Employing both inter-branch and intra-branch correlation to improve the accuracy of branch prediction for superscalar processors

Abstract

Today's superscalar processors use branch prediction to reduce the influence of control hazards. Conventional two-level branch predictors make predictions based on either intra-brance correlation or inter-branch correlation . In the paper, the authors proposed two new branch predictors, called PGXg and PGAg, which make predictions by employing both intra-branch correlation and inter-branch correction. It is shown that the proposed branch predictors can achieve a better cost-performance ratio than conventional two-level branch predictors, such as PAg and gshare.