Evaluating the impact of memory system performance on software prefetching and locality optimizations

• DOI: 10.1145/377792.377906
• Publication Date: 2003-11-14T12:54:41Z
• Authors: Abdel-Hameed A. B...; Aneesh Aggarwal; Donald Yeung; Chau-Wen Tseng
• Keywords: Memory bandwidth; Padding; CAS latency

Software prefetching and locality optimizations are techniques for overcoming the speed gap between processor memory. In this paper, we evaluate impact of memory trends on effectiveness software three types applications: regular scientific codes, irregular pointer-chasing codes. We find many applications, outperforms when there is sufficient bandwidth, but outperform under bandwidth-limited conditions. The break-even point (for 1 Ghz processors) occurs at roughly 2.5 GBytes/sec today's systems, will increase future systems. also study interactions applied in concert. Naively combining provides robustness to changes bandwidth latency, does not yield additional performance gains. propose several algorithms better integrate optimizations, including a modified tiling algorithm, padding prefetching, index prefetching.