Implementing the generalized matrix product on a systolic array parallel architecture

The generalized matrix product includes in its formulation many common array manipulations. It also provides a framework for the expression of a number of important image processing algorithms. It is shown that the generalized matrix product may be implemented in its full generality on systolic array architectures. Two approaches are presented. One approach is to regard the generalized matrix product as a collection of products of small matrices and then consider arrangements of systolic configurations common to the smaller products. A second approach is to embed the two factors of the generalized matrix product in sparse matrices and multiply the sparse matrices using a conventional systolic array.