A limitation study into access decoupling

Abstract

This paper presents a study into the theoretical limits of a latency hiding technique called access decoupling. Access decoupling is effective at hiding memory latency for low ILP and conservative dependency analysis [9, 12, 13]. We assess if this result still applies for maximum ILP and perfect dependency analysis. We find that access decoupling with a basic decoupled memory model fails to hide latency. However, when the memory system is optimised to capture temporal locality exposed by decoupling, sensitivity to memory-latency is almost removed. The optimised memory is also found to be a powerful bandwidth filter.