Reducing Memory-access Latency of Block Execution Model Through Data Prefetching
Yongqing Ren · Journal of Chinese Computer Systems · 2010
Block execution model makes a serial program run in parallel by partitioning instructions into a sequence of execution blocks,which can exploit potential instruction-level parallelism(ILP)in applications,However,memory access latency has become a major challenge to increase ILP in this model.Data prefetching can effectively reduce the memory access latency in traditional execution model,and it also has strong applicability to block execution model.In this paper,we analyze the feasibility of introducing data prefetch into block execution model,and give a study on the improved latency and cache hit rate of memory access instructions.Simulation results show that data prefetching can significantly reduced memory access latency in block execution model.