An algorithm for multiplication of concatenated matrices (Q911227)

The matrices to be multiplied on an \(m*m\) array of nearest neighbour type processors (i.e. each processor has links with the nearest four neighbours) are partitioned into \(m*m\) blocks. A parallel algorithm is shown whose efficiency for a large number of matrix factors tends to one and is therefore optimal. The algorithm uses relatively complex shifting operations and is demonstrated for a \(3*3\) partition at the end of the paper.