Abstract
In this paper, we introduce a new concept of run-time collaboration between hardware and software prefetching mechanisms. An association bit is added to a memory access instruction (MAI) to indicate if any software PREFETCH instruction corresponding to the MAI has been inserted into the program. This bit is set by the compiler. Default hardware prefetching might be triggered for a MAI only if a “1” is detected in this bit. Simulation on SPEC95 shows that this association concept is very useful in HW/SW hybrid prefetching; its performance improvement in floating point applications ranges from a few percent to about 60%, with an average of 28.63%. This concept is important because its requirements for hardware and compiler support are very minimal. Furthermore, most existing architectures actually have unused encoding space that can be used to hold the association information.
Chapter PDF
References
Callahan, D., Kennedy, K., Porterfield, A., “Software Prefetching,” Proceedings of the Four International Conference on Architectural Support for Programming Languages and Operating Systems, April 1991, pp. 40–52.
Chi, C.H., Cheung, C.M., “Hardware-Driven Prefetching for Pointer Data References,” Proceedings of the 1997 ACM International Conference on Supercomputing, July 1998.
Kane, G., PA-RISC 2.0 Architecture, Prentice-Hall Press, 1996.
Manku, G.S., Prasad, M.R., Patterson, D.A., “A New Voting Based Hardware Data Prefetch Scheme,” Proceedings of 4 th International Conference on High Performance Computing, Dec. 1997, pp. 100–105.
Wang, K., Franklin, M., “Highly Accurate Data Value Prediction using Hybrid Predictors,” Proceedings of the MICRO-30, 1997, pp. 281–290.
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2002 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Chi, CH., Yuan, J. (2002). Runtime Association of Software Prefetch Control to Memory Access Instructions. In: Monien, B., Feldmann, R. (eds) Euro-Par 2002 Parallel Processing. Euro-Par 2002. Lecture Notes in Computer Science, vol 2400. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-45706-2_66
Download citation
DOI: https://doi.org/10.1007/3-540-45706-2_66
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-44049-9
Online ISBN: 978-3-540-45706-0
eBook Packages: Springer Book Archive