Comment (by Roland Haas):

Looking at the omp single and in particular its implied omp barrier at the end, I suspect the current code may also run the risk of creating a deadlock when the scheduled functions are called in a multi-threaded way in CarpetX. Namely since the barier assumes that all threads will encounter it, there would be a deadlock if the the number of tiles processed by each thread is not identical to that of any of the other threads. A possible workaround seems to be to remove the barrier by using nowait which however makes copyprivate impossible requiring that the cctki_bndsize arrays be static so that they are shared among the threads.

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2446/cactus-new-cgh-fields-tile_min-tile_max