[OpenMP][OMPT] Fix memory leak when using serialized GCC compatibility code

[jlpeyton]

Summary

Serialized parallels allocate lightweight task teams on the heap
but never free them in the corresponding join. This patch adds a wrapper
around the allocation (if ompt enabled) and also adds the corresponding
free in the join call.

[protze.joachim]

Lgtm

[jlpeyton]

My mistake here. Uploading diff with full context for the record.

openmp/runtime/src/kmp_runtime.cpp

#if OMPT_SUPPORT
                   return_address);
             }
             master_th->th.ompt_thread_info.state = ompt_state_overhead;
           }
 #endif
         }
       } else if (call_context == fork_context_gnu) {
 #if OMPT_SUPPORT
+        if (ompt_enabled.enabled) {
           ompt_lw_taskteam_t lwt;
           __ompt_lw_taskteam_init(&lwt, master_th, gtid, &ompt_parallel_data,
                                   return_address);
 
           lwt.ompt_task_info.frame.exit_frame = ompt_data_none;
           __ompt_lw_taskteam_link(&lwt, master_th, 1);
+        }
 // don't use lw_taskteam after linking. content was swaped
 #endif
 
         // we were called from GNU native code
         KA_TRACE(20, ("__kmp_fork_call: T#%d serial exit\n", gtid));
         return FALSE;
       } else {
         KMP_ASSERT2(call_context < fork_context_last,
if (master_th->th.th_teams_microtask) {
         // __kmpc_end_serialized_parallel
         team->t.t_serialized++;
       }
     }
     __kmpc_end_serialized_parallel(loc, gtid);
 
 #if OMPT_SUPPORT
     if (ompt_enabled.enabled) {
+      if (fork_context == fork_context_gnu) {
+        __ompt_lw_taskteam_unlink(master_th);
+      }
       __kmp_join_restore_state(master_th, parent_team);
     }
 #endif
 
     return;
   }
 
   master_active = team->t.t_master_active;