Analysis and Accuracy Improvement for Perceptron-based Branch Prediction Method

In the modern microarchitecture, high accuracy of branch predictors is essential for improving the overall performance of the processors. With the emergence of the Perceptron-based predictors, prediction based on long global branch history becomes possible, and this advantage gives this type of strategies much potential in achieving extremely high prediction accuracy. This paper devises two methods to obtain further accuracy improvement through slight modifications on the original structure, and presents the analysis of the limitation of indexing function refinement on Perceptron-based branch predictor. The performance of the two methods and the limitation analysis are evaluated with the championship branch prediction framework.