LOCAL INTERCONNECTION SCHEME FOR PARALLEL PROCESSING ARCHITECTURES

Method and apparatus for interconnecting the processing cells of a parallel processing machine configured in a two-dimensional rectangular array in which each cell includes a plurality of ports, each port having a unique port address, and a plurality of cells have similarly addressed ports. The cells are interconnected, via the cell ports, to form cell clusters having a central cell and eight neighboring cells such that a plurality of neighboring cells share a common connection to the central cell and further such that no two similarly addressed ports are coupled to one another. During a data transfer operation, in accordance with the single instruction multiple data (SIMD) format, each cell transmits data from one port and receives data from another port such that all cells transmit data from similarly addressed ports and receive data at similarly addressed ports to provide data transfer throughout the array in a uniform direction.