【作者】 王庆； 季振洲； 刘涛；

【Author】 Wang Qing,Ji Zhenzhou,and Liu Tao (School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001)

【机构】 哈尔滨工业大学计算机科学与技术学院；

【摘要】 当前嵌入式多核处理器的应用越来越普遍,如何充分认识与利用嵌入式多核的并行计算效率已经成为嵌入式多核并行计算的热门问题.讨论了两种不同的共享式嵌入式多核存储层次结构,即无二级共享Cache和拥有二级共享Cache的多核存储结构,在这些嵌入式多核系统中进行实际的OpenMP并行应用计算,并通过增加OpenMP对循环分块的支持来对并行程序进行数据优化.实验结果对比分析表明,在嵌入式多核中采用共享二级Cache对并行程序性能影响不大,而在对并行程序进行优化后性能平均有6%的提高.更多还原

【Abstract】 The current multicore processors for embedded applications become increasingly popular, how to improve the parallel computing efficiency of the embedded multicore has become a key issue. This paper discussed two different shared-memory architectures for embedded multicore,namely level 2 Cache non-sharing and level 2 Cache sharing.We present our solution to implement OpenMP programs on these embedded multicore architectures,and to optimize the parallel programs with supporting loop tiling.Performance evaluations show that there is little influence of performance between Cache non-sharing and Cache sharing.By transforming the OpenMP programs in a localityaware manner,we observe up to 6%performance increase.更多还原