A Fast Hybrid Cache System for Pipelined Stream Processing[J]. 2015, 32(2): 1-4.DOI:
A Fast Hybrid Cache System for Pipelined Stream Processing
摘要
针对多核流应用处理器中存在的缓存效率低下和一致性问题
提出了数据流应用的流水化分割处理方案
在此基础上提出了一种基于一级私有缓存和二级共有缓存的混合型缓存结构
并深入分析了该结构运行过程中的完整性和一致性问题.通过对常用的FIR、FFT等DSP算法的验证
该高速缓存结构以一定的面积开销将访问下级缓存的可能性降低30%左右
显著降低了流应用的运行时间.
Abstract
To address the low performance and coherence problem in multicore stream processors
this paper proposes a pipeline with stream-based architecture.A hybrid cache system with private first-level cache and shared second-level is also introduced based on the architecture. We analysis the consistency and coherency of the architecture and then verify it on DSP algorithms such as FIR and FFT. According to the result
although the area is increased
the memory access time of the new system decreases by 30%.