On Thu, Aug 11, 2011 at 03:33:16PM -0400, Izaak Beekman wrote: > I was wondering about collective H5 calls (and perhaps the definition of > collective MPI calls). Is the idea that every process in MPI_comm_world must > participate (i.e. EVERY available processor) , or every process in a given > communicator?
Collective routines take a communicator as one of the arguments: all processes in that communicator must participate. I/O routines are a little different, as the communicator was specified at open time, but again, all processes who participated in open need to participate in the collective i/o. similarly, in hdf5 you associate a communicator with an hdf5 file (H5Pset_fapl_mpio). Does not have to be MPI_COMM_WORLD. ==rob -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA _______________________________________________ Hdf-forum is for HDF software users discussion. [email protected] http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
