Automatic synthesis of systolic arrays from uniform recurrent equations

摘要: We describe a systematic method for the design of systolic arrays. This may be used algorithms that can expressed as set uniform recurrent equations over convex D Cartesian coordinates. Most already considered implementation represented in this way. The methods consists two steps: finding timing-function computations is compatible with dependences introduced by equations, then mapping domain onto another finite coordinates, each representing processor array, such way concurrent are mapped different processors. scheduling and functions meet conditions allow full automation method. exemplified on convolution product matrix product.