Abstract
Recent research has shown that it is possible to overcome the dataflow limit by predicting instruction results based on previously produced values or a sequence thereof. Instruction results often depend on the path used to arrive at that instruction; by maintaining different data value histories for the different paths leading up to an instruction, it is possible to do better predictions. This paper studies the effect of control flow correlation schemes on stride-based data value predictors. Our studies indicate that if enough hardware resources can be provided for storing the histories corresponding to different paths, the prediction accuracy increases steadily as the degree of correlation is increased.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
Lipasti, M.H., Shen, J.P.: Exceeding the Dataflow Limit via Value Prediction. In: Proc. 29th Int’l Symposium on Microarchitecture, pp. 226–237 (1996)
Nair, R.: Dynamic Path Based Branch Correlation. In: Proc. 28th Int’l symposium on Microarchitecture (1995)
Nakra, T., Gupta, R., Soffa, M.L.: Global Context-Based Value Prediction. In: Proc. Int’l Symposium on High Performance Computer Architecture (1999)
Wang, K., Franklin, M.: Highly Accurate Data Value Prediction. In: Proc. 3rd Int’l Conference on High Performance Computing, pp. 358-363 (1997)
Young, C., Smith, M.: Improving the Accuracy of Static Branch Prediction using Branch Correlation. In: Proc. ASPLOS-VI (1994)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 1999 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Mohan, W., Franklin, M. (1999). Improving Data Value Prediction Accuracy Using Path Correlation. In: Banerjee, P., Prasanna, V.K., Sinha, B.P. (eds) High Performance Computing – HiPC’99. HiPC 1999. Lecture Notes in Computer Science, vol 1745. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-46642-0_4
Download citation
DOI: https://doi.org/10.1007/978-3-540-46642-0_4
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-66907-4
Online ISBN: 978-3-540-46642-0
eBook Packages: Springer Book Archive