基于OpenCL的图形处理器FDTD算法仿真研究

FDTD Simulation Using Graphic Processing Units Based on OpenCL

大型电磁仿真计算的时域有限差分(FDTD)仿真计算通常是十分耗时的,通用图形处理器(GPGPU)技术为其提供了一种合适的解决方案。通过分析FDTD算法特征以及Courant稳定性及数值色散稳定条件,阐述其在并行计算方面的优势。OpenCL是一种新的开放的行业标准,可以用来开发在CPUs,GPUs及其它各种平台上通用的程序。通过阐述OpenCL硬件基础,执行环境,实现方法来增进对其概念的掌握。为充分发挥异构处理平台下GPU的计算能力,提出了基于开放运算语言(OpenCL)模型,并且利用图形处理器并行FDTD仿真的实现方法。并与传统CPU计算相比较,验证计算结果的精确性。通过分析不同网格数量的速度提升情况,结果表明基于OpenCL的GPU计算速度与单CPU相比可以提升几十倍。

It usually takes a lot of time to simulate Large-scale electromagnetic with Finite Difference Time Domain(FDTD) method, but now General Purpose Graphics Processing Units(GPGPU) has provided a viable solution to solve this problem. It can be found that FDTD is an inherently data parallel algorithm by analyzing its algorithm characteristics and its courant stability condition and its numerically stable condition. OpenCL is a new open industry standard that can be used to program CPUs, GPUs, and other devices from different vendors. By introducing OpenCL hardware basis, execution environment, realization method, it could make us understand its conception easier. In order to fully exploit the capability of GPU for general—purpose computing under heterogeneous processing platforms, Parallel method for FDTD algorithm was put forward by using GPU based-on Open Computing Language(OpenCL) model. Comparing with the calculation with traditional CPU, its accuracy was verified. By analyzing calculation speed in different sizes of Yee cells, the acceleration based on OpenCL could be come up to tens of times compare to single CPU.