A systematic partitioning approach for LS and SVD problems to fixed size arrays with constraints

Design methodologies which incorporate partitioning in the projection of an algorithm onto a physical array of processors are of theoretical and practical interest in systolic/wavefront designs. A systematic partitioning technique utilizing the integral matrix approach for designing fixed-size arrays with desired user-imposed constraints is presented. These include I/O and interconnection constraints to obtain efficient communication link properties. Applications to the QR algorithm for eigenvalue decomposition and least square problems are given