Modeling Out-of-Order Superscalar Processor Performance Quickly and Accurately with Traces

Fast and accurate processor simulation is essential in processor design. Trace-driven simulation is a widely practiced fast simulation method. However, serious accuracy issues arise when an out-of-order superscalar processor is considered. In this thesis, trace-driven simulation methods are suggested to quickly and accurately model out-of-order superscalar processor performance with reduced traces. The approaches abstract the processor core and focus on the processor's uncore events rather than the processor's internal events. As a result, fast simulation speed is achieved while maintaining fairly small error compared with an execution-driven simulator. Traces can be generated either by a cycle-accurate simulator or an abstract timing model on top of a simple functional simulator. Simulation results are more accurate with the method using traces generated from a cycle-accurate simulator. Faster trace generation speed is achieved with the abstract timing model. The methods determine how to treat a cache miss with respect to other cache misses recorded in the trace by dynamically reconstructing the reorder buffer state during simulation and honoring the dependencies between the trace items. This approach preserves a processor's dynamic uncore access patterns and accurately predicts the relative performance change when the processor's uncore-level parameters are changed. The methods are attractive especially in the early design stages due to its fast simulation speed.