Refining algorithm mappings for linear systolic arrays

摘要

Nested FOR loop algorithms are the most common applications ofsystolic arrays. All the techniques for mapping nested FOR loops ontosystolic arrays have the `curse of dimensionality'. The authors proposethe approach of first finding coarse-grained mappings that are easier todetermine and then refining these mappings through some simpletransformations, to obtain efficient fine-grained mappings. The authorspropose three transformation techniques that can derive efficientfine-grained mappings for linear systolic arrays. These techniques canbe used to derive easily efficient mappings for some commonly knownalgorithms