Record Branch Prediction: An Optimized Scheme for Two-level Branch Predictors

Abstract

Modern processors get great performance improvement by branch prediction, and two-level branch predictor is widely used because of its simple structure and high performance. But this kind of predictor has 3 limitations-Branch Alias, History Noise and Multi-way Access. In this paper, we propose an optimized mechanism named Record Branch Prediction, which records the PCs and GHR of previous executed branches in order to guide current branch prediction. We apply Record Branch Prediction on the traditional two-level branch predictors and implement Record-Gshare and Record2Bc-gskew. The results show that record predictors can get less prediction miss than original predictor without Record Branch Prediction. Besides, longer PC-Records and larger PHT capacity can reduce more prediction miss percentage for original predictor in general. As an example, Records-Gshare can reduce 7.6% prediction miss compared with Gshare and Records-2Bc-gskew can reduce 14.5% prediction miss compared with 2Bc-gskew on average when number of PC-Records entries is 20 and number of PHT entries is 4096.

An Improved Dynamic Branch Predictor by Selective Access of a Specific Element in 4-Way Cache

“…Bimodal은 최근의 분기 결과를 BHT (Branch History [7] . 기존의 분기 예측기는 대부분 범용 프로세서를 기준으로 설계되므로 전력 소모가 많고 분기의 예측률을 높이기 위해 노력하 였다 [8][9][10][11] . [12] .…”

“…T. Chen은 2단으로 구성된 분기 예측기를 사용하였다. 이런 방법들은 Gshare 분기 예측기의 예측률을 높이는데 치중하여 전력소모가 크므로, 제한된 전력하에서 분기 예측률이 높은 동 적 분기 예측기가 필요하다 [10] .…”