GPU Accelerating for Rapid Multi-core Cache Simulation

To find the best memory system for emerging workloads, traces are obtained during application´s execution, then caches with different configurations are simulated using these traces. Since program traces can be several gigabytes, simulation of cache performance is a time consuming process. Compute unified device architecture (CUDA) is a software development platform which enables programmers to accelerate the general-purpose applications on the graphics processing unit (GPU). This paper presents a real time multi-core cache simulator, which was built based on the Pin tool to get the memory reference, and fast method for multi-core cache simulation using the CUDA-enabled GPU. The proposed method is accelerated by the following techniques: execution parallelism exploration, memory latency hiding, a novel trace compression methodology. We describe how these techniques can be incorporated into CUDA code. Experimental results show that the hybrid parallel method of time-partitioning combines with set-partitioning presented here is 11.10× speedup compared to the CPU serial simulation algorithm. The present simulator can characterize cache performance of single-threaded or multi-threaded workloads at the speeds of 6-15 MIPS. It can simulates 6 cache configurations within one single pass at this speeds compared to CMP$im, which can only simulate one cache configuration each simulation pass at the speeds of 4-10 MIPS.