Exploration of Performance of Dynamic Branch Predictors used in Mitigating Cost of Branching

Abstract

Today, Branch Predictors have become a sophisticated component in all modern processors. While contemporary predictors have good accuracy, the cost of a misprediction is quite high. Hence, active research is being conducted towards improving the performance of branch predictors, while reducing the resources consumed. Modern computer architectures increasingly rely on speculation to boost instruction-level parallelism (ILP). As the pipelines deepen and instructions issued per cycle increase, the accurate branch prediction mechanisms become more and more important in increasing the processor performance. This study explores some branch predictors and compares them on the basis of their performance on certain benchmarks applications.