A study of perceptron based branch prediction on Simplescalar platform

This paper presents a perceptron based branch prediction approach. The method of implementation of the perceptron approach on the Simplescalar platform is provided. Mibench benchmark sets have been run to prove the optimal branch prediction rate of perceptron predictors comparing with conventional predictors. Further investigations have been outlined to indicate the impact of history table length and number of perceptron on the proposed branch predictors. Results show that perceptron based prediction scheme outperforms the other existing methods. Furthermore, the history table, within a certain bound, has an obvious impact on the prediction result, while the number of perceptron´s impact is trivial.