OpenMP在多核处理器上的应用研究
The Application Research of OpenMP on Multicore Processors
DOI: 10.12677/CSA.2014.49025, PDF, HTML, 下载: 2,473  浏览: 6,053  国家科技经费支持
作者: 王孝平, 董秀成, 谢维成:西华大学电气信息学院,成都
关键词: 多核编程并行计算OpenMPMulticore Programming Parallel Computing OpenMP
摘要: 多核操作系统在解决通用进程的分配和调度等宏观问题上表现优越,而对于应用层的多核数据并行计算没有做到精确控制。本文深入研究了OpenMP的特性,提出了一种应用层开发多核并行程序的机制,对其性能进行了全方位的研究比较。实验结果表明,利用这种机制,可以方便灵活的控制多核数据并行粒度,充分利用多核计算机资源,提升应用程序的实时性。对于千万级别的运算量,运行时间上有较大的改进:双核并行为非并行的0.5,四核并行为非并行的0.38,八核并行为非并行的0.19,十六核并行为非并行的0.14。
Abstract: Multicore operating system is superior in assigning and scheduling the general processes, but it is unsatisfying in processing parallel data of the application layer. We study the characteristics of OpenMP and propose a mechanism for the development of application-layer multi-core parallel program. We conduct a comprehensive study and compare its performance. Our experimental re-sults show that the mechanism is convenient and flexible to control the multicore parallel granu-larity of data, take advantage of multicore computing resources, and enhance the real-time of ap-plications. To the ten million of computation, there are greater improvements in running time: the ratio is 0.5:1 between the dual-core and the non-parallel systems; the ratio is 0.38:1 between the quad-core and the non-parallel systems; the ratio is 0.19:1 between the eight-core and the non- parallel systems; the ratio is 0.14:1 between the sixteen-core and the non-parallel systems.
文章引用:王孝平, 董秀成, 谢维成. OpenMP在多核处理器上的应用研究[J]. 计算机科学与应用, 2014, 4(9): 175-180. http://dx.doi.org/10.12677/CSA.2014.49025

参考文献

[1] 于慧莉, 李勤新, 宫春明 (2014) 一种基于OpenMP和MPI的非序贯蒙塔卡罗暂态稳定评估的动态混合并行化方法. 现代电力, 2, 31-36.
[2] 潘亮, 郭改枝, 宋鑫梦 (2014) 基于OpenMP矩阵相乘并行算法的设计. 宝鸡文理学院学报(自然科学版), 1, 21-23.
[3] 熊雨前, 冯丙春, 薛军超 (2014) OpenMP在PDF技术中的应用. 数字与缩微影像, 1, 43-45.
[4] 万波, 魏帆, 金钟 (2014) 自洽场方法在Xeon Phi上的并行实现. 科研信息化技术与应用, 2, 34-42.
[5] 白洪涛, 李昂, 欧阳丹彤, 邢书豪, 刘雪飞 (2014) 基于多核的粗粒度2.5维电磁场正演并行算法. 吉林大学学报(理学版), 3, 509-514.
[6] 蒋沁谷 (2014) GRAPES全球模式MPI + OpenMP混合并行方法. 中国气象科学研究院, 北京.