| And the Electromagnetic scattering is one of the main development directions of computational electromagnetics and has the value of scientific research and national defense strategic significance to study the electromagnetic scattering problems of electrically large size targets with complicated shape. The moment method based on integral equation has become one of the most popular methods in the field of numerical calculation of electromagnetic scattering problems. The moment method is a numerical method based on the integral equation of electromagnetic field, but the global moment method generates dense matrix at the most time. With the continuous development of computer technology study on fast algorithm of the method of moments has become a hot.In this paper, we make parallel acceleration on the most time consuming and space of the method of moments, including the parallel filling algorithm and the parallel iterative algorithm of the impedance matrix of the method of moments based on GPU computation framework. We put forward different fast algorithms by analyzing the scale of impedance matrix, allocating the desired thread and thread block reasonably and making use of the shared memory and constant memory reasonably. Not only that, we test the performance of the algorithm. According the results of the performance test, we propose the improved algorithm to maximize the utilization of hardware resources to achieve higher speedup ratio and load balancing. We make a test on Jacobi, Gauss-Seidel and CGN in different orders and obtain the optimal iterative method. In the iterative process we mainly use the CUBLAS library to realize the matrix vector multiplication.In this paper, we convert moment method calculation results into 3D graphics by the 3D visualization technique. In order to show the surface current of the target under different time, we picture the surface current based on the OpenSceneGraph. So we can observe the target on the real time. |
PDF Full Text Request