arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1505.08023cs.DCcs.MS
收藏

基于MiniFE的并行有限元算法研究与优化

The Research and Optimization of Parallel Finite Element Algorithm based on MiniFE

Meng Wu, Can Yang, Taoran Xiang, Daning Cheng

摘要:

有限元法是现代工程设计和分析中最重要的数值方法之一。由于传统的串行有限元法难以高效、准确地求解大型有限元问题,高性能并行有限元法已成为解决实际工程问题的重要途径之一。本文以国家能源研究科学计算中心(NERSC)发布的MiniFE程序为基础,分析了并行有限元的具体步骤、关键计算模式和并行机制。根据实验结果,分析了各模块计算量的比例,并总结出程序的主要性能瓶颈。在此基础上,在服务器平台上对MiniFE程序进行了优化。针对程序的瓶颈-- SpMV内核进行了优化,并采用了高效的存储格式BCRS。并提出了MPI+OpenMP混合编程的改进方案。实验结果表明,优化后的程序在SpMV内核和同步方面都有较好的性能。它可以将程序的性能平均提高8.31%。关键词:有限元,并行,MiniFE,SpMV,性能优化

英文摘要:

Finite element method (FEM) is one of the most important numerical methods in modern engineering design and analysis. Since traditional serial FEM is difficult to solve large FE problems efficiently and accurately, high-performance parallel FEM has become one of the essential way to solve practical engineering problems. Based on MiniFE program, which is released by National Energy Research Scientific Computing Center(NERSC), this work analyzes concrete steps, key computing pattern and parallel mechanism of parallel FEM. According to experimental results, this work analyzes the proportion of calculation amount of each module and concludes the main performance bottleneck of the program. Based on that, we optimize the MiniFE program on a server platform. The optimization focuses on the bottleneck of the program - SpMV kernel, and uses an efficient storage format named BCRS. Moreover, an improving plan of hybrid MPI+OpenMP programming is provided. Experimental results show that the optimized program performs better in both SpMV kernel and synchronization. It can increase the performance of the program, on average, by 8.31%. Keywords : finite element, parallel, MiniFE, SpMV, performance optimization

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑