[Gerry Kan]

Howdy folks:

A bit of a question in MPI.

For a simple 2D code for a rectangular uniform grid, I have defined my own domain topology. That is, for each rank (kP), the corresponding (i, j) coordinates, problem size (ni x nj), grid offset (i0, j0) and its neighbors in the four principal directions (kE, kW, kN, kS) have been determined manually and stored in arrays.

Now I am thinking of replacing the part of this with MPI_Cart_* functions. I can see that, with this, I don't have to store the coordinate and neighbor information. However, I am still responsible for calculating the offset and problem size for the subdomains.

So this begs the question, whether it are other advantages (I could only think of potential use of MPI I/O, and perhaps some saving in memory) by using MPI_Cart* functions. If not I can only only justify this as being an "exercise" for using this feature.

Any opinions or comments are welcome.

Thanks in advance, Gerry.