int procs_cart_rank, procs_coords[2]; dims[0] = dims[1] = procs_size_sqrt; periods[0] = periods[1] = true; MPI_Cart_create(MPI_COMM_WORLD, 2, dims, periods, false, &cart_all); MPI_Comm_rank(cart_all, &procs_cart_rank); MPI_Cart_coords(cart_all, procs_cart_rank, 2, procs_coords); MPI_Comm_split(cart_all, procs_coords[0], procs_coords[1], &cart_row); MPI_Comm_split(cart_all, procs_coords[1], procs_coords[0], &cart_col); int rank_cart_row, rank_cart_col; MPI_Comm_rank(cart_row, & rank_cart_row); MPI_Comm_rank(cart_col, & rank_cart_col); // 计算并传递 for (int round = 0; round < procs_size_sqrt; ++ round) ...{ MPI_Isend(B, n_sqr, MPI_INT, (procs_coords[0] - 1 + procs_size_sqrt) % procs_size_sqrt, 1, cart_col, &reqSend); int broader = (round + procs_coords[0]) % procs_size_sqrt; if (broader == procs_coords[1]) std::copy(A,A+n_sqr,T); MPI_Bcast(T, n_sqr, MPI_INT, broader , cart_row); for (int row=0; row

} // 释放 MPI_Comm_free(&cart_col); MPI_Comm_free(&cart_row); MPI_Comm_free(&cart_all); delete []A; delete []B; delete []C; delete []T; end_time = MPI_Wtime(); MPI_Finalize(); printf("task %d consumed %lf seconds ", procs_id, end_time-start_time); return 0; }