Fetch Size Adaptation vs. Stream Buffer for Media Benchmarks

ABSTRACT In urrent a he designs, a he size and line size are often xed and determined by the spatial and temporal lo ality of ben hmarks used to evaluate the targeted pro essors. Stream bu er and large fet h size are two te hniques that exploit spatial lo ality to hide memory laten y. In this paper, we ompare the e e tiveness of stream bu er and fet h size on media ben hmarks. We have presented a dynami predi tion me hanism for fet h size adaptation. Our results show that fet h size adaptation is as e e tive as stream bu er and both an improve the performan e of media ben hmarks by up to 10%. Due to its simple hardware design, fet h size adaptation is a viable alternative to stream bu er in hiding memory laten y.