Author(s)

O.B. Efremides and M.P. Bekakos

Abstract

Processor-time optimal systolic matrix multiplication O.B. Efremides & M.P. Bekakos Department of Informatics, Athens University of Economics and Business, 104 $4 Athens Abstract The systolic algorithm implementation on processor arrays using the data dependence method is based on the linear transformations applied on the data dependencies and on the index set the algorithm. The implementation obtained by using this method has a regular data flow, but still it is not possible for processor-time optimal Systolic .Arrays (SAs) to be synthesized. Herein, a new approach is introduced to synthesize processor-time optimal SAs for the matrix multiplication problem. The proposed procedure is based on the dependence method, but nonlinear transformations and the concept of the double processing streams for the input data of the algorithm are also utilized. 1 Introduction One of the most challenging problems in systolic processing is the develop- ment of a methodology for mapping algorithms in

Keywords