Formal Analysis of Branch Prediction Algorithms

In modern microprocessors deep pipelines are widely used to speed up execution. The importance of a good branch predictor to the eeectiveness of a deep pipeline in the presence of conditional branches is well-known. There are several branch prediction schemes described in literature. The common way to evaluate the performance of branch prediction algorithms is simulation. In this paper we present a formal methodology for evaluating branch prediction algorithms. Our goal is to provide a technique for analyzing branch prediction which is based on rigorous mathematical foundations and thus provide a deeper insight into the analysis and design of branch prediction algorithms. Our technique hinges on constructing a Markov model or a probabilis-tic automaton corresponding to the test suite. We also show the eeectiveness of our method by experimental results.