A limitation study into access decoupling

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.