Advancing the RISC-V Performance Simulation Ecosystem with Data Prefetching
摘要
As processor speeds continue to outpace memory access times, memory latency remains a critical bottleneck in modern computing systems. Addressing this challenge requires effective memory access optimizations, such as data prefetching, which can anticipate memory requests and reduce stalls. Meanwhile, RISC-V has emerged as a compelling open-source alternative to proprietary ISAs with increasing adoption in domains ranging from embedded systems to high-performance computing. However, as RISC-V systems scale in complexity, there is a growing need for accurate and efficient performance modeling to guide architectural optimizations. Accordingly, this paper enhances the RISC-V Olympia trace-based performance simulator by extending it to support prefetching mechanisms. Additional contributions include modifications to the Spike simulator and the development of a complementary parser for generating an open-source simulation trace format. The effectiveness of these prefetching strategies is evaluated, demonstrating their potential to reduce memory access latency and enhance RISC-V system performance.