摘要
研究基于GPU的有限元求解中的总刚矩阵生成和线性方程组求解问题.通过对单元着色和分组完成总刚矩阵的生成,并以行压缩存储(Compressed Sparse Row,CSR)格式存储,用预处理共轭梯度法求解所生成的大规模线性稀疏方程组.在CUDA(Compute Unified Device Architecture)平台上完成程序设计,并用GT430 GPU对弹性力学的平面问题和空间问题进行试验.结果表明,总刚矩阵生成和方程组求解分别得到最高11.7和8的计算加速比.
The global stiffness matrix generation and the linear equations solution in finite element solution based on GPU is researched. The global stiffness matrix is generated using element coloring and grouping technique and stored in Compressed Sparse Row （CSR） format, and the preconditioned conjugate gradient method is used to solve the generated large-scale sparse linear equations. The code is programmed on Compute Unified Device Architecture （CUDA） platform and the plane and 3D elasticity matters are tested by GT430 GPU. The results show that the calculation speedups of the global stiffness matrix generation and linear equations solution reach 11.7 and 8 respectively.
出处
《计算机辅助工程》
2014年第2期41-45,共5页
Computer Aided Engineering
基金
国家自然科学基金(51109072)
关键词
GPU计算
有限元法
刚度矩阵
预处理共轭梯度法
GPU calculation
finite element method
stiffness matrix
preconditioned conjugate gradientmethod
作者简介
张健飞（1977-）,男,江苏海门人,讲师,博士,研究方向为高性能计算、应用数值分析、计算力学与工程仿真,（E-mail）zhjf77@163.com
