5 files changed, 375 insertions, 0 deletions
diff --git a/openmp/runtime/test/ompt/callback.h b/openmp/runtime/test/ompt/callback.h
new file mode 100644
index 00000000000..f92d50ae32e
--- /dev/null
+++ b/openmp/runtime/test/ompt/callback.h
@@ -0,0 +1,80 @@
+#include <stdio.h>
+#include <inttypes.h>
+#include <ompt.h>
+
+static ompt_get_thread_id_t ompt_get_thread_id;
+
+static void
+on_ompt_event_barrier_begin(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t task_id)
+{
+  printf("%" PRIu64 ": ompt_event_barrier_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+}
+
+static void
+on_ompt_event_barrier_end(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t task_id)
+{
+  printf("%" PRIu64 ": ompt_event_barrier_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+}
+
+static void
+on_ompt_event_implicit_task_begin(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t task_id)
+{
+  printf("%" PRIu64 ": ompt_event_implicit_task_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+}
+
+static void
+on_ompt_event_implicit_task_end(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t task_id)
+{
+  printf("%" PRIu64 ": ompt_event_implicit_task_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+}
+
+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,
+  uint32_t requested_team_size,
+  void *parallel_function,
+  ompt_invoker_t invoker)
+{
+  printf("%" PRIu64 ": ompt_event_parallel_begin: parent_task_id=%" PRIu64 ", parent_task_frame=%p, parallel_id=%" PRIu64 ", requested_team_size=%" PRIu32 ", parallel_function=%p, invoker=%d\n", ompt_get_thread_id(), parent_task_id, parent_task_frame, parallel_id, requested_team_size, parallel_function, invoker);
+}
+
+static void
+on_ompt_event_parallel_end(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t task_id,
+  ompt_invoker_t invoker)
+{
+  printf("%" PRIu64 ": ompt_event_parallel_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", invoker=%d\n", ompt_get_thread_id(), parallel_id, task_id, invoker);
+}
+
+
+void ompt_initialize(
+  ompt_function_lookup_t lookup,
+  const char *runtime_version,
+  unsigned int ompt_version)
+{
+  ompt_set_callback_t ompt_set_callback = (ompt_set_callback_t) lookup("ompt_set_callback");
+  ompt_get_thread_id = (ompt_get_thread_id_t) lookup("ompt_get_thread_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_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);
+}
+
+ompt_initialize_t ompt_tool()
+{
+  return &ompt_initialize;
+}
diff --git a/openmp/runtime/test/ompt/parallel/nested.c b/openmp/runtime/test/ompt/parallel/nested.c
new file mode 100644
index 00000000000..3e25a06fc3c
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/nested.c
@@ -0,0 +1,146 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// REQUIRES: ompt, ompt-trace
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  omp_set_nested(1);
+
+  #pragma omp parallel num_threads(4)
+  {
+    #pragma omp parallel num_threads(4)
+    {
+    }
+  }
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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]]
+
+  // 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: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id={{[0-9]+}}, parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=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]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_barrier_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID]]
+  // 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=[[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: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // 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_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // 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_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // 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_parallel_end: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]], invoker=[[PARALLEL_INVOKER]]
+  // 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]]
+
+  // 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]]: 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/nested_serialized.c b/openmp/runtime/test/ompt/parallel/nested_serialized.c
new file mode 100644
index 00000000000..4caa9fbc2e0
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/nested_serialized.c
@@ -0,0 +1,76 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// REQUIRES: ompt, ompt-trace
+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  omp_set_nested(0);
+
+  #pragma omp parallel num_threads(4)
+  {
+    #pragma omp parallel num_threads(4)
+    {
+    }
+  }
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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]]
+
+  // 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: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id={{[0-9]+}}, parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=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]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // 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=[[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: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], 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]]: 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], 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]]: 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_parallel_begin: parent_task_id=[[IMPLICIT_TASK_ID]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[NESTED_PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=[[NESTED_PARALLEL_FUNCTION]], invoker=[[PARALLEL_INVOKER]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_begin: parallel_id=[[NESTED_PARALLEL_ID]], task_id=[[NESTED_IMPLICIT_TASK_ID:[0-9]+]]
+  // THREADS: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id=[[NESTED_PARALLEL_ID]], 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]]: 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]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/normal.c b/openmp/runtime/test/ompt/parallel/normal.c
new file mode 100644
index 00000000000..ca4af7eec62
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/normal.c
@@ -0,0 +1,54 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck --check-prefix=THREADS %s
+// REQUIRES: ompt, ompt-trace
+#include "callback.h"
+
+int main()
+{
+  #pragma omp parallel num_threads(4)
+  {
+  }
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // 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]]
+
+  // 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: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id={{[0-9]+}}, parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[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]]: 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: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // 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]]
+
+  return 0;
+}
diff --git a/openmp/runtime/test/ompt/parallel/serialized.c b/openmp/runtime/test/ompt/parallel/serialized.c
new file mode 100644
index 00000000000..6338e00e90a
--- /dev/null
+++ b/openmp/runtime/test/ompt/parallel/serialized.c
@@ -0,0 +1,19 @@
+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt, ompt-trace
+#include "callback.h"
+
+int main()
+{
+  #pragma omp parallel num_threads(1)
+  {
+  }
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=1, parallel_function=0x{{[0-f]+}}, invoker=[[PARALLEL_INVOKER:.+]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: 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]]
+
+  return 0;
+}