Split Temporal/Spatial Cache: A Survey and Reevaluation of Performance

The purpose of this paper is to reevaluate the performance of the Split Temporal/Spatial (STS) cache. First we briefly survey the split cache designs found in the open literature. Then we propose quantitative definitions for both temporal and spatial locality. These definitions can be used to represent each split cache design (or any other method for optimized locality exploitation) as a line in a temporal-spatial locality plane. Then we explain the particular process used to evaluate the STS cache design, and finally we present the results of that evaluation. We conclude with possible improvements pointed to by our evaluation results. Introduction In recent years, the speed gap between dynamic memories and microprocessors has been steadily increasing. For this reason, a lot of effort is invested into finding ways to reduce or hide memory latency. One of the oldest and most powerful ways of reducing the memory latency is through use of cache memories. Caches exploit the locality of d..