Efficient Cycle-Accurate Simulation of the UltraSPARC III CPU

Abstract

This paper presents a novel technique for cycleaccurate simulation of the Central Processing Unit (CPU) of a modern superscalar processor, the Ultra-SPARC III Cu processor. The technique is based on adding a module to an existing fetch-decode-execute style of CPU simulator, rather than the traditional method of fully modelling the CPU microarchitecture. It is also suitable for accurate SMP modelling. The main functions of the module are the simulation of instruction grouping, register interlocks and the store buffer. Its simple table-driven implementation permits easy modification for exploring microarchitectural variations. The technique results in a 40% loss of simulation speed, instead of a 10 times or greater performance loss by fully implementing the detailed micro-architecture. The technique is validated against an actual UltraSPARC III Cu processor, and achieves high levels of accuracy over a range of scientific benchmarks. Show more