毕业论文基于gpu的fdtd算法.docVIP

ABSTRACT 摘 要 PAGE 1 摘 要 本文研究的是基于GPU的FDTD算法。GPU（Programmable Graphics Processing Unit），可编程图形处理单元，是英伟达公司首先提出的一个概念，现在一般叫图形处理器，它是将数据流并行处理的方法引入到了硬件的结构中，使其具有高性能的数据并行处理能力。GPU初始是用来处理图形渲染的，经过了从处理2D图像到3D图像的发展历程，随着GPU编程思想的确立，CUDA框架的推出，GPU发展的愈渐成熟，GPU的应用范围也不再局限于计算机图形学本身，而扩展到了各个领域。本文主要是介绍其在通用计算方面的应用，也就是研究它对于C程序的GPU加速过程。 FDTD（Finite Difference Time Domain），是指时域有限差分，这是一种天然的并行计算方法，在电磁场数值问题的计算上有广泛的应用。FDTD的计算过程可以很容易的划分为多个子计算过程，而子计算过程之问同时进行着相似的计算。这个方法主要是把麦克斯韦的方程式在时间上和空间上进行差分化，并且通过时间领域上的更新来模仿电磁场的变化来计算问题。GPU解决了FDTD计算时所需要的巨大内存，而基于GPU的FDTD算法可以对计算进行加速，节省计算时间，降低计算量，这就是研究该课题的原因所在。 关键词：GPU加速，FDTD算法，图形处理器，时域有限差分，CUDA编程 ABSTRACT The paper study the FDTD algorithm by GPU. GPU is graphics processing unit, which NVIDIA first put forword the concept.It is the data stream parallel processing method is introduced into the hardware structure, so it has high performance data parallel processing capability..The original GPU is used to process the graphics rendering, after processing the 2D image to 3D image development.As the GPU programming thought establishment and CUDA framework for the launch, GPU developing more mature.The application scope of GPU is not limited to computer graphics itself, but extend to all areas.This paper mainly introduces its the universal computing applications, is to study it for C program GPU acceleration process. FDTD is finite difference time domain.This is a natural parallel computation method in electromagnetic field numerical calculation, the problem has wide application.FDTD calculation process can be easily divided into a plurality of sub calculation process, and calculation process. At the same time a similar calculation.This method is mainly to Maxwells equations in time and space were poorly differentiated, and through time domain update to mimic the field changes to computing problems. GPU solves the FDTD’s calculation needed huge memory.FDTD based on the GPU algorithm can calculate faster, save time, reduce the amount of calculation, which is the study of the