Fine grain parallel processors and real-time applications: MIMDcontroller/SIMD array

摘要

Important issues related to the applications of massively parallelarchitectures for real-time cognition are discussed. One of these isreal-time control of the array processor. Systems accommodating massivedata parallelism are faced with the critical issue of how to incorporateadaptive programming decisions without losing machine efficiency. Thisinvolves reexamination of functions to be performed by the softwareversus those performed by the hardware. The discussion is based on amultiple-instruction multiple-data (MIMD) controller that has beenconstructed for a single-instruction multiple-data (SIMD) array. Theefficiency of an SIMD architecture as a computational model is discussedwith reference to several important functions in vision applicationsthat have been benchmarked on the system. Applications that involvereal-time control are described