Earlier this year, Arianna Delsante defended her Master’s Thesis in computer science at Uppsala University. Her thesis topic was to speed up cache and branch prediction simulation in Simics, and in the end she got a speed up of about 10x compared to previous implementations in Simics. I explain a bit more about cache simulation in fast functional simulators and what she did in my latest Intel Developer Zone Blog post, “Speeding-Up Cache Simulation in Simics by 10x“.
In short, the thesis work used the new instrumentation API introduced as a standard product feature in Simics 6 instead of older less efficient APIs. The work also used batching to process multiple memory accesses at once instead of sending each individual access to the cache model.