Subroutines to exchange boundary data

Current implementations provide a single routine to update the ghost boundaries for a distributed array:

        subroutine mesh_update_bdry(array)
        real array(....)

This subroutine updates the ghost boundaries of array (by exchanging boundary information with neighboring processes). The corners of the ghost boundaries are updated, but ghost boundaries that correspond to the boundaries of the global array are not. (Observe that the width of the ghost boundaries -- i.e., the width of the boundary to exchange -- is specified by the user via PARAMETER NGHOST.)

NOTE: Some implementations also include alternative boundary-exchange operations, including:

• A boundary exchange operation that does not update the corners of the ghost section -- this is potentially somewhat faster.
• A boundary exchange operation that treats the grid as a torus rather than a mesh -- i.e., fills in ghost boundaries for processes on the boundaries of the process grid with values from processes on the opposite border. (E.g., the left ghost boundary of a process on the left edge of the process grid contains values from the right boundary of the corresponding process on the right edge of the process grid.)

Refer to comments in the implementation code for details.