MPI_Cart_shift

MPI_Cart_shift computes the ranks of neighboring processes.

C

 int MPI_Cart_shift(MPI_Comm comm, int direction, int disp,
                    int *rank_source, int *rank_dest)

Fortran

MPI_CART_SHIFT(COMM, DIRECTION, DISP, RANK_SOURCE, RANK_DEST, IERROR)
    INTEGER COMM, DIRECTION, DISP, RANK_SOURCE, RANK_DEST, IERROR

This routine does not actually shift data. Instead, it returns the correct ranks for a shift which can be made in subsequent communication calls. If the shift would put some coordinate out of the topology, MPI_PROC_NULL is returned for the source and/or destination.

Arguments:

direction

dimension in which the shift should be made

disp

size of the shift in processor coordinates (+ or -)

rank_source

rank from which the caller should receive a message

rank_dest

rank to which the caller should send a message

MPI_Cart_sub

MPI_Cart_sub partitions a Cartesian process grid.

C

int MPI_Cart_sub(MPI_Comm comm, int *remain_dims, MPI_Comm *newcomm)

Fortran

MPI_CART_SUB(COMM, REMIAN_DIMS, NEWCOMM, IERROR)
    INTEGER COMM, NEWCOMM, IERROR
    LOGICAL REMAIN_DIMS(*)

Often one wants to do an operation only on a slice of an existing Cartesian topology.

• MPI_Cart_sub generates a new communicator for each slice.
• Each slice can then perform its own collective communications.
• Example: if comm is a 2x3x4 grid, and remain_dims={TRUE,FALSE,TRUE}, then three new communicators are created, each being a 2x4 grid.
• The calling process receives back only the new communicator it is in.