A general approach to solving arbitrarily large problems in a fixed size systolic array

Solving large problems in a small systolic array is considered difficult because the data arrays that pipeline through the array have to meet at the right place at the right time. Thus, partitioning and scheduling of data arrays for systolic computation is generally performed ad hoc. A general partitioning and scheduling method for solving large problems in a small systolic array is presented. Given a systolic array and its algorithm, a scheduling program that schedules the data blocks resulting from partitioning the data arrays can be obtained in a straightforward manner from the algorithm. The method, which can be easily mechanized, is applicable to loop algorithms and systolic arrays of various structures and shapes with or without local memory. Array emulation is an effective approach to solve large problems in a small processor array efficiently. Various array emulation methods are discussed, and the partitioning and scheduling method presented is shown to be useful for array emulation also.<>