Two-Level Alloyed Branch Predictor based on Genetic Algorithm for Deep Pipelining Processors

To gain improved performance in multiple issue superscalar processors, the increment in instruction fetch and issue rate is pretty necessary. Evasion of control hazard is a primary source to get peak instruction level parallelism in superscalar processors. Conditional branch prediction can help in improving the performance of processors only when these predictors are equipped with algorithms to give higher accuracy. The Increment in single miss-prediction rate can cause wastage of more than 20% of the instructions cycles, which leads us to an exploration of new techniques and algorithms that increase the accuracy of branch prediction. Alloying is a way to exploit the local and global history of different predictors in the same structure and sometimes also called hybrid branch prediction. In this paper, we aim to design a more accurate and robust two-level alloyed predictor, whose behavior is more dynamic on changing branch direction.