关于多核系统并行程序效率的编程因素及其研究  

Programming Factors about Efficiency of Parallel Program in Multi-core System and its Research

作　　者：王文义[1] 冉晓龙[1] 

机构地区：[1]中原工学院并行处理技术研究所,郑州450007

出　　处：《计算机科学》2015年第8期28-31,59,共5页Computer Science

基　　金：国家自然科学基金项目(61379079);河南省基础与前沿技术研究项目(082300410300)资助

摘　　要：着重分析了多核架构系统中内存对齐技术与cache利用率等因素对并行程序性能的影响。用共享存储环境OpenMP分析了并行计算量与处理器核心数目之间的关系,通过用MPI编程实现的矩阵相乘的行划分和CANNON算法等实例分析,指出了只有综合考虑了多核系统的结构特征、系统软件、多核编程语言环境以及正确运用算法等,才能设计出高效且能耗又小的并行应用程序。This paper emphatically analyzed the effects of the memory alignment technology, cache using ratio and such as these factors on parallel program performance in the multi-core architecture system. This paper used shared memory environment OpenMP to analyze the relationship between parallel amount of calculation and the number of the processor core. The results of experiment, which were programmed by MPI to implement the row-divided algorithm and CAN- NON algorithm of matrix multiplication, point out that only in the comprehensive consideration of the architecture fea- ture of multi-core system, system software, multi-core programming language environment and the correct use of algo- rithm can we design a better parallel application--being capable of high efficiency and small energy consumption.

关 键 词：绿色计算 内存对齐 OPENMP CANNON算法 多核处理器 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]