Evaluation of hardware and software data prefetching

Jean Loup Baer*, Tien-Fu Chen

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

Prefetching, a technique that exploits the overlap of processor computations with data accesses, is one of several approaches for tolerating large memory latencies. Prefetching can be either hardware-based or software-directed. Hardware-based prefetching, requiring some support unit connected to the cache, handles prefetches at run-time without compiler intervention. Software-directed approaches, relying on compiler technology, insert explicit prefetch instructions. The common requirement for both approaches is a lockup-free cache supporting multiple outstanding memory requests. In this paper, we evaluate and compare the substantive performance gains that can be achieved with hardware-based and software-directed prefetching in a shared-memory multiprocessor environment. Results from direct-execution simulations on four benchmarks show that in the domain of linear array references both hardware and software schemes are able to efficiently generate prefetches to reduce the memory latency and execution time. When complex data access patterns are considered, the software approach can generate more accurate prefetches but at the cost of extra instruction execution overhead. Our results also show that negative effects such as increases in network traffic and data pollution in the cache are not severe.

Original languageEnglish
Title of host publicationIFIP Transactions A
Subtitle of host publicationComputer Science and Technology
PublisherPubl by Elsevier Science Publishers B.V.
Pages257-266
Number of pages10
EditionA-44
ISBN (Print)0444818707
StatePublished - 1 Jan 1994
EventProceedings of the IFIP WG10.3 Working Conference on Applications in Parallel and Distributed Computing - Caracas, Venez
Duration: 18 Apr 199422 Apr 1994

Conference

ConferenceProceedings of the IFIP WG10.3 Working Conference on Applications in Parallel and Distributed Computing
CityCaracas, Venez
Period18/04/9422/04/94

Fingerprint Dive into the research topics of 'Evaluation of hardware and software data prefetching'. Together they form a unique fingerprint.

  • Cite this

    Baer, J. L., & Chen, T-F. (1994). Evaluation of hardware and software data prefetching. In IFIP Transactions A: Computer Science and Technology (A-44 ed., pp. 257-266). Publ by Elsevier Science Publishers B.V..