Comparison of the Cellular Logic Image Processor (CLIP) 4; Distributed Array Processor (DAP) and Massively Parallel Processor (MPP) Processor-Array Implementations

摘要

Image processing applications of the SIMD processor array are compared. It is shown that, for sufficiently long algorithms, the performance of processor arrays which use column or row parallel I/0 overlapped with processing is equivalent to the performance of processor arrays which use image parallel I/0. The relative speed up gained by overlapping I/0 and processing in the cases of column or row parallel I/0 as a function of algorithm length is shown to be sharply peaked.