多核架构下的数据处理算法优化策略综述  

Optimization Strategies of Data Processing Algorithms under Multi-Core Architecture

作　　者：陈伟[1,2] 杜凌霞[1,2] 陈红[1,2] 

机构地区：[1]中国人民大学数据工程与知识工程教育部重点实验室,北京100872 [2]中国人民大学信息学院,北京100872

出　　处：《计算机科学与探索》2011年第12期1057-1075,共19页Journal of Frontiers of Computer Science and Technology

基　　金：国家高技术研究发展计划(863)No.2008AA01Z120;国家教育部新世纪优秀人才支持计划;国家教育部博士点基金No.20090004110002~~

摘　　要：多核处理器,尤其是单芯片多处理器(chip multi-processor,CMP)能够提供强大的共享内存的并行资源,然而单核处理器上的程序和算法并不能充分利用多核架构提供的并行计算资源,因此必须针对多核体系架构特点,对算法进行改进优化,提高算法的执行性能。以优化程序局部性、减少cache访问冲突、提高线程并行度、充分利用单指令多数据流(single instruction multipledata,SIMD)并行和带宽优化等几方面为出发点,归纳和分析了多核处理器上数据处理算法的相关优化策略,并对多核算法进行了总结评述。最后阐述了该领域亟待解决的诸多问题,展望了未来的研究发展方向。Multi-core processors, especially the single chip multi-processors （CMP）, can provide strong parallel resources of sharing memory. However, the majority of programs and algorithms from single-core processors can not make full use of computing resources of multi-core architecture. Therefore, according to some characteristics of multi-core architecture, researchers should optimize those algorithms and substantially increase the performance of algorithms. This paper reviews related optimization strategies of data processing algorithms on multi-core processors around several aspects, including improving the locality of programs, decreasing cache access conflict, enhancing the thread parallelism, taking advantage of single instruction multiple data （SIMD） parallelism and optimizing bandwidth usage. In addtion, it summarizes and comments on the algorithms on multi-core processors. Finally, it introduces several interesting issues that people should address and predicts the prospect of the research on multi-core processors in future.

关 键 词：多核 单芯片多处理器(CMP) 数据级别并行(DLP) 线程级别并行(TLP) 单指令多数据流(SIMD) 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]