Worst-Case Execution Time Analysis for Dynamic Branch Predictors

  • Authors:
  • Iain Bate;Ralf Reutemann

  • Affiliations:
  • University of York;University of York

  • Venue:
  • ECRTS '04 Proceedings of the 16th Euromicro Conference on Real-Time Systems
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

Branch prediction mechanisms are becoming commonplacewithin modern microprocessors. For developers ofreal-time control systems, the mechanisms present predictabilityproblems. The reasons are they increase thedifficulty in analysing software for its Worst-Case ExecutionTime without introducing unmanageable pessimismand they increase the variability of the softwareýs executiontimes. In this paper, we improve upon existing branchprediction analysis by taking into account the semanticcontext of the branches in the source code in order toclassify them as either being easy-to-predict or hard-to-predict.Based on this classification we provide a staticanalysis approach for bimodal and global-history branchprediction schemes. The analysis is applied to a previouslypublished example with the benefit that a more detailed explanationof its results is obtained.