| OpenCL is a new open industry standard,which can be used to program on CPUs, GPUs, and other devices from different vendors. OpenCL can link up CPU and GPU together properly in order to reduce energy consumption, reduce hardware costs. In addition, it can be used on different kinds of host language such as C, C++, Java, Python, JavaScript, Delphi and so on. Aa a result of that, it can be easily mastered by programmers from different professional backgrounds. Programmers can put their main focus on accelerating algorithm to improve its eficiency, in this way eficient algorithms can be easily written.FDTD is one numberical computing method to study the transmission of electromagnetic to solve Maxwell equations in time domain. In time and space domain,electric fields or magnetic fields alternates turn by turn, in this way Maxwell equations are transformed into difference equations. Then, recursive way can be used to solve these difference equations. This calculation determines it that the characteristics of FDTD requires a lot of iterations. FDTD space step and time step can not be too large because of numerical dispersion and numerical stability limitation. Especially when we use FDTD calculations to solve complex structures electromagnetic simulation problems, it usually costs lots of time and memory.The natural parallel nature of FDTD algorithm determines it that we can use parallel computing method to reduce computing time. Network parallel computing technology can be used to slove this problem in a certain extent but network performance, load balancing, and matching between hardware and software limits its development. The price of FPGA base equipment is very high but its effect isn’t ideal, the versatility and portability of CUDA programming between diffenent isn’t satisfactory either. This paper maily elaborates FDTD algorithm and its basic principle, studies hardware and software environment based on OpenCL model in order to achieve Two-dimensional TM waves FDTD simulation on GPU.The main achievements of my research are as follows:Firstly,the simulation platform is implemented. This paper achieves uniform grid Two-dimensional TM wave electromagnetic simulation based on CPU and GPU.The experimental simulation platforms are AMD Athlon (tm) II X3440CPU and NVIDIA Tesla C1060GPU. This article is derived the basic algorithm recursive formula of FDTD and its convergence condition on Two-dimensional cartesian uniform grid. When we achieve FDTD parallel algorithm based on GPU, CPU and GPU work in the heterogeneous system, GPU takes charge of most most of the calculation while CPU is responsible for memory calling, allocating and conditions controling. When the number of Yee cells reaches million level of magnitude, the advantages of parallel computing on GPU can be fully demonstrated. In this way, the efficiency of computing can be improved so as to reduce the time of electromagnetic simulation.Secondly, the simulation results of Two-dimensional TM wave is analyzed. Each simulation data obtained during the simulation is combined with RGB system through reasonable manner. During the time-domain process, the simulation data is displayed via a dynamic Two-dimensional FDTD images. Firstly,we must verify the accuracy of the simulation results and the result is that the simulation data based on CPU and GPU is consistent. When the Yee cells number reaches millions level of magnitude, by comparison the simulation data between CPU and GPU, OpenCL based on GPU can get a stable acceleration about40times. It just proves it that OpenCL can be used to reduce FDTD parallel computing time greatly to accelerate FDTD parallel algorithm popularization and application in our real life. |
PDF Full Text Request