Improving Accuracy of Perceptron Predictor Through Correlating Data Values in SMT Processors

摘要

Simultaneous Multithreaded (SMT) processors improve the instruction throughput by allowing fetching and running instructions from several threads simultaneously at a single cycle. With the pipeline deepen and issue widths increase, the branch predictor plays a more important role in improving the performance of an SMT processor. Many predictors based on neural network, especially on perception, are proposed to provide a more accurate dynamic branch prediction than before in the literature. In this paper, we propose an effective method to improve the accuracy of a perceptron predictor through correlating data values in SMT processors. The key idea is using a dynamic bias input, which comes from some information independent on the branch histories (data values for example), to realize the objective of improving accuracy. The implementation of our method is simple, and the predicting latency is not lengthened. Execution-driven simulation results show that our method works successfully on improving the accuracy of a perceptron predictor and increasing the overall instruction throughput of SMT processors.