计算机工程与应用 ›› 2010, Vol. 46 ›› Issue (10): 57-59.DOI: 10.3778/j.issn.1002-8331.2010.10.019

• 研发、设计、测试 • 上一篇    下一篇

采用PETSc的有限元并行计算实现与优化

张健飞   

  1. 河海大学 土木工程学院,南京 210098
  • 收稿日期:2008-11-07 修回日期:2008-12-22 出版日期:2010-04-01 发布日期:2010-04-01
  • 通讯作者: 张健飞

Parallel implementation and tuning of finite element computing with PETSc

ZHANG Jian-fei   

  1. School of Civil Engineering,Hohai University,Nanjing 210098,China
  • Received:2008-11-07 Revised:2008-12-22 Online:2010-04-01 Published:2010-04-01
  • Contact: ZHANG Jian-fei

摘要: 可移植可扩展科学计算工具箱PETSc提供了高性能求解偏微分方程组的大量对象和解法库,基于此进行结构有限元并行计算,可降低难度和成本。给出了基于PETS的结构有限元并行计算实现方法,包括有限元方程组的并行形成和并行求解的实现。根据PETSc的特点,提出了提高计算性能的优化措施,即数据局部化和存储预分配。数值实验表明实现方法可行,优化措施效果明显。

关键词: 可移植可扩展科学计算工具箱(PETSc), 有限元, 并行计算, 数据局部化, 存储预分配

Abstract: Abstract: Portable Extensible Toolkit for Scientific computation(PETSc) provides many high performance objects and solvers for partial differential equations.The difficulty and cost of structural parallel finite element computing with PETSc will come down.The parallel implementation of structural finite element computing with PETSc is presented,including parallel formation and parallel solving of finite element equations.According to the characteristics of PETSc,two performance tuning measures are presented,they are data localization and storage pre-allocation.Numerical test shows that the implementing method is practicable and performance tuning measures are effective.

Key words: Portable Extensible Toolkit for Scientific computation(PETSc), finite element, parallel computing, data localization, storage pre-allocation

中图分类号: