分支预测器
计算机科学
投机性执行
管道(软件)
并行计算
中央处理器
感知器
执行时间
人工智能
操作系统
人工神经网络
作者
Akash D. Halke,Ashwini Kulkarni
标识
DOI:10.1109/upcon52273.2021.9667663
摘要
The branch predictors instruct the CPU, which direction is to be taken while executing branches so that the CPU will not stall fetching instructions in the pipeline. If the predicted direction is wrong, then the pipeline stalls and refetches the correct instructions. This penalty has to be paid by the CPU for its wrong prediction. So constantly working on reducing mispredictions is at most important. The difference between static and dynamic branch prediction is that static gives prediction at the time of compilation while dynamic gives prediction at the time of execution. Dynamic branch predictors maintain a history table that keeps a record of the last few branches' behaviors. So, improving dynamic branch prediction is of great significance as it contributes to the speed of the processor. We are attempting to simulate and analyse such a predictor using perceptron to investigate its performance as compared to counter-based predictors.
科研通智能强力驱动
Strongly Powered by AbleSci AI