Improve MPI_Waitall performance for MPI_THREAD_MULTIPLE

Avoid atomic cmpxchng operations for MPI requests that are already
complete. This improves the performance in message rate benchmarks.

Signed-off-by: Austen Lauria <awlauria@us.ibm.com>

Author: nysal

ompi/request/req_wait.c

@@ -243,7 +243,7 @@ int ompi_request_default_wait_all( size_t count,
             continue;
         }
 
-        if (!OPAL_ATOMIC_COMPARE_EXCHANGE_STRONG_PTR(&request->req_complete, &_tmp_ptr, &sync)) {
+        if (REQUEST_COMPLETE(request) || !OPAL_ATOMIC_COMPARE_EXCHANGE_STRONG_PTR(&request->req_complete, &_tmp_ptr, &sync)) {
             if( OPAL_LIKELY( REQUEST_COMPLETE(request) ) ) {
                 if( OPAL_UNLIKELY( MPI_SUCCESS != request->req_status.MPI_ERROR ) ) {
                     failed++;