本文已被：浏览 1134次 下载 2994次
Received:April 23, 2015 Revised:May 23, 2015
Received:April 23, 2015 Revised:May 23, 2015
中文摘要: OpenMP规范了一系列的编译制导、环境变量和运行库,具有简单、可移植、支持增量并行等优点.但同时,采用FORK-JOIN模型所引起的频繁的线程管理开销也是制约OpenMP程序性能的瓶颈之一.本文讨论了如何利用并行区的合并与扩展,实现并行区的重构,并在此基础上利用Open64的IPA优化部件所提供的全局间过程分析能力,实现跨越过程边界的并行块的合并.最终实验表明,该方法有效地改进了OpenMP程序的运行性能.
Abstract:OpenMP regulates a series of compilation guidance, environment variables and runtime routines, having the advantages of simple operation, portability and supporting incremental parallel. But at the same time, the use of frequent thread management overhead FORK-JOIN model is one of the bottlenecks caused by OpenMP program performance constraints.. This article discusses how to use the merge and extend parallel zone, reconstruct the parallel section, and on this basis, using global process analysis ability which is provided by IPA, one of the Open64 optimization components, realize the parallel block mergeing which acrosses process boundaries. The final experimental results show that the method improves the performance of OpenMP programs effectively.
文章编号： 中图分类号： 文献标志码：
基金项目:"核高基"重大专项(2012ZX01034-001-001)
引用文本：
刘京,郑启龙,李彭勇,郭连伟.面向Open64的OpenMP程序优化.计算机系统应用,2016,25(1):154-159
LIU Jing,ZHENG Qi-Long,LI Peng-Yong,GUO Lian-Wei.OpenMP Program Optimization Based on Open64.COMPUTER SYSTEMS APPLICATIONS,2016,25(1):154-159
刘京,郑启龙,李彭勇,郭连伟.面向Open64的OpenMP程序优化.计算机系统应用,2016,25(1):154-159
LIU Jing,ZHENG Qi-Long,LI Peng-Yong,GUO Lian-Wei.OpenMP Program Optimization Based on Open64.COMPUTER SYSTEMS APPLICATIONS,2016,25(1):154-159
