False communication-computation overlap (postpone-wait)

Version's name: False communication-computation overlap (postpone-wait) ; a version of the False communication-computation overlap program.
Repository: [home] and version downloads: [.zip] [.tar.gz] [.tar.bz2] [.tar]
Implemented best practices: Postpone the execution of non-blocking send waits operations ·

The postpone-wait version of the False communication-computation overlap kernel changes when the MPI_Waitall for the MPI_Isend is done. The call is postponed until the computation ends, succesfully overlapping the communication and the computation.

The following code snippet shows the change introduced to the code:

    for(n = 0; n < n_neighbours; n++){
        MPI_Irecv(r_buffer, rSIZE, ..., neighbours[n], ..., &irecv_req[n]);
        MPI_Isend(s_buffer, sSIZE, ..., neighbours[n], ..., &isend_req[n]);
        MPI_Waitall(n_neighbours, irecv_req, irecv_stat);
    MPI_Waitall(n_neighbours, isend_req, isend_stat); //This wait here!
The following experiments have been registered: