Fast cycle-accurate behavioral simulation for pipelined processors using early pipeline evaluation

Modeling and simulating pipelined processors in procedural languages such as C/C++ requires lots of cost in handling concurrent events, which hinders fast simulation. A number of researches on simulation have devised speed-up techniques to reduce the number of events. This paper presents a new simulation approach developed to enhance the simulation of pipelined processors. The proposed approach is based on early pipeline evaluation that all the intermediate values of an instruction are computed in advance, creating a future state for the next instructions. The future state allows the next instructions to be computed without considering data dependencies between nearby instructions. We apply this concept to building a cycle-accurate simulator for a pipelined RISC processor and achieve almost the same speed as the instruction-level simulator.