Newsgroups: comp.parallel.mpi From: Michael Pernice Subject: Re: data types Organization: Center for High Performance Computing Date: Mon, 01 Dec 1997 12:07:39 -0700 Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <34830AFB.98731C18@sneffels.usi.utah.edu> Glenn Sollie wrote: > > I am relatively new at MPI, and am trying to apply MPI to a three > dimensional CFD code. I need to send and receive planes of data for > boundary conditions, so I need to define vector datatypes for 3D arrays, > as outlined in appendix E.1.3 of "Using MPI". The description there > appears to be pretty straightforward, but I'm having a terrible time > making it work correctly, especially the procedure for passing a y-z > plane of data. Can anyone out there help? > > Thanks, > > gsollie@mindspring.com Glenn, I wrote some user-defined datatypes for passing faces in a 3D electromagnetic wave propagation code that seemed to work. There was one set of faces that was particularly tricky. I needed to define an intermediate type using MPI_TYPE_VECTOR to define rows that were constructed from non-contiguous data, and put together these rows to form the faces by using MPI_TYPE_HVECTOR. I needed to use MPI_TYPE_EXTENT to get the strides right. The particulars depend on your data structure, such as how you handle ghost cells for holding neighbor data and the number of ghost cells you need. If these hints aren't enough to get you started, I can send you a code fragment. Does anyone know a better way to do this? BTW, while writing this code, I ran across a tool at LLNL called MPIMap (http://www.llnl.gov/liv_comp/mpimap/) that's supposed to help with this process, but I could never get it to work. Anyone out there with experience using it? -- Michael Pernice usimap@sneffels.usi.utah.edu University of Utah Phone: (801) 581-7708 Center for High Performance Computing Fax: (801) 585-5366