CoGS-Sim- CoPhase-Guided Small-Sample Simulation of

The emergence and ubiquity of multithreaded and multicore architectures in modern proces-sors presents many new challenges in processor design. These architectures present signicantly larger design and benchmark spaces than tradi-tional single-context architectures. Such characteris-tics put ever-higher strain on ecient cycle-accurate simulation which remains a vital tool in the de-sign process. In amny cases in multicontext asrchitectures, in-dependent processes are running simultaneously. By tracking the program phase of each of the running processes and using that information to guide small-sample simulation, it is possible to perform very fast and accurate sampled simulation of a multicontext ar-chitecture. As with any sampled simulation, speed-up is achieved by minimizing the amount of detailed, cycle-accurate simulation. By tracking the variance of performance in each co-phase, it is possible to mini-mize the amount of simulation in each co-phase and of the programs as a whole, further accelerating the simulation process and providing an automatic adap-tive mechanism to adjust the amount of detailed sim-ulation as needed. Finally, by tracking phase behav-ior, it is possible to quickly adapt to co-phase changes on the y during simulation without prior knowledge of the running programs. This paper presents this technique, which is called CoGS-Sim for CoPhase-Guided, Small-Sample Simulation. 1