now that we have the unexpected message queue for unknown communicators, there is no need to have this additional synchronization operation for multi-threaded communicator creations.

This commit was SVN r20046.

ompi/communicator/comm_cid.c

@@ -231,13 +231,6 @@ int ompi_comm_nextcid ( ompi_communicator_t* newcomm,
         newcomm->c_f_to_c_index = newcomm->c_contextid;
         opal_pointer_array_set_item (&ompi_mpi_communicators, nextcid, newcomm);
         
-        /* for synchronization purposes, avoids receiving fragments for 
-           a communicator id, which might not yet been known. For single-threaded
-           scenarios, this call is in ompi_comm_activate, for multi-threaded
-           scenarios, it has to be already here ( before releasing another
-           thread into the cid-allocation loop ) */
-        (allredfnct)(&response, &glresponse, 1, MPI_MIN, comm, bridgecomm,
-                     local_leader, remote_leader, send_first );
         OPAL_THREAD_LOCK(&ompi_cid_lock);
         ompi_comm_unregister_cid (comm->c_contextid);
         OPAL_THREAD_UNLOCK(&ompi_cid_lock);