Modeling the dynamic behavior of an ILP processor

Abstract

We introduce a new approach to evaluate the effects of branch and value prediction. So far, all the papers published in this area used trace-driven or execution-driven simulation in order to obtain an estimation of the potential to use these techniques. We introduce an analytical model based on fluid stochastic Petri nets (FSPNs). This model is used to capture the dynamic behavior of an ILP processor with aggressive use of prediction techniques and speculative execution. We define the FSPN model and derive the state equations for the underlying stochastic process. Our implementation-independent approach confirms recently published results about control and data speculation and reveals considerable potential for further research using numerical solution of systems of partial differential equations.