基于CUDA的晶格Boltzmann并行算法的综合优化设计

摘要

In this paper,a parallel algorithm for lattice Boltzmann method (LBM) is implemented based on GPU in the CUDA framework. In order to improve the efficiency of the algorithm,several typical op-rntimization strategies are investigated,and a new comprehensive optimization solution is obtained. The results of numerical experiments on 3-D Poiseuille flows in a tube show thatithe LBM parallel algorithm designed with the new optimal solution is much more efficient than those from existing optimal solutions.%本文在CUDA框架下设计与实现基于GPU的晶格Boltzmann方法(LBM)的并行算法.为进一步提高计算效率,本文分别研究几种典型的优化策略对计算效率的影响,并给出了一种集多优化策略为一体的综合优化解决方案.以圆管内Poiseuille流为算例的实验表明,采用新综合优化方案设计的LBM并行算法能够获得更高的计算效率.