基于CUDA平台的时域有限差分算法研究  

Study of FDTD algorithm based on CUDA platform

作　　者：沈琛[1] 王璐[1] 胡玉娟[1] 吴先良[1] 

机构地区：[1]合肥师范学院公共计算机教学部,安徽合肥230061

出　　处：《合肥工业大学学报（自然科学版）》2012年第5期644-647,共4页Journal of Hefei University of Technology：Natural Science

基　　金：高等学校省级优秀青年人才基金资助项目(2012SQRL160)

摘　　要：文章针对传统时域有限差分(FDTD)算法的不足,以图形加速卡为核心,通过理论分析和数值模拟,研究并实现了基于CUDA平台的FDTD并行算法。CUDA是最新的可编程多线程的通用计算GPU模型,由于FDTD算法在空间上具有天然的并行性,因此非常适合在GPU上实现并行算。文章描述了在CUDA编程模型上的FDTD算法的设计以及优化过程,并通过数值仿真实验结果证明了基于GPU的并行FDTD算法可以大大减少计算时间,基于GPU加速已成为电磁场数值计算的研究热点之一。In this paper,in view of the deficiencies in traditional finite-difference time-domain（FDTD） algorithm,the FDTD parallel algorithm based on Compute Unified Device Architecture（CUDA） platform is studied and realized by taking the graphics processing unit（GPU） as the core and using theoretical analysis and numerical simulation.The CUDA is a fairly new technology to program inexpensive multi-threaded GPUs.The FDTD algorithm is well suited for parallel processing across spatial domain,thus it is a good candidate for the execution on a GPU.The design and optimization of FDTD algorithm for CUDA programming model are discussed.The results of numerical experiments demonstrate that the computation time of GPU based parallel FDTD algorithm decreases significantly.The GPU based acceleration has been a hot topic in the studies of electromagnetic field numerical computation.

关 键 词：计算机图形处理器(GPU) 统一计算架构(CUDA) 时域有限差分(FDTD)算法 

分 类 号：TN011[电子电信—物理电子学]