pbdDMAT: 'pbdR' Distributed Matrix Methods

redistribute

R Documentation

Distribute/Redistribute matrices across the process grid

Description

Takes either an R matrix and distributes it as a distributed matrix, or takes a distributed matrix and redistributes it across a (possibly) new BLACS context, using a (possibly) new blocking dimension.

Usage

reblock(dx, bldim = dx@bldim, ICTXT = .pbd_env$ICTXT)

redistribute(dx, bldim = dx@bldim, ICTXT = .pbd_env$ICTXT)

Arguments

`dx`	numeric distributed matrix
`bldim`	the blocking dimension for block-cyclically distributing the matrix across the process grid.
`ICTXT`	BLACS context number for return.

Details

distribute() takes an R matrix x stored on the processes in some fashion and distributes it across the process grid belonging to ICTXT. If a process is to call distribute() and does not yet have any ownership of the matrix x, then that process should store NULL for x.

How one might typically use this is to read in a non-distributed matrix on the first process, store that result as the R matrix x, and then have the other processes store NULL for x. Then calling distribute() returns the distributed matrix which was distributed according to the options bldim and ICTXT.

Using an ICTXT value other than zero is not recommended unless you have a good reason to. Use of other such contexts should only be considered for advanced users, preferably those with knowledge of ScaLAPACK.

redistribute() takes a distributed matrix and redistributes it to the (possibly) new process grid with BLACS context ICTXT and with the (possibly) new blocking dimension bldim. The original BLACS context is dx@ICTXT and the original blocking dimension is dx@bldim.

These two functions are essentially simple wrappers for the ScaLAPACK function PDGEMR2D, with the above described behavior. Of note, for distribute(), dx@ICTXT and ICTXT must share at least one process in common. Likewise for redistribute() with xCTXT and ICTXT.

Very general redistributions can be done with redistribute(), but thinking in these terms is an acquired skill. For this reason, several simple interfaces to this function have been written.