A comparison of hardware prefetching techniques for multimedia benchmarks

Abstract

Data prefetching is a well known technique for improving cache performance. While several studies have examined prefetch strategies for scientific and commercial applications, no published work has studied the special memory requirements of multimedia applications. This paper presents data for three types of hardware prefetching schemes: stream buffers, stride prediction tables, and a hybrid combination of the two, the stream cache. Use of the stride prediction table is shown to eliminate up to 90% of the misses that would otherwise be incurred in a moderate or large sized cache with no prefetching hardware. The stream cache, proposed for the first time in this paper, has the potential to cut execution times by half with the addition of a relatively small amount of additional hardware.