Efficient Worst-Case Execution Time Analysis of Dynamic Branch Prediction

Abstract

Dynamic branch prediction is commonly found in modern processors, but notoriously difficult to model for worst-case execution time analysis. This is particularly true for global dynamic branch predictors, where predictions are influenced by the global branch history. Prior research in this area has concluded that modeling of global branch prediction is too costly for practical use. This paper presents an approach to model global branch prediction while keeping the analysis effort reasonably low. The approach separates the branch history analysis from the integer linear programming formulation of the worst-case execution time problem. Consequently, the proposed approach scales to longer branch history lengths than previous approaches.