An empirical study on performance of branch predictors with varying storage budgets

Abstract

The objective of this paper is to examine the common branch predictor designs available in literature, and characterize their accuracy versus storage performance. As expected, many of the predictors which are known to have high accuracy in general, lose out on performance when exercised in low storage scenarios. This paper presents an empirical evaluation of different branch predictors at various storage points and the resulting effect on processor performance in terms of prediction accuracy and latency. We present our findings using the branch predictors and the traces of the Championship Branch Predictor-2 benchmarks. We believe that our study will be extremely beneficial for choosing a branch predictor design for embedded processors working in resource constrained environments.