Closed-loop adaptive and stochastic prefetch mechanism for data array

Nowadays, one of the main limiting factor in multiprocessor development is the increasing speed gap between efficient processing elements and slow main memories. To reduce this limitation, prefetching mechanisms, implemented in memory hierarchy, attempt to predict the future data needed in local memory. However, classical proposed solutions are no longer efficient for typical access sequences in the context of image processing algorithms. In this paper, a prefetching mechanism is proposed, based on the use of the index array of access rather than memory addresses. The adaptive prediction policy takes into account both application and main memory characteristics. This mechanism enhances the Cache nD-AP, formerly statically set, to improve its performance. Experiments show that the proposed adaptive strategy is able to maintain its efficiency when the characteristics of the system dynamically change.