
@@ -0,0 +1,54 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  omp_lock_t lock;
+  omp_init_lock(&lock);
+  print_current_address(1);
+
+  omp_test_lock(&lock);
+  print_current_address(2);
+  omp_unset_lock(&lock);
+  print_current_address(3);
+
+  omp_set_lock(&lock);
+  print_current_address(4);
+  omp_test_lock(&lock);
+  print_current_address(5);
+  omp_unset_lock(&lock);
+  print_current_address(6);
+
+  omp_destroy_lock(&lock);
+  print_current_address(7);
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_init_lock: wait_id=[[WAIT_ID:[0-9]+]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]] 
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]]  
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]  
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]]  
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]  
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]  
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_destroy_lock: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]  
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/test_nest_lock.c b/openmp/runtime/test/ompt/synchronization/test_nest_lock.c
new file mode 100644
index 00000000000..ad02d32c9c8
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/test_nest_lock.c
@@ -0,0 +1,42 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  omp_nest_lock_t nest_lock;
+  omp_init_nest_lock(&nest_lock);
+
+  omp_test_nest_lock(&nest_lock);
+  omp_unset_nest_lock(&nest_lock);
+
+  omp_set_nest_lock(&nest_lock);
+  omp_test_nest_lock(&nest_lock);
+  omp_unset_nest_lock(&nest_lock);
+  omp_unset_nest_lock(&nest_lock);
+
+  omp_destroy_nest_lock(&nest_lock);
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_init_nest_lock: wait_id=[[WAIT_ID:[0-9]+]], hint=0, impl={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_first: wait_id=[[WAIT_ID]], codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_last: wait_id=[[WAIT_ID]], codeptr_ra={{0x[0-f]+}}  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_first: wait_id=[[WAIT_ID]], codeptr_ra={{0x[0-f]+}}  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_next: wait_id=[[WAIT_ID]], codeptr_ra={{0x[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_prev: wait_id=[[WAIT_ID]], codeptr_ra={{0x[0-f]+}}  
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_last: wait_id=[[WAIT_ID]], codeptr_ra={{0x[0-f]+}}  
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/synchronization/test_nest_lock_parallel.c b/openmp/runtime/test/ompt/synchronization/test_nest_lock_parallel.c
new file mode 100644
index 00000000000..0e1dcece8c1
--- /dev/null
+++ b/openmp/runtime/test/ompt/synchronization/test_nest_lock_parallel.c
@@ -0,0 +1,59 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  omp_nest_lock_t nest_lock;
+  omp_init_nest_lock(&nest_lock);
+
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {
+      omp_set_nest_lock(&nest_lock);
+      print_current_address(1);
+    }
+    #pragma omp barrier
+    omp_test_nest_lock(&nest_lock); //should fail for non-master
+    print_current_address(2);
+    #pragma omp barrier
+    #pragma omp master
+    {
+      omp_unset_nest_lock(&nest_lock);
+      print_current_address(3);
+      omp_unset_nest_lock(&nest_lock);
+      print_current_address(4);
+    }
+  }
+
+  omp_destroy_nest_lock(&nest_lock);
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_nest_lock'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID:[0-9]+]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]] 
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_first: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]] 
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]] 
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_acquired_nest_lock_next: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS]] 
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_prev: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]] 
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_release_nest_lock_last: wait_id=[[WAIT_ID]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]] 
+  // CHECK-NEXT: {{^}}[[MASTER_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_destroy_nest_lock: wait_id=[[WAIT_ID]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_wait_nest_lock: wait_id=[[WAIT_ID]], hint=0, impl={{[0-9]+}}, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]] 
+  // CHECK-NOT: {{^}}[[THREAD_ID]]: ompt_event_acquired_nest_lock_next: wait_id=[[WAIT_ID]]
+  // CHECK-NEXT: {{^}}[[THREAD_ID]]: current_address={{.*}}[[RETURN_ADDRESS]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/dependences.c b/openmp/runtime/test/ompt/tasks/dependences.c
new file mode 100644
index 00000000000..32817c4e2dd
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/dependences.c
@@ -0,0 +1,53 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <omp.h>   
+#include <math.h>
+#include <unistd.h>
+
+int main()
+{
+  int x = 0;
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {  
+      #pragma omp task depend(out:x)
+      {
+        x++;
+        delay(100);
+      }
+      print_fuzzy_address(1);
+    
+      #pragma omp task depend(in:x)
+      {
+        x = -1;
+      }
+    }
+  }
+
+  x++;
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_dependences'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_dependence'
+  
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: new_task_data initially not null
+
+  // CHECK: {{^}}{{[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter=[[NULL]], new_task_id=[[FIRST_TASK:[0-f]+]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, task_type=ompt_task_explicit=4, has_dependences=yes
+  // CHECK: {{^}}{{[0-9]+}}: ompt_event_task_dependences: task_id=[[FIRST_TASK]], deps={{0x[0-f]+}}, ndeps=1
+  // CHECK: {{^}}{{[0-9]+}}: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+
+  // CHECK: {{^}}{{[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter=[[NULL]], new_task_id=[[SECOND_TASK:[0-f]+]], codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit=4, has_dependences=yes
+  // CHECK: {{^}}{{[0-9]+}}: ompt_event_task_dependences: task_id=[[SECOND_TASK]], deps={{0x[0-f]+}}, ndeps=1
+  // CHECK: {{^}}{{[0-9]+}}: ompt_event_task_dependence_pair: first_task_id=[[FIRST_TASK]], second_task_id=[[SECOND_TASK]]
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/explicit_task.c b/openmp/runtime/test/ompt/tasks/explicit_task.c
new file mode 100644
index 00000000000..491192426ec
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/explicit_task.c
@@ -0,0 +1,100 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h> 
+
+int main()
+{
+  int condition=0;
+  omp_set_nested(0);
+  print_frame(0);
+  #pragma omp parallel num_threads(2)
+  {
+    print_frame(1);
+    print_ids(0);
+    print_ids(1);
+    print_frame(0);
+    #pragma omp master
+    {
+      print_ids(0);
+      #pragma omp task shared(condition)
+      {
+        OMPT_SIGNAL(condition);
+        print_frame(1);
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+      }
+      print_fuzzy_address(1);
+      OMPT_WAIT(condition,1);
+      print_ids(0);
+    }
+    #pragma omp barrier
+    print_ids(0);
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: new_task_data initially not null
+  
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address(0)=[[MAIN_REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+  // nested parallel masters
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // <- ompt_event_task_create would be expected here
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[EXIT]], parent_task_frame.reenter=[[REENTER]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // explicit barrier after master
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // implicit barrier parallel
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // this is expected to come earlier and at MASTER:
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[IMPLICIT_TASK_ID]], second_task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[TASK_EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], exit_frame=[[TASK_EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[TASK_ID]], second_task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_end: task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/serialized.c b/openmp/runtime/test/ompt/tasks/serialized.c
new file mode 100644
index 00000000000..1fb6f066481
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/serialized.c
@@ -0,0 +1,93 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>   
+#include <math.h>
+
+int main()
+{
+  omp_set_nested(0);
+  print_frame(0);
+  #pragma omp parallel num_threads(2)
+  {
+    print_frame(1);
+    print_ids(0);
+    print_ids(1);
+    print_frame(0);
+    #pragma omp master
+    {
+      print_ids(0);
+      int t = (int)sin(0.1);
+      #pragma omp task if(t)
+      {
+        print_frame(1);
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+      }
+      print_fuzzy_address(1);
+      print_ids(0);
+    }
+    print_ids(0);
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+  
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: new_task_data initially not null
+  
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[NULL]], new_task_id={{[0-9]+}}, codeptr_ra=[[NULL]], task_type=ompt_task_initial=1, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[MAIN_REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+
+  // nested parallel masters
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[EXIT]], parent_task_frame.reenter=[[REENTER]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // <- ompt_event_task_schedule ([[IMPLICIT_TASK_ID]], [[TASK_ID]]) would be expected here
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[IMPLICIT_TASK_ID]], second_task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[TASK_EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], exit_frame=[[TASK_EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // <- ompt_event_task_schedule ([[TASK_ID]], [[IMPLICIT_TASK_ID]]) would be expected here
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[TASK_ID]], second_task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_end: task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reen
+
+  // implicit barrier parallel
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/task_in_joinbarrier.c b/openmp/runtime/test/ompt/tasks/task_in_joinbarrier.c
new file mode 100644
index 00000000000..74d8cc7eec2
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/task_in_joinbarrier.c
@@ -0,0 +1,90 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h> 
+
+int main()
+{
+  int condition=0;
+  omp_set_nested(0);
+  print_frame(0);
+  #pragma omp parallel num_threads(2)
+  {
+    print_frame(1);
+    print_ids(0);
+    print_ids(1);
+    print_frame(0);
+    #pragma omp master
+    {
+      print_ids(0);
+      #pragma omp task shared(condition)
+      {
+        OMPT_SIGNAL(condition);
+        print_frame(1);
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+      }
+      OMPT_WAIT(condition,1);
+      print_ids(0);
+    }
+    print_ids(0);
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: new_task_data initially not null
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address(0)=[[MAIN_REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+  // nested parallel masters
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // <- ompt_event_task_create would be expected here
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[EXIT]], parent_task_frame.reenter=[[REENTER]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[TASK_FUNCTION:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // implicit barrier parallel
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // implicit barrier parallel
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[IMPLICIT_TASK_ID]], second_task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[TASK_EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], exit_frame=[[TASK_EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[TASK_ID]], second_task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_end: task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/task_types.c b/openmp/runtime/test/ompt/tasks/task_types.c
new file mode 100644
index 00000000000..7a4f775cc48
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/task_types.c
@@ -0,0 +1,112 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h>
+#include <math.h>
+
+
+void print_task_type(int id)
+{
+  #pragma omp critical
+  {
+    int task_type;
+    char buffer[2048];
+    ompt_get_task_info(0, &task_type, NULL, NULL, NULL, NULL);
+    format_task_type(task_type, buffer);
+    printf("%" PRIu64 ": id=%d task_type=%s=%d\n", ompt_get_thread_data()->value, id, buffer, task_type);
+  }
+};
+
+int main()
+{
+  //initial task
+  print_task_type(0);
+
+  int x;
+  //implicit task
+  #pragma omp parallel num_threads(1)
+  {
+    print_task_type(1);
+    x++;
+  }
+
+  #pragma omp parallel num_threads(2)
+  #pragma omp master
+  {
+    //explicit task
+    #pragma omp task
+    {
+      print_task_type(2);
+      x++;
+    }
+
+    //explicit task with undeferred
+    #pragma omp task if(0)
+    {
+      print_task_type(3);
+      x++;
+    }
+
+    //explicit task with untied
+    #pragma omp task untied
+    {
+      print_task_type(4);
+      x++;
+    }
+
+    //explicit task with final
+    #pragma omp task final(1)
+    {
+      print_task_type(5);
+      x++;
+      //nested explicit task with final and undeferred
+      #pragma omp task
+      {
+        print_task_type(6);
+        x++;
+      }
+    }
+
+    //Mergeable task test deactivated for now
+    //explicit task with mergeable
+    /*
+    #pragma omp task mergeable if((int)sin(0))
+    {
+      print_task_type(7);
+      x++;
+    }
+    */
+
+    //TODO: merged task
+  }
+
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_task_create: parent_task_id=0, parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[NULL]], new_task_id={{[0-9]+}}, codeptr_ra=[[NULL]], task_type=ompt_task_initial=1, has_dependences=no
+  // CHECK-NOT: 0: parallel_data initially not null
+  // CHECK: {{^}}[[MASTER_ID]]: id=0 task_type=ompt_task_initial=1
+  // CHECK: {{^}}[[MASTER_ID]]: id=1 task_type=ompt_task_implicit|ompt_task_undeferred=134217730
+
+  // CHECK-DAG: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit=4, has_dependences=no
+  // CHECK-DAG: {{^[0-9]+}}: id=2 task_type=ompt_task_explicit=4
+
+  // CHECK-DAG: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred=134217732, has_dependences=no
+  // CHECK-DAG: {{^[0-9]+}}: id=3 task_type=ompt_task_explicit|ompt_task_undeferred=134217732
+
+  // CHECK-DAG: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_untied=268435460, has_dependences=no
+  // CHECK-DAG: {{^[0-9]+}}: id=4 task_type=ompt_task_explicit|ompt_task_untied=268435460
+
+  // CHECK-DAG: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_final=536870916, has_dependences=no
+  // CHECK-DAG: {{^[0-9]+}}: id=5 task_type=ompt_task_explicit|ompt_task_final=536870916
+
+  // CHECK-DAG: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_final=671088644, has_dependences=no
+  // CHECK-DAG: {{^[0-9]+}}: id=6 task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_final=671088644
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/task_types_serialized.c b/openmp/runtime/test/ompt/tasks/task_types_serialized.c
new file mode 100644
index 00000000000..dfd5598d24d
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/task_types_serialized.c
@@ -0,0 +1,112 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <omp.h>
+
+void print_task_type(int id)
+{
+  #pragma omp critical
+  {
+    int task_type;
+    char buffer[2048];
+    ompt_get_task_info(0, &task_type, NULL, NULL, NULL, NULL);
+    format_task_type(task_type, buffer);
+    printf("%" PRIu64 ": id=%d task_type=%s=%d\n", ompt_get_thread_data()->value, id, buffer, task_type);
+  }
+};
+
+int main()
+{
+  //initial task
+  print_task_type(0);
+
+  int x;
+  //implicit task
+  #pragma omp parallel num_threads(1)
+  {
+    print_task_type(1);
+    x++;
+  }
+
+  #pragma omp parallel num_threads(1)
+  #pragma omp master
+  {
+    //explicit task
+    #pragma omp task
+    {
+      print_task_type(2);
+      x++;
+    }
+
+    //explicit task with undeferred
+    #pragma omp task if(0)
+    {
+      print_task_type(3);
+      x++;
+    }
+
+    //explicit task with untied
+    #pragma omp task untied
+    {
+      print_task_type(4);
+      x++;
+    }
+
+    //explicit task with final
+    #pragma omp task final(1)
+    {
+      print_task_type(5);
+      x++;
+      //nested explicit task with final and undeferred
+      #pragma omp task
+      {
+        print_task_type(6);
+        x++;
+      }
+    }
+
+/*
+    //TODO:not working
+    //explicit task with mergeable
+    #pragma omp task mergeable
+    {
+      print_task_type(7);
+      x++;
+    }
+*/
+
+    //TODO: merged task
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_task_create: parent_task_id=0, parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[NULL]], new_task_id={{[0-9]+}}, codeptr_ra=[[NULL]], task_type=ompt_task_initial=1, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: id=0 task_type=ompt_task_initial=1
+  // CHECK: {{^}}[[MASTER_ID]]: id=1 task_type=ompt_task_implicit|ompt_task_undeferred=134217730
+
+  // CHECK: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred=134217732, has_dependences=no
+  // CHECK: {{^[0-9]+}}: id=2 task_type=ompt_task_explicit|ompt_task_undeferred=134217732
+
+  // CHECK: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred=134217732, has_dependences=no
+  // CHECK: {{^[0-9]+}}: id=3 task_type=ompt_task_explicit|ompt_task_undeferred=134217732
+
+  // CHECK: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_untied=402653188, has_dependences=no
+  // CHECK: {{^[0-9]+}}: id=4 task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_untied=402653188
+
+  // CHECK: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_final=671088644, has_dependences=no
+  // CHECK: {{^[0-9]+}}: id=5 task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_final=671088644
+
+  // CHECK: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_final=671088644, has_dependences=no
+  // CHECK: {{^[0-9]+}}: id=6 task_type=ompt_task_explicit|ompt_task_undeferred|ompt_task_final=671088644
+
+  // ___CHECK: {{^[0-9]+}}: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id={{[0-9]+}}, codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit|ompt_task_undeferred=134217732, has_dependences=no
+  // ___CHECK: {{^[0-9]+}}: id=7 task_type=ompt_task_explicit|ompt_task_undeferred=134217732
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/taskyield.c b/openmp/runtime/test/ompt/tasks/taskyield.c
new file mode 100644
index 00000000000..56a46970cf3
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/taskyield.c
@@ -0,0 +1,62 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// Current GOMP interface implements taskyield as stub
+// XFAIL: gcc
+
+#include "callback.h"
+#include <omp.h>   
+#include <unistd.h>
+
+int main()
+{
+  int condition=0, x=0;
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp master
+    {
+        #pragma omp task shared(condition)
+        {
+          OMPT_SIGNAL(condition);
+          OMPT_WAIT(condition,2);
+        }
+        OMPT_WAIT(condition,1);
+        #pragma omp task shared(x)
+        {
+          x++;
+        }
+        printf("%" PRIu64 ": before yield\n", ompt_get_thread_data()->value);
+        #pragma omp taskyield
+        printf("%" PRIu64 ": after yield\n", ompt_get_thread_data()->value);
+        OMPT_SIGNAL(condition);
+    }
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_implicit_task'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: new_task_data initially not null
+  
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID:[0-9]+]], team_size={{[0-9]+}}, thread_num={{[0-9]+}}
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id=[[WORKER_TASK:[0-9]+]], codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit=4, has_dependences=no
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id={{[0-9]+}}, parent_task_frame.exit={{0x[0-f]+}}, parent_task_frame.reenter={{0x[0-f]+}}, new_task_id=[[MAIN_TASK:[0-9]+]], codeptr_ra={{0x[0-f]+}}, task_type=ompt_task_explicit=4, has_dependences=no
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[IMPLICIT_TASK_ID]], second_task_id=[[MAIN_TASK]], prior_task_status=ompt_task_yield=2
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_schedule: first_task_id=[[MAIN_TASK]], second_task_id=[[IMPLICIT_TASK_ID]], prior_task_status=ompt_task_complete=1
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_task_schedule: first_task_id={{[0-9]+}}, second_task_id=[[WORKER_TASK]], prior_task_status=ompt_task_others=4
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[WORKER_TASK]], second_task_id={{[0-9]+}}, prior_task_status=ompt_task_complete=1
+
+
+
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/tasks/untied_task.c b/openmp/runtime/test/ompt/tasks/untied_task.c
new file mode 100644
index 00000000000..1adfe8f96fe
--- /dev/null
+++ b/openmp/runtime/test/ompt/tasks/untied_task.c
@@ -0,0 +1,107 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+#include "callback.h"
+#include <omp.h> 
+
+int main()
+{
+  int condition=0;
+  omp_set_nested(0);
+  print_frame(0);
+  #pragma omp parallel num_threads(2)
+  {
+    print_frame(1);
+    print_ids(0);
+    print_ids(1);
+    print_frame(0);
+    #pragma omp master
+    {
+      print_ids(0);
+      #pragma omp task untied shared(condition)
+      {
+        OMPT_SIGNAL(condition);
+        print_frame(1);
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+        #pragma omp task if(0)
+        {
+          print_ids(0);
+          print_ids(1);
+          print_ids(2);
+        }
+        print_ids(0);
+        print_ids(1);
+        print_ids(2);
+      }
+      OMPT_WAIT(condition,1);
+      print_ids(0);
+    }
+    #pragma omp barrier
+    print_ids(0);
+  }
+
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_schedule'
+  // 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 'ompt_callback_mutex_acquire'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_acquired'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_mutex_released'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+
+  // make sure initial data pointers are null
+  // CHECK-NOT: 0: new_task_data initially not null
+  
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: __builtin_frame_address(0)=[[MAIN_REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[MAIN_REENTER]], parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=2, codeptr_ra=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:[0-9]+]]
+  // nested parallel masters
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID:[0-9]+]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // <- ompt_event_task_create would be expected here
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame.exit=[[EXIT]], parent_task_frame.reenter=[[REENTER]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[TASK_FUNCTION:0x[0-f]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // explicit barrier after master
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // implicit barrier parallel
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(0)=[[REENTER:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // this is expected to come earlier and at MASTER:
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[IMPLICIT_TASK_ID]], second_task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: __builtin_frame_address(1)=[[TASK_EXIT:0x[0-f]+]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], exit_frame=[[TASK_EXIT]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 1: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[EXIT]], reenter_frame=[[REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 2: parallel_id=[[IMPLICIT_PARALLEL_ID]], task_id=[[PARENT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[MAIN_REENTER]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_schedule: first_task_id=[[TASK_ID]], second_task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_task_end: task_id=[[TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: task level 0: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], exit_frame=[[NULL]], reenter_frame=[[NULL]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_barrier_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
+
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/worksharing/for/auto_split.c b/openmp/runtime/test/ompt/worksharing/for/auto_split.c
new file mode 100644
index 00000000000..d82e3fd1ef6
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/for/auto_split.c
@@ -0,0 +1,8 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_split.h
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=CHECK-LOOP %S/base_split.h
+// REQUIRES: ompt
+// GCC doesn't call runtime for auto = static schedule
+// XFAIL: gcc
+
+#define SCHEDULE auto
+#include "base_split.h"
diff --git a/openmp/runtime/test/ompt/worksharing/for/base.h b/openmp/runtime/test/ompt/worksharing/for/base.h
index 73988792680..96022d08ad5 100644
--- a/openmp/runtime/test/ompt/worksharing/for/base.h
+++ b/openmp/runtime/test/ompt/worksharing/for/base.h
@@ -9,28 +9,35 @@ int main()
   for (i = 0; i < 4; i++) {
   }
 
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_work'
+
+
   // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
-  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: 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, parallel_function=0x{{[0-f]+}}, invoker={{.*}}
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: 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=0x{{[0-f]+}}, invoker={{[0-9]+}}
 
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], workshare_function=0x{{[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], workshare_function=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], workshare_function=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], workshare_function=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   return 0;
 }
diff --git a/openmp/runtime/test/ompt/worksharing/for/base_serialized.h b/openmp/runtime/test/ompt/worksharing/for/base_serialized.h
index bd8b226e003..3376b3702ea 100644
--- a/openmp/runtime/test/ompt/worksharing/for/base_serialized.h
+++ b/openmp/runtime/test/ompt/worksharing/for/base_serialized.h
@@ -8,14 +8,21 @@ int main()
   #pragma omp parallel for num_threads(1) schedule(SCHEDULE)
   for (i = 0; i < 1; i++) {
   }
+  
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_work'
+
 
   // CHECK: 0: NULL_POINTER=[[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=1, parallel_function=0x{{[0-f]+}}, invoker={{.+}}
+  // 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=1, codeptr_ra=0x{{[0-f]+}}, invoker={{[0-9]+}}
 
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], workshare_function=0x{{[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[PARALLEL_ID,0]}}, task_id=[[IMPLICIT_TASK_ID]]
 
   return 0;
 }
diff --git a/openmp/runtime/test/ompt/worksharing/for/base_split.h b/openmp/runtime/test/ompt/worksharing/for/base_split.h
new file mode 100644
index 00000000000..0f1fed3c449
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/for/base_split.h
@@ -0,0 +1,66 @@
+#include "callback.h"
+#include <omp.h>
+
+/* With the combined parallel-for construct (base.h), the return-addresses are hard to compare.
+   With the separate parallel and for-nowait construct, the addresses become more predictable,
+   but the begin of the for-loop still generates additional code, so the offset of loop-begin 
+   to the label is >4 Byte.
+*/
+
+int main()
+{
+  unsigned int i;
+
+  #pragma omp parallel num_threads(4) 
+  {
+    print_current_address(0);
+    #pragma omp for schedule(SCHEDULE) nowait
+    for (i = 0; i < 4; i++) {
+      print_fuzzy_address(1);
+    }
+    print_fuzzy_address(2);
+  }
+  print_fuzzy_address(3);
+
+  // Check if libomp supports the callbacks for this test.
+  // 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 'ompt_callback_work'
+
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: 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=[[PARALLEL_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}, invoker={{[0-9]+}}
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id={{[0-9]+}}, codeptr_ra=[[LOOP_BEGIN_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, codeptr_ra=[[LOOP_END_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[LOOP_END_RETURN_ADDRESS]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, invoker={{[0-9]+}}, codeptr_ra=[[PARALLEL_RETURN_ADDRESS]]
+  // CHECK: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[PARALLEL_RETURN_ADDRESS]]
+  
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, codeptr_ra=[[LOOP_END_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[LOOP_END_RETURN_ADDRESS]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, codeptr_ra=[[LOOP_END_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[LOOP_END_RETURN_ADDRESS]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id={{[0-9]+}}, codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, codeptr_ra=[[LOOP_END_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[LOOP_END_RETURN_ADDRESS]]
+
+
+  // CHECK-LOOP: 0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK-LOOP: 0: ompt_event_runtime_shutdown
+  // CHECK-LOOP: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: 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={{0x[0-f]+}}, invoker={{[0-9]+}}
+  // CHECK-LOOP: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id={{[0-9]+}}, codeptr_ra=[[LOOP_BEGIN_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
+  // CHECK-LOOP: {{^}}{{[0-9]+}}: fuzzy_address={{.*}}[[LOOP_BEGIN_RETURN_ADDRESS]]
+  // CHECK-LOOP: {{^}}{{[0-9]+}}: fuzzy_address={{.*}}[[LOOP_BEGIN_RETURN_ADDRESS]]
+  // CHECK-LOOP: {{^}}{{[0-9]+}}: fuzzy_address={{.*}}[[LOOP_BEGIN_RETURN_ADDRESS]]
+  // CHECK-LOOP: {{^}}{{[0-9]+}}: fuzzy_address={{.*}}[[LOOP_BEGIN_RETURN_ADDRESS]]
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/worksharing/for/dynamic_split.c b/openmp/runtime/test/ompt/worksharing/for/dynamic_split.c
new file mode 100644
index 00000000000..3886f71319e
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/for/dynamic_split.c
@@ -0,0 +1,6 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_split.h
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=CHECK-LOOP %S/base_split.h
+// REQUIRES: ompt
+
+#define SCHEDULE dynamic
+#include "base_split.h"
diff --git a/openmp/runtime/test/ompt/worksharing/for/guided_split.c b/openmp/runtime/test/ompt/worksharing/for/guided_split.c
new file mode 100644
index 00000000000..aaefbb31d86
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/for/guided_split.c
@@ -0,0 +1,6 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_split.h
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=CHECK-LOOP %S/base_split.h
+// REQUIRES: ompt
+
+#define SCHEDULE guided
+#include "base_split.h"
diff --git a/openmp/runtime/test/ompt/worksharing/for/runtime_split.c b/openmp/runtime/test/ompt/worksharing/for/runtime_split.c
new file mode 100644
index 00000000000..032353e4127
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/for/runtime_split.c
@@ -0,0 +1,6 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_split.h
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=CHECK-LOOP %S/base_split.h
+// REQUIRES: ompt
+
+#define SCHEDULE runtime
+#include "base_split.h"
diff --git a/openmp/runtime/test/ompt/worksharing/for/static_split.c b/openmp/runtime/test/ompt/worksharing/for/static_split.c
new file mode 100644
index 00000000000..d8c88dd015f
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/for/static_split.c
@@ -0,0 +1,8 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_split.h
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=CHECK-LOOP %S/base_split.h
+// REQUIRES: ompt
+// GCC doesn't call runtime for static schedule
+// XFAIL: gcc
+
+#define SCHEDULE static
+#include "base_split.h"
diff --git a/openmp/runtime/test/ompt/worksharing/sections.c b/openmp/runtime/test/ompt/worksharing/sections.c
new file mode 100644
index 00000000000..7d274c4dfe6
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/sections.c
@@ -0,0 +1,36 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+// GCC generates code that does not distinguish between sections and loops
+// XFAIL: gcc
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  #pragma omp parallel sections num_threads(2)
+  {
+    #pragma omp section
+    {
+      printf("%lu: section 1\n", ompt_get_thread_data()->value);
+    }
+    #pragma omp section
+    {
+      printf("%lu: section 2\n", ompt_get_thread_data()->value);
+    }
+  }
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_work'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_sections_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], parent_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[SECT_BEGIN:0x[0-f]+]], count=2
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_sections_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, codeptr_ra=[[SECT_END:0x[0-f]+]]
+
+  // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_sections_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[SECT_BEGIN]], count=2
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_sections_end: parallel_id=[[PARALLEL_ID]], task_id={{[0-9]+}}, codeptr_ra=[[SECT_END]]
+
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/worksharing/single.c b/openmp/runtime/test/ompt/worksharing/single.c
new file mode 100644
index 00000000000..94865c4ee99
--- /dev/null
+++ b/openmp/runtime/test/ompt/worksharing/single.c
@@ -0,0 +1,36 @@
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// GCC generates code that does not call the runtime for the single construct
+// XFAIL: gcc
+
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  int x = 0;
+  #pragma omp parallel num_threads(2)
+  {
+    #pragma omp single
+    {
+      x++;
+    }
+  }
+
+  printf("x=%d\n", x);
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_work'
+
+  // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
+
+  // CHECK: {{^}}[[THREAD_ID_1:[0-9]+]]: ompt_event_single_in_block_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], parent_task_id=[[TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]+}}, count=1
+  // CHECK: {{^}}[[THREAD_ID_1]]: ompt_event_single_in_block_end: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], codeptr_ra={{0x[0-f]+}}, count=1
+
+  // CHECK: {{^}}[[THREAD_ID_2:[0-9]+]]: ompt_event_single_others_begin: parallel_id=[[PARALLEL_ID:[0-9]+]], task_id=[[TASK_ID:[0-9]+]], codeptr_ra={{0x[0-f]+}}, count=1
+  // CHECK: {{^}}[[THREAD_ID_2]]: ompt_event_single_others_end: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID]], codeptr_ra={{0x[0-f]+}}, count=1
+
+
+
+  return 0;
+}