改进的基于O-GEHL预测技术的EDGE块预测器
Improved Next-Block Predictor Based on O-GEHL for EDGE Architectures
-
摘要: 块预测器赋予了EDGE体系结构高效的控制流推测能力。针对基于O-GEHL预测技术的块预测器的缺点,提出了两种改进策略:1) 无选择器的O-GEHL出口预测;2) 直接使用OGEHL预测器的二值预测能力单独对每个出口进行预测。性能评价结果表明,无选择器的O-GEHL出口预测方案比文献4中的O-GEHL块预测器性能平均提高0.7%;对8个出口分别使用二值O-GEHL预测器进行预测的方案,在硬件资源较多时,性能提高3%;只对前4个出口采用二值O-GEHL预测器的方案,性能平均提高2%.Abstract: Next-block predictors enable high efficiency control-flow speculation of EDGE architecture. This paper analyzes defects of next-block predictor based on O-GEHL prediction technology, and proposes two improvements: the exit prediction without chooser and the exit prediction with binary O-GEHL prediction. Performance evaluation results show that: the one without chooser in the exit prediction stage outperforms previously published one by 0.7%; the one using 8 conditional O-GEHL predictor improves the performance by 3% with the largest resource; the one using 4 conditional O-GEHL predictor only for the first 4 exits improves the performance by 2%.