Reciprocative multirelation arrays

A processor array architecture based on a novel dataflow scheme is reported. It is shown that the reciprocative multirelation arrays (RMA) dataflow provides optimal efficiency of processor utilization. An abstract framework is used as a description of the general principle, and different operational units are 'plugged' into the framework to implement different algorithms. Suitable applications are in the area of symbolic computing. Details of an RMA implementation of a sorter are presented, where the advantages are very high processor utilization, simple and localized control, pipelined execution of data sets without any significant control overhead, and a modular and compact design.<
>