This MR cleans up the CUDA communication module, i.e., follow best practice approaches and use unambiguous statements.