Add OMPT support for teams construct

[hbae]

Summary

This change adds OMPT support for events from teams construct.

[protze.joachim]

The reported events look good to me.

runtime/src/kmp_csupport.cpp

if (ompt_enabled.enabled &&
 
     // reset clear the task id only after unlinking the task
     ompt_data_t *parent_task_data;
     __ompt_get_task_info_internal(1, NULL, &parent_task_data, NULL, NULL, NULL);
 
     if (ompt_enabled.ompt_callback_parallel_end) {
       ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
           &(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
-          ompt_parallel_invoker_program, OMPT_LOAD_RETURN_ADDRESS(global_tid));
+          ompt_parallel_invoker_program | ompt_parallel_team,
+          OMPT_LOAD_RETURN_ADDRESS(global_tid));
     }
     __ompt_lw_taskteam_unlink(this_thr);
     this_thr->th.ompt_thread_info.state = ompt_state_overhead;
   }
 #endif
 
   /* If necessary, pop the internal control stack values and replace the team
    * values */

runtime/src/kmp_runtime.cpp

if (ompt_enabled.enabled &&
     parent_task_info = OMPT_CUR_TASK_INFO(this_thr);
 
     parent_task_info->frame.enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
     if (ompt_enabled.ompt_callback_parallel_begin) {
       int team_size = 1;
 
       ompt_callbacks.ompt_callback(ompt_callback_parallel_begin)(
           &(parent_task_info->task_data), &(parent_task_info->frame),
-          &ompt_parallel_data, team_size, ompt_parallel_invoker_program,
-          codeptr);
+          &ompt_parallel_data, team_size,
+          ompt_parallel_invoker_program | ompt_parallel_team, codeptr);
     }
   }
 #endif // OMPT_SUPPORT
 
   if (this_thr->th.th_team != serial_team) {
     // Nested level will be an index in the nested nthreads array
     int level = this_thr->th.th_team->t.t_level;
 

 #endif
 
 #if OMPT_SUPPORT
     if (ompt_enabled.enabled) {
       if (ompt_enabled.ompt_callback_parallel_begin) {
         int team_size = master_set_numthreads
                             ? master_set_numthreads
                             : get__nproc_2(parent_team, master_tid);
+        int flags = OMPT_INVOKER(call_context) |
+                    ((microtask == (microtask_t)__kmp_teams_master)
+                         ? ompt_parallel_league
+                         : ompt_parallel_team);
         ompt_callbacks.ompt_callback(ompt_callback_parallel_begin)(
-            parent_task_data, ompt_frame, &ompt_parallel_data, team_size,
-            OMPT_INVOKER(call_context), return_address);
+            parent_task_data, ompt_frame, &ompt_parallel_data, team_size, flags,
+            return_address);
       }
       master_th->th.ompt_thread_info.state = ompt_state_overhead;
     }
 #endif
 
     master_th->th.th_ident = loc;
 
 #if OMP_40_ENABLED

 #else
         *argv++ = va_arg(ap, void *);
 #endif
       // Increment our nested depth levels, but not increase the serialization
       if (parent_team == master_th->th.th_serial_team) {
         // AC: we are in serialized parallel
         __kmpc_serialized_parallel(loc, gtid);
         KMP_DEBUG_ASSERT(parent_team->t.t_serialized > 1);
-        // AC: need this in order enquiry functions work
-        // correctly, will restore at join time
-        parent_team->t.t_serialized--;
+
 #if OMPT_SUPPORT
         void *dummy;
         void **exit_runtime_p;
 
         ompt_lw_taskteam_t lw_taskteam;
 
         if (ompt_enabled.enabled) {
           __ompt_lw_taskteam_init(&lw_taskteam, master_th, gtid,
                                   &ompt_parallel_data, return_address);
           exit_runtime_p = &(lw_taskteam.ompt_task_info.frame.exit_frame.ptr);
 
           __ompt_lw_taskteam_link(&lw_taskteam, master_th, 0);
           // don't use lw_taskteam after linking. content was swaped
 
           /* OMPT implicit task begin */
           implicit_task_data = OMPT_CUR_TASK_DATA(master_th);
           if (ompt_enabled.ompt_callback_implicit_task) {
-            ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
-                ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
-                implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
             OMPT_CUR_TASK_INFO(master_th)
                 ->thread_num = __kmp_tid_from_gtid(gtid);
+            ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+                ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
+                implicit_task_data, 1,
+                OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit);
           }
 
           /* OMPT state */
           master_th->th.ompt_thread_info.state = ompt_state_work_parallel;
         } else {
           exit_runtime_p = &dummy;
         }
 #endif
+        // AC: need to decrement t_serialized for enquiry functions to work
+        // correctly, will restore at join time
+        parent_team->t.t_serialized--;
 
         {
           KMP_TIME_PARTITIONED_BLOCK(OMP_parallel);
           KMP_SET_THREAD_STATE_BLOCK(IMPLICIT_TASK);
           __kmp_invoke_microtask(microtask, gtid, 0, argc, parent_team->t.t_argv
 #if OMPT_SUPPORT
                                  ,
                                  exit_runtime_p
 #endif
                                  );
         }
 
 #if OMPT_SUPPORT
         *exit_runtime_p = NULL;
         if (ompt_enabled.enabled) {
           OMPT_CUR_TASK_INFO(master_th)->frame.exit_frame = ompt_data_none;
           if (ompt_enabled.ompt_callback_implicit_task) {
             ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
                 ompt_scope_end, NULL, implicit_task_data, 1,
-                OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+                OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit);
           }
+          ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
           __ompt_lw_taskteam_unlink(master_th);
-
           if (ompt_enabled.ompt_callback_parallel_end) {
             ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
-                OMPT_CUR_TEAM_DATA(master_th), OMPT_CUR_TASK_DATA(master_th),
-                OMPT_INVOKER(call_context), return_address);
+                &ompt_parallel_data, OMPT_CUR_TASK_DATA(master_th),
+                OMPT_INVOKER(call_context) | ompt_parallel_team,
+                return_address);
           }
           master_th->th.ompt_thread_info.state = ompt_state_overhead;
         }
 #endif
         return TRUE;
       }
 
       parent_team->t.t_pkfn = microtask;
       parent_team->t.t_invoke = invoker;
       KMP_ATOMIC_INC(&root->r.r_in_parallel);
       parent_team->t.t_active_level++;
       parent_team->t.t_level++;
 #if OMP_50_ENABLED
       parent_team->t.t_def_allocator = master_th->th.th_def_allocator; // save
+
+#if OMPT_SUPPORT
+      if (ompt_enabled.enabled) {
+        ompt_lw_taskteam_t lw_taskteam;
+        __ompt_lw_taskteam_init(&lw_taskteam, master_th, gtid,
+                                &ompt_parallel_data, return_address);
+        __ompt_lw_taskteam_link(&lw_taskteam, master_th, 1, true);
+      }
+#endif
 #endif
 
       /* Change number of threads in the team if requested */
       if (master_set_numthreads) { // The parallel has num_threads clause
         if (master_set_numthreads < master_th->th.th_teams_size.nth) {
           // AC: only can reduce number of threads dynamically, can't increase
           kmp_info_t **other_threads = parent_team->t.t_threads;
           parent_team->t.t_nproc = master_set_numthreads;
#if OMPT_SUPPORT
                                     &ompt_parallel_data, return_address);
 
             __ompt_lw_taskteam_link(&lw_taskteam, master_th, 0);
             // don't use lw_taskteam after linking. content was swaped
 
             task_info = OMPT_CUR_TASK_INFO(master_th);
             exit_runtime_p = &(task_info->frame.exit_frame.ptr);
             if (ompt_enabled.ompt_callback_implicit_task) {
-              ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
-                  ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
-                  &(task_info->task_data), 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
               OMPT_CUR_TASK_INFO(master_th)
                   ->thread_num = __kmp_tid_from_gtid(gtid);
+              ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+                  ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
+                  &(task_info->task_data), 1,
+                  OMPT_CUR_TASK_INFO(master_th)->thread_num,
+                  ompt_task_implicit);
             }
 
             /* OMPT state */
             master_th->th.ompt_thread_info.state = ompt_state_work_parallel;
           } else {
             exit_runtime_p = &dummy;
           }
 #endif
#endif
           }
 
 #if OMPT_SUPPORT
           if (ompt_enabled.enabled) {
             exit_runtime_p = NULL;
             if (ompt_enabled.ompt_callback_implicit_task) {
               ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
                   ompt_scope_end, NULL, &(task_info->task_data), 1,
-                  OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+                  OMPT_CUR_TASK_INFO(master_th)->thread_num,
+                  ompt_task_implicit);
             }
-
+            ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
             __ompt_lw_taskteam_unlink(master_th);
             if (ompt_enabled.ompt_callback_parallel_end) {
               ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
-                  OMPT_CUR_TEAM_DATA(master_th), parent_task_data,
-                  OMPT_INVOKER(call_context), return_address);
+                  &ompt_parallel_data, parent_task_data,
+                  OMPT_INVOKER(call_context) | ompt_parallel_team,
+                  return_address);
             }
             master_th->th.ompt_thread_info.state = ompt_state_overhead;
           }
 #endif
         } else if (microtask == (microtask_t)__kmp_teams_master) {
           KMP_DEBUG_ASSERT(master_th->th.th_team ==
                            master_th->th.th_serial_team);
           team = master_th->th.th_team;
#endif
               // Get args from parent team for teams construct
               argv[i] = parent_team->t.t_argv[i];
           }
           // AC: revert change made in __kmpc_serialized_parallel()
           //     because initial code in teams should have level=0
           team->t.t_level--;
           // AC: call special invoker for outer "parallel" of teams construct
           invoker(gtid);
+#if OMPT_SUPPORT
+          if (ompt_enabled.enabled) {
+            ompt_task_info_t *task_info = OMPT_CUR_TASK_INFO(master_th);
+            if (ompt_enabled.ompt_callback_implicit_task) {
+              ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+                  ompt_scope_end, NULL, &(task_info->task_data), 0,
+                  OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_initial);
+            }
+            if (ompt_enabled.ompt_callback_parallel_end) {
+              ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
+                  &ompt_parallel_data, parent_task_data,
+                  OMPT_INVOKER(call_context) | ompt_parallel_league,
+                  return_address);
+            }
+            master_th->th.ompt_thread_info.state = ompt_state_overhead;
+          }
+#endif
         } else {
 #endif /* OMP_40_ENABLED */
           argv = args;
           for (i = argc - 1; i >= 0; --i)
 // TODO: revert workaround for Intel(R) 64 tracker #96
 #if (KMP_ARCH_X86_64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64) && KMP_OS_LINUX
             *argv++ = va_arg(*ap, void *);
 #else
#if OMPT_SUPPORT
             task_info = OMPT_CUR_TASK_INFO(master_th);
             exit_runtime_p = &(task_info->frame.exit_frame.ptr);
 
             /* OMPT implicit task begin */
             implicit_task_data = OMPT_CUR_TASK_DATA(master_th);
             if (ompt_enabled.ompt_callback_implicit_task) {
               ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
                   ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
-                  implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+                  implicit_task_data, 1, __kmp_tid_from_gtid(gtid),
+                  ompt_task_implicit);
               OMPT_CUR_TASK_INFO(master_th)
                   ->thread_num = __kmp_tid_from_gtid(gtid);
             }
 
             /* OMPT state */
             master_th->th.ompt_thread_info.state = ompt_state_work_parallel;
           } else {
             exit_runtime_p = &dummy;
#endif
           }
 
 #if OMPT_SUPPORT
           if (ompt_enabled.enabled) {
             *exit_runtime_p = NULL;
             if (ompt_enabled.ompt_callback_implicit_task) {
               ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
                   ompt_scope_end, NULL, &(task_info->task_data), 1,
-                  OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+                  OMPT_CUR_TASK_INFO(master_th)->thread_num,
+                  ompt_task_implicit);
             }
 
             ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
             __ompt_lw_taskteam_unlink(master_th);
             if (ompt_enabled.ompt_callback_parallel_end) {
               ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
                   &ompt_parallel_data, parent_task_data,
-                  OMPT_INVOKER(call_context), return_address);
+                  OMPT_INVOKER(call_context) | ompt_parallel_team,
+                  return_address);
             }
             master_th->th.ompt_thread_info.state = ompt_state_overhead;
           }
 #endif
 #if OMP_40_ENABLED
         }
 #endif /* OMP_40_ENABLED */
       } else if (call_context == fork_context_gnu) {
static inline void __kmp_join_restore_state(kmp_info_t *thread,
   // restore state outside the region
   thread->th.ompt_thread_info.state =
       ((team->t.t_serialized) ? ompt_state_work_serial
                               : ompt_state_work_parallel);
 }
 
 static inline void __kmp_join_ompt(int gtid, kmp_info_t *thread,
                                    kmp_team_t *team, ompt_data_t *parallel_data,
-                                   fork_context_e fork_context, void *codeptr) {
+                                   int flags, void *codeptr) {
   ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
   if (ompt_enabled.ompt_callback_parallel_end) {
     ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
-        parallel_data, &(task_info->task_data), OMPT_INVOKER(fork_context),
-        codeptr);
+        parallel_data, &(task_info->task_data), flags, codeptr);
   }
 
   task_info->frame.enter_frame = ompt_data_none;
   __kmp_join_restore_state(thread, team);
 }
 #endif
 
 void __kmp_join_call(ident_t *loc, int gtid
#endif /* OMP_40_ENABLED */
   master_th = __kmp_threads[gtid];
   root = master_th->th.th_root;
   team = master_th->th.th_team;
   parent_team = team->t.t_parent;
 
   master_th->th.th_ident = loc;
 
 #if OMPT_SUPPORT
+  void *team_microtask = (void *)team->t.t_pkfn;
   if (ompt_enabled.enabled) {
     master_th->th.ompt_thread_info.state = ompt_state_overhead;
   }
 #endif
 
 #if KMP_DEBUG
   if (__kmp_tasking_mode != tskm_immediate_exec && !exit_teams) {
     KA_TRACE(20, ("__kmp_join_call: T#%d, old team = %p old task_team = %p, "
else if ((__itt_frame_end_v3_ptr || KMP_ITT_DEBUG) &&
       __kmp_itt_region_joined(gtid);
   } // active_level == 1
 #endif /* USE_ITT_BUILD */
 
 #if OMP_40_ENABLED
   if (master_th->th.th_teams_microtask && !exit_teams &&
       team->t.t_pkfn != (microtask_t)__kmp_teams_master &&
       team->t.t_level == master_th->th.th_teams_level + 1) {
-    // AC: We need to leave the team structure intact at the end of parallel
-    // inside the teams construct, so that at the next parallel same (hot) team
-    // works, only adjust nesting levels
-
+#if OMPT_SUPPORT
+    ompt_data_t ompt_parallel_data = ompt_data_none;
+    if (ompt_enabled.enabled) {
+      ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
+      if (ompt_enabled.ompt_callback_implicit_task) {
+        int ompt_team_size = team->t.t_nproc;
+        ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+            ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size,
+            OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit);
+      }
+      task_info->frame.exit_frame = ompt_data_none;
+      task_info->task_data = ompt_data_none;
+      ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
+      __ompt_lw_taskteam_unlink(master_th);
+    }
+#endif
     /* Decrement our nested depth level */
     team->t.t_level--;
     team->t.t_active_level--;
     KMP_ATOMIC_DEC(&root->r.r_in_parallel);
 
     // Restore number of threads in the team if needed. This code relies on
     // the proper adjustment of th_teams_size.nth after the fork in
     // __kmp_teams_master on each teams master in the case that
#endif
           // Synchronize thread's task state
           other_threads[i]->th.th_task_state = master_th->th.th_task_state;
         }
       }
     }
 
 #if OMPT_SUPPORT
     if (ompt_enabled.enabled) {
-      __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, fork_context,
-                      codeptr);
+      __kmp_join_ompt(gtid, master_th, parent_team, &ompt_parallel_data,
+                      OMPT_INVOKER(fork_context) | ompt_parallel_team, codeptr);
     }
 #endif
 
     return;
   }
 #endif /* OMP_40_ENABLED */
 
   /* do cleanup and restore the parent team */
#endif /* OMP_40_ENABLED */
     KMP_ATOMIC_DEC(&root->r.r_in_parallel);
   }
   KMP_DEBUG_ASSERT(root->r.r_in_parallel >= 0);
 
 #if OMPT_SUPPORT
   if (ompt_enabled.enabled) {
     ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
     if (ompt_enabled.ompt_callback_implicit_task) {
-      int ompt_team_size = team->t.t_nproc;
+      int flags = (team_microtask == (void *)__kmp_teams_master)
+                      ? ompt_task_initial
+                      : ompt_task_implicit;
+      int ompt_team_size = (flags == ompt_task_initial) ? 0 : team->t.t_nproc;
       ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
           ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size,
-          OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+          OMPT_CUR_TASK_INFO(master_th)->thread_num, flags);
     }
-
     task_info->frame.exit_frame = ompt_data_none;
     task_info->task_data = ompt_data_none;
   }
 #endif
 
   KF_TRACE(10, ("__kmp_join_call1: T#%d, this_thread=%p team=%p\n", 0,
                 master_th, team));
   __kmp_pop_current_task_from_thread(master_th);
#endif
   // TODO: GEH - cannot do this assertion because root thread not set up as
   // executing
   // KMP_ASSERT( master_th->th.th_current_task->td_flags.executing == 0 );
   master_th->th.th_current_task->td_flags.executing = 1;
 
   __kmp_release_bootstrap_lock(&__kmp_forkjoin_lock);
 
 #if OMPT_SUPPORT
+  int flags =
+      OMPT_INVOKER(fork_context) |
+      ((team_microtask == (void *)__kmp_teams_master) ? ompt_parallel_league
+                                                      : ompt_parallel_team);
   if (ompt_enabled.enabled) {
-    __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, fork_context,
+    __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, flags,
                     codeptr);
   }
 #endif
 
   KMP_MB();
   KA_TRACE(20, ("__kmp_join_call: exit T#%d\n", gtid));
 }
 
#if OMPT_SUPPORT
   ompt_data_t *thread_data;
   if (ompt_enabled.enabled) {
     thread_data = &(this_thr->th.ompt_thread_info.thread_data);
     *thread_data = ompt_data_none;
 
     this_thr->th.ompt_thread_info.state = ompt_state_overhead;
     this_thr->th.ompt_thread_info.wait_id = 0;
     this_thr->th.ompt_thread_info.idle_frame = OMPT_GET_FRAME_ADDRESS(0);
+    this_thr->th.ompt_thread_info.parallel_flags = 0;
     if (ompt_enabled.ompt_callback_thread_begin) {
       ompt_callbacks.ompt_callback(ompt_callback_thread_begin)(
           ompt_thread_worker, thread_data);
     }
-  }
-#endif
-
-#if OMPT_SUPPORT
-  if (ompt_enabled.enabled) {
     this_thr->th.ompt_thread_info.state = ompt_state_idle;
   }
 #endif
+
   /* This is the place where threads wait for work */
   while (!TCR_4(__kmp_global.g.g_done)) {
     KMP_DEBUG_ASSERT(this_thr == __kmp_threads[gtid]);
     KMP_MB();
 
     /* wait for work to do */
     KA_TRACE(20, ("__kmp_launch_thread: T#%d waiting for work\n", gtid));
 
#if OMPT_SUPPORT
 
   my_task_data =
       &(team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data);
   my_parallel_data = &(team->t.ompt_team_info.parallel_data);
   if (ompt_enabled.ompt_callback_implicit_task) {
     ompt_team_size = team->t.t_nproc;
     ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
         ompt_scope_begin, my_parallel_data, my_task_data, ompt_team_size,
-        __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+        __kmp_tid_from_gtid(gtid), ompt_task_implicit);
     OMPT_CUR_TASK_INFO(this_thr)->thread_num = __kmp_tid_from_gtid(gtid);
   }
 #endif
 
 #if KMP_STATS_ENABLED
   stats_state_e previous_state = KMP_GET_THREAD_STATE();
   if (previous_state == stats_state_e::TEAMS_REGION) {
     KMP_PUSH_PARTITIONED_TIMER(OMP_teams);
   } else {
     KMP_PUSH_PARTITIONED_TIMER(OMP_parallel);
   }
   KMP_SET_THREAD_STATE(IMPLICIT_TASK);
 #endif
 
   rc = __kmp_invoke_microtask((microtask_t)TCR_SYNC_PTR(team->t.t_pkfn), gtid,
                               tid, (int)team->t.t_argc, (void **)team->t.t_argv
 #if OMPT_SUPPORT
                               ,
                               exit_runtime_p
 #endif
                               );
 #if OMPT_SUPPORT
   *exit_runtime_p = NULL;
+  this_thr->th.ompt_thread_info.parallel_flags |= ompt_parallel_team;
 #endif
 
 #if KMP_STATS_ENABLED
   if (previous_state == stats_state_e::TEAMS_REGION) {
     KMP_SET_THREAD_STATE(previous_state);
   }
   KMP_POP_PARTITIONED_TIMER();
 #endif
int __kmp_invoke_teams_master(int gtid) {
   kmp_info_t *this_thr = __kmp_threads[gtid];
   kmp_team_t *team = this_thr->th.th_team;
 #if KMP_DEBUG
   if (!__kmp_threads[gtid]->th.th_team->t.t_serialized)
     KMP_DEBUG_ASSERT((void *)__kmp_threads[gtid]->th.th_team->t.t_pkfn ==
                      (void *)__kmp_teams_master);
 #endif
   __kmp_run_before_invoked_task(gtid, 0, this_thr, team);
+#if OMPT_SUPPORT
+  int tid = __kmp_tid_from_gtid(gtid);
+  ompt_data_t *task_data =
+      &team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data;
+  ompt_data_t *parallel_data = &team->t.ompt_team_info.parallel_data;
+  if (ompt_enabled.ompt_callback_implicit_task) {
+    ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+        ompt_scope_begin, parallel_data, task_data, team->t.t_nproc, tid,
+        ompt_task_initial);
+    OMPT_CUR_TASK_INFO(this_thr)->thread_num = tid;
+  }
+#endif
   __kmp_teams_master(gtid);
+#if OMPT_SUPPORT
+  this_thr->th.ompt_thread_info.parallel_flags |= ompt_parallel_league;
+#endif
   __kmp_run_after_invoked_task(gtid, 0, this_thr, team);
   return 1;
 }
 #endif /* OMP_40_ENABLED */
 
 /* this sets the requested number of threads for the next parallel region
    encountered by this team. since this should be enclosed in the forkjoin
    critical section it should avoid race conditions with assymmetrical nested

runtime/src/kmp_wait_release.h

#if OMPT_OPTIONAL
     if (ompt_enabled.ompt_callback_sync_region) {
       ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
           ompt_sync_region_barrier_implicit, ompt_scope_end, NULL, tId,
           codeptr);
     }
 #endif
     if (!KMP_MASTER_TID(ds_tid)) {
       if (ompt_enabled.ompt_callback_implicit_task) {
+        int flags = this_thr->th.ompt_thread_info.parallel_flags;
+        flags = (flags & ompt_parallel_league) ? ompt_task_initial
+                                               : ompt_task_implicit;
         ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
-            ompt_scope_end, NULL, tId, 0, ds_tid, ompt_task_implicit);
+            ompt_scope_end, NULL, tId, 0, ds_tid, flags);
       }
       // return to idle state
       this_thr->th.ompt_thread_info.state = ompt_state_idle;
     } else {
       this_thr->th.ompt_thread_info.state = ompt_state_overhead;
     }
   }
 }

runtime/src/ompt-internal.h

 typedef struct {
   ompt_data_t thread_data;
   ompt_data_t task_data; /* stored here from implicit barrier-begin until
                             implicit-task-end */
   void *return_address; /* stored here on entry of runtime */
   ompt_state_t state;
   ompt_wait_id_t wait_id;
   int ompt_task_yielded;
+  int parallel_flags; // information for the last parallel region invoked
   void *idle_frame;
 } ompt_thread_info_t;
 
 extern ompt_callbacks_internal_t ompt_callbacks;
 
 #if OMP_40_ENABLED && OMPT_SUPPORT && OMPT_OPTIONAL
 #if USE_FAST_MEMORY
 #define KMP_OMPT_DEPS_ALLOC __kmp_fast_allocate

runtime/src/ompt-specific.h

 
 void __ompt_team_assign_id(kmp_team_t *team, ompt_data_t ompt_pid);
 void __ompt_thread_assign_wait_id(void *variable);
 
 void __ompt_lw_taskteam_init(ompt_lw_taskteam_t *lwt, kmp_info_t *thr,
                              int gtid, ompt_data_t *ompt_pid, void *codeptr);
 
 void __ompt_lw_taskteam_link(ompt_lw_taskteam_t *lwt, kmp_info_t *thr,
-                             int on_heap);
+                             int on_heap, bool always = false);
 
 void __ompt_lw_taskteam_unlink(kmp_info_t *thr);
 
 ompt_team_info_t *__ompt_get_teaminfo(int depth, int *size);
 
 ompt_task_info_t *__ompt_get_task_info_object(int depth);
 
 int __ompt_get_parallel_info_internal(int ancestor_level,

runtime/src/ompt-specific.cpp

void __ompt_lw_taskteam_init(ompt_lw_taskteam_t *lwt, kmp_info_t *thr, int gtid,
   lwt->ompt_task_info.scheduling_parent = NULL;
   lwt->ompt_task_info.deps = NULL;
   lwt->ompt_task_info.ndeps = 0;
   lwt->heap = 0;
   lwt->parent = 0;
 }
 
 void __ompt_lw_taskteam_link(ompt_lw_taskteam_t *lwt, kmp_info_t *thr,
-                             int on_heap) {
+                             int on_heap, bool always) {
   ompt_lw_taskteam_t *link_lwt = lwt;
-  if (thr->th.th_team->t.t_serialized >
+  if (always ||
+      thr->th.th_team->t.t_serialized >
           1) { // we already have a team, so link the new team and swap values
     if (on_heap) { // the lw_taskteam cannot stay on stack, allocate it on heap
       link_lwt =
           (ompt_lw_taskteam_t *)__kmp_allocate(sizeof(ompt_lw_taskteam_t));
     }
     link_lwt->heap = on_heap;
 
     // would be swap in the (on_stack) case.
     ompt_team_info_t tmp_team = lwt->ompt_team_info;

runtime/test/ompt/callback.h

case ompt_scope_begin:
 
       //there is no parallel_begin callback for implicit parallel region
       //thus it is initialized in initial task
       if(flags & ompt_task_initial)
       {
         char buffer[2048];
 
         format_task_type(flags, buffer);
-        if(parallel_data->ptr)
+        // Only check initial task not created by teams construct
+        if (team_size == 1 && thread_num == 1 && parallel_data->ptr)
           printf("%s\n", "0: parallel_data initially not null");
         parallel_data->value = ompt_get_unique_id();
         printf("%" PRIu64 ": ompt_event_initial_task_begin: parallel_id=%" PRIu64 ", task_id=%" PRIu64 ", actual_parallelism=%" PRIu32 ", index=%" PRIu32 ", flags=%" PRIu32 "\n", ompt_get_thread_data()->value, parallel_data->value, task_data->value, team_size, thread_num, flags);
       } else {
         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:
       if(flags & ompt_task_initial){
-        printf("%" PRIu64 ": ompt_event_initial_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);
+        printf("%" PRIu64 ": ompt_event_initial_task_end: parallel_id=%" PRIu64
+               ", task_id=%" PRIu64 ", actual_parallelism=%" PRIu32
+               ", index=%" PRIu32 "\n",
+               ompt_get_thread_data()->value,
+               (parallel_data) ? parallel_data->value : 0, task_data->value,
+               team_size, thread_num);
       } else {
         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

 
 static void on_ompt_callback_parallel_begin(
     ompt_data_t *encountering_task_data,
     const ompt_frame_t *encountering_task_frame, ompt_data_t *parallel_data,
     uint32_t requested_team_size, int flag, const void *codeptr_ra) {
   if(parallel_data->ptr)
     printf("0: parallel_data initially not null\n");
   parallel_data->value = ompt_get_unique_id();
-  printf("%" PRIu64 ": ompt_event_parallel_begin: parent_task_id=%" PRIu64
+  int invoker = flag & 0xF;
+  const char *event = (flag & ompt_parallel_team) ? "parallel" : "teams";
+  const char *size = (flag & ompt_parallel_team) ? "team_size" : "num_teams";
+  printf("%" PRIu64 ": ompt_event_%s_begin: parent_task_id=%" PRIu64
          ", parent_task_frame.exit=%p, parent_task_frame.reenter=%p, "
-         "parallel_id=%" PRIu64 ", requested_team_size=%" PRIu32
+         "parallel_id=%" PRIu64 ", requested_%s=%" PRIu32
          ", codeptr_ra=%p, invoker=%d\n",
-         ompt_get_thread_data()->value, encountering_task_data->value,
+         ompt_get_thread_data()->value, event, encountering_task_data->value,
          encountering_task_frame->exit_frame.ptr,
-         encountering_task_frame->enter_frame.ptr, parallel_data->value,
-         requested_team_size, codeptr_ra, flag);
+         encountering_task_frame->enter_frame.ptr, parallel_data->value, size,
+         requested_team_size, codeptr_ra, invoker);
 }
 
 static void on_ompt_callback_parallel_end(ompt_data_t *parallel_data,
                                           ompt_data_t *encountering_task_data,
                                           int flag, const void *codeptr_ra) {
-  printf("%" PRIu64 ": ompt_event_parallel_end: parallel_id=%" PRIu64
+  int invoker = flag & 0xF;
+  const char *event = (flag & ompt_parallel_team) ? "parallel" : "teams";
+  printf("%" PRIu64 ": ompt_event_%s_end: parallel_id=%" PRIu64
          ", task_id=%" PRIu64 ", invoker=%d, codeptr_ra=%p\n",
-         ompt_get_thread_data()->value, parallel_data->value,
-         encountering_task_data->value, flag, codeptr_ra);
+         ompt_get_thread_data()->value, event, parallel_data->value,
+         encountering_task_data->value, invoker, codeptr_ra);
 }
 
 static void
 on_ompt_callback_task_create(
     ompt_data_t *encountering_task_data,
     const ompt_frame_t *encountering_task_frame,
     ompt_data_t* new_task_data,
     int type,

runtime/test/ompt/misc/interoperability.cpp

 // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
 
 // first master thread
 // CHECK: {{^}}[[MASTER_ID_1:[0-9]+]]: ompt_event_thread_begin:
 // CHECK-SAME: thread_type=ompt_thread_initial=1, thread_id=[[MASTER_ID_1]]
 
 
 // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_initial_task_begin: parallel_id={{[0-9]+}}
-// CHECK-SAME: task_id=[[PARENT_TASK_ID_1:[0-9]+]], actual_parallelism=1, index=1, flags=1 
+// CHECK-SAME: task_id=[[PARENT_TASK_ID_1:[0-9]+]], actual_parallelism=1,
+// CHECK-SAME: index=1, flags=1
 
 // CHECK: {{^}}[[MASTER_ID_1]]: ompt_event_parallel_begin:
 // CHECK-SAME: parent_task_id=[[PARENT_TASK_ID_1]]
 // CHECK-SAME: parent_task_frame.exit=[[NULL]]
 // CHECK-SAME: parent_task_frame.reenter={{0x[0-f]+}}
 // CHECK-SAME: parallel_id=[[PARALLEL_ID_1:[0-9]+]], requested_team_size=2
 // CHECK-SAME: codeptr_ra=0x{{[0-f]+}}, invoker={{.*}}
 
 // CHECK: {{^}}[[MASTER_ID_1]]: ompt_event_parallel_end:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID_1]], task_id=[[PARENT_TASK_ID_1]]
 // CHECK-SAME: invoker={{[0-9]+}}
 
 // CHECK: {{^}}[[MASTER_ID_1]]: ompt_event_initial_task_end:
 // CHECK-SAME: parallel_id={{[0-9]+}}, task_id=[[PARENT_TASK_ID_1]],
-// CHECK-SAME: team_size=0, thread_num=1
+// CHECK-SAME: actual_parallelism=0, index=1
 
 // CHECK: {{^}}[[MASTER_ID_1]]: ompt_event_thread_end:
 // CHECK-SAME: thread_id=[[MASTER_ID_1]]
 
 // second master thread
 // CHECK: {{^}}[[MASTER_ID_2:[0-9]+]]: ompt_event_thread_begin:
 // CHECK-SAME: thread_type=ompt_thread_initial=1, thread_id=[[MASTER_ID_2]]
 
 // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_initial_task_begin: parallel_id={{[0-9]+}}
-// CHECK-SAME: task_id=[[PARENT_TASK_ID_2:[0-9]+]], actual_parallelism=1, index=1, flags=1 
+// CHECK-SAME: task_id=[[PARENT_TASK_ID_2:[0-9]+]], actual_parallelism=1,
+// CHECK-SAME: index=1, flags=1
 
 // CHECK: {{^}}[[MASTER_ID_2]]: ompt_event_parallel_begin:
 // CHECK-SAME: parent_task_id=[[PARENT_TASK_ID_2]]
 // CHECK-SAME: parent_task_frame.exit=[[NULL]]
 // CHECK-SAME: parent_task_frame.reenter={{0x[0-f]+}}
 // CHECK-SAME: parallel_id=[[PARALLEL_ID_2:[0-9]+]]
 // CHECK-SAME: requested_team_size=2, codeptr_ra=0x{{[0-f]+}}
 // CHECK-SAME: invoker={{.*}}
 
 // CHECK: {{^}}[[MASTER_ID_2]]: ompt_event_parallel_end:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID_2]], task_id=[[PARENT_TASK_ID_2]]
 // CHECK-SAME: invoker={{[0-9]+}}
 
 // CHECK: {{^}}[[MASTER_ID_2]]: ompt_event_initial_task_end:
 // CHECK-SAME: parallel_id={{[0-9]+}}, task_id=[[PARENT_TASK_ID_2]],
-// CHECK-SAME: team_size=0, thread_num=1
+// CHECK-SAME: actual_parallelism=0, index=1
 
 // CHECK: {{^}}[[MASTER_ID_2]]: ompt_event_thread_end:
 // CHECK-SAME: thread_id=[[MASTER_ID_2]]
 
 // first worker thread
 // CHECK: {{^}}[[THREAD_ID_1:[0-9]+]]: ompt_event_thread_begin:
 // CHECK-SAME: thread_type=ompt_thread_worker=2, thread_id=[[THREAD_ID_1]]
 // CHECK-NOT: {{^}}[[THREAD_ID_1:[0-9]+]]: ompt_event_initial_task_end:

runtime/test/ompt/teams/parallel_team.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// UNSUPPORTED: gcc-4, gcc-5, gcc-6, gcc-7
+#include "callback.h"
+
+int main() {
+#pragma omp target teams num_teams(1) thread_limit(2)
+#pragma omp parallel num_threads(2)
+  { printf("In teams\n"); }
+  return 0;
+}
+
+// CHECK: 0: NULL_POINTER=[[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:[0-9]+]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK:[0-9]+]], {{.*}}, index=1
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_teams_begin:
+// CHECK-SAME: parent_task_id=[[INIT_TASK]]
+// CHECK-SAME: {{.*}} requested_num_teams=1
+// CHECK-SAME: {{.*}} invoker=[[TEAMS_FLAGS:[0-9]+]]
+
+//
+// team 0/thread 0
+//
+// initial task in the teams construct
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK_0:[0-9]+]], actual_parallelism=1, index=0
+
+// parallel region forked by runtime
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[INIT_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0:[0-9]+]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[IMPL_TASK_0:[0-9]+]]
+
+// user parallel region
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[IMPL_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00:[0-9]+]]
+// CHECK-SAME: {{.*}} requested_team_size=2
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00]], task_id=[[IMPL_TASK_00:[0-9]+]]
+// CHECK-SAME: {{.*}} team_size=2, thread_num=0
+//
+// barrier event is here
+//
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_00]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00]], task_id=[[IMPL_TASK_0]]
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_0]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[INIT_TASK_0]]
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK_0]], actual_parallelism=0, index=0
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_teams_end:
+// CHECK-SAME: {{.*}} task_id=[[INIT_TASK]], invoker=[[TEAMS_FLAGS]]
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK]], {{.*}}, index=1
+
+//
+// team 0/thread 1
+//
+// CHECK: {{^}}[[WORKER:[0-9]+]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00]], task_id=[[IMPL_TASK_01:[0-9]+]]
+// CHECK-SAME: {{.*}} team_size=2, thread_num=1
+//
+// barrier event is here
+//
+// CHECK: {{^}}[[WORKER]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_01]]

runtime/test/ompt/teams/serial_teams.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+// UNSUPPORTED: gcc-4, gcc-5, gcc-6, gcc-7
+#include "callback.h"
+
+int main() {
+#pragma omp target teams num_teams(2) thread_limit(1)
+#pragma omp parallel num_threads(1)
+  { printf("In teams parallel\n"); }
+  return 0;
+}
+
+// CHECK: 0: NULL_POINTER=[[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_0:[0-9]+]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK:[0-9]+]], {{.*}}, index=1
+
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_teams_begin:
+// CHECK-SAME: parent_task_id=[[INIT_TASK]]
+// CHECK-SAME: {{.*}} requested_num_teams=2
+// CHECK-SAME: {{.*}} invoker=[[TEAMS_FLAGS:[0-9]+]]
+
+//
+// team 0
+//
+// initial task in the teams construct
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK_0:[0-9]+]], actual_parallelism=2, index=0
+
+// parallel region forked by runtime
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[INIT_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0:[0-9]+]]
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[IMPL_TASK_0:[0-9]+]]
+
+// user parallel region
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[IMPL_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00:[0-9]+]]
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00]], task_id=[[IMPL_TASK_0]]
+
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_0]]
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[INIT_TASK_0]]
+
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK_0]], actual_parallelism=0, index=0
+
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_teams_end:
+// CHECK-SAME: {{.*}} task_id=[[INIT_TASK]], invoker=[[TEAMS_FLAGS]]
+
+// CHECK: {{^}}[[MASTER_0]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK]], {{.*}}, index=1
+
+//
+// team 1
+//
+// initial task in the teams construct
+// CHECK: {{^}}[[MASTER_1:[0-9]+]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK_1:[0-9]+]], actual_parallelism=2, index=1
+
+// parallel region forked by runtime
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[INIT_TASK_1]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_ID_1:[0-9]+]]
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_ID_1]], task_id=[[IMPL_TASK_1:[0-9]+]]
+
+// user parallel region
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[IMPL_TASK_1]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_ID_11:[0-9]+]]
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_ID_11]], task_id=[[IMPL_TASK_1]]
+
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_1]]
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_ID_1]], task_id=[[INIT_TASK_1]]
+
+// CHECK: {{^}}[[MASTER_1]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK_1]], actual_parallelism=0, index=1

runtime/test/ompt/teams/serialized.c

+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+// UNSUPPORTED: gcc-4, gcc-5, gcc-6, gcc-7
+#include "callback.h"
+
+int main() {
+#pragma omp target teams num_teams(1) thread_limit(1)
+#pragma omp parallel num_threads(1)
+  { printf("In teams\n"); }
+  return 0;
+}
+
+// CHECK: 0: NULL_POINTER=[[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:[0-9]+]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK:[0-9]+]], {{.*}}, index=1
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_teams_begin:
+// CHECK-SAME: parent_task_id=[[INIT_TASK]]
+// CHECK-SAME: {{.*}} requested_num_teams=1
+// CHECK-SAME: {{.*}} invoker=[[TEAMS_FLAGS:[0-9]+]]
+
+// initial task in the teams construct starts
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK_0:[0-9]+]], actual_parallelism=1, index=0
+
+// parallel region forked by runtime
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[INIT_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0:[0-9]+]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[IMPL_TASK_0:[0-9]+]]
+
+// user parallel region
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[IMPL_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00:[0-9]+]]
+// CHECK-SAME: {{.*}} requested_team_size=1
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00]], task_id=[[IMPL_TASK_00:[0-9]+]]
+// CHECK-SAME: {{.*}} team_size=1, thread_num=0
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_00]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_00]], task_id=[[IMPL_TASK_0]]
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[INIT_TASK_0]]
+
+// initial task in the teams construct ends
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK_0]], actual_parallelism=0, index=0
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_teams_end:
+// CHECK-SAME: {{.*}} task_id=[[INIT_TASK]], invoker=[[TEAMS_FLAGS]]
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK]], {{.*}}, index=1

runtime/test/ompt/teams/team.c

+// RUN: %libomp-compile-and-run | FileCheck %s
+// REQUIRES: ompt
+// UNSUPPORTED: gcc-4, gcc-5, gcc-6, gcc-7
+#include "callback.h"
+
+int main() {
+#pragma omp target teams num_teams(1) thread_limit(1)
+  { printf("In teams\n"); }
+  return 0;
+}
+
+// CHECK: 0: NULL_POINTER=[[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:[0-9]+]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK:[0-9]+]], {{.*}}, index=1
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_teams_begin:
+// CHECK-SAME: parent_task_id=[[INIT_TASK]]
+// CHECK-SAME: {{.*}} requested_num_teams=1
+// CHECK-SAME: {{.*}} invoker=[[TEAMS_FLAGS:[0-9]+]]
+
+// initial task in the teams construct starts
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_begin:
+// CHECK-SAME: task_id=[[INIT_TASK_0:[0-9]+]], actual_parallelism=1, index=0
+
+// parallel region forked by runtime
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_begin:
+// CHECK-SAME: {{.*}} parent_task_id=[[INIT_TASK_0]]
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0:[0-9]+]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_begin:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[IMPL_TASK_0:[0-9]+]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_implicit_task_end:
+// CHECK-SAME: {{.*}} parallel_id={{[0-9]+}}, task_id=[[IMPL_TASK_0]]
+// CHECK: {{^}}[[MASTER]]: ompt_event_parallel_end:
+// CHECK-SAME: {{.*}} parallel_id=[[PAR_0]], task_id=[[INIT_TASK_0]]
+
+// initial task in the teams construct ends
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK_0]], actual_parallelism=0, index=0
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_teams_end:
+// CHECK-SAME: {{.*}} task_id=[[INIT_TASK]], invoker=[[TEAMS_FLAGS]]
+
+// CHECK: {{^}}[[MASTER]]: ompt_event_initial_task_end:
+// CHECK-SAME: task_id=[[INIT_TASK]], {{.*}}, index=1