CUDA架构下的三维弹性静力学边界元并行计算

摘要

The low efficiency of the conventional boundary element method (BEM) is caused by the high computational complexity. In order to accelerate BEM computation, this paper presents a computing unified device architecture (CUDA) based GPU parallel algorithm which is applied in the BEM to solve 3D elastostatics problems. First, the parallelism of different coefficient integrals of the BEM in 3D elastostatics is analyzed and the corresponding GPU parallel algorithm is described. Then the solution method and its parallel strategy of the BEM equations are presented. The example results show that the GPU parallel algorithm is much faster than the conventional method.%针对传统边界元法计算量大、计算效率低的问题,以三维弹性静力学的边界元法为对象,将基于CUDA的GPU并行计算应用到其边界元计算中,提出了基于CUDA架构的GPU并行算法.该算法首先对不同类型的边界元系数积分进行并行性分析,描述了相关的GPU并行算法,然后阐述了边界元方程组的求解方法及其并行策略.实验结果表明,文中算法较传统算法具有显著的加速效果.