I've added this and posted an update to the wiki. There are a few bugs in this; please check the wiki version. Bill On Apr 19, 2011, at 4:31 PM, James Dinan wrote:
Hi All,
Example is below, please give any feedback. I'll work on compressing this down the match the format in the writeup.
Caption: The following example shows how request-based operations can be used to overlap communication with computation. Each process fetches, processes, and writes the result for $NSTEPS$ chunks of data. Instead of a single buffer, $M$ local buffers are used to allow up to $M$ communication operations to overlap with computation.
~Jim.
---snip---
int i, j; MPI_Window win; MPI_Request put_req[M] = { MPI_REQUEST_NULL }; MPI_Request get_req; double data[M][N];
/* Create win: size M*N*sizeof(double), displacement unit MPI_DOUBLE */
MPI_Win_lock_all(0, win);
for (i = 0; i < NSTEPS; i++) { MPI_Waitany(M, put_req, &j, MPI_STATUS_IGNORE); MPI_RGet(data[j], N, MPI_DOUBLE, target, i*N, N, MPI_DOUBLE, win, &get_req); MPI_Wait(get_req); compute(i, data[M], ...); MPI_RPut(data[j], N, MPI_DOUBLE, target, i*N, N, MPI_DOUBLE, win, &R[j]); }
MPI_Waitall(M, put_req, MPI_STATUSES_IGNORE); MPI_Win_unlock_all(win); _______________________________________________ mpi3-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign