Font Size: a A A

Gpu-based Electromagnetic Scattering Integral Equation Method For Parallel Numerical Solution

Posted on:2011-05-22Degree:MasterType:Thesis
Country:ChinaCandidate:J Y XuFull Text:PDF
GTID:2190360308966177Subject:Electromagnetic field and microwave technology
Abstract/Summary:PDF Full Text Request
The analysis of the scattering characteristic of electrically-large object is an important direction in computational electromagnetic, with huge number of unknowns, long computing time and high requirement of hardware as the main feature. We always apply MPI,OPENMP programming technology on large scale workstation cluster or high performance multicore workstation platform for rapid parallel solution. In the last years, a new parallel technology based on personal computer appears, promotes and widely used, this is the CUDA programming model base on GPU. This article implement parallel solution for integral equation about scattering problems apply CUDA programming model based on GPU.This article is an inheritance work, the concrete content is achieve parallel MOM and parallel MLFMA base on GPU. We have already implement about 140 acceleration ratio in MOM and about 7 acceleration ratio in MLFMA, the drawback is MOM method can only handle no more than 10 thousand unknowns and MLFMA can only implement low acceleration ratio. In this background, after reading the program, understanding the program and test the program, this article come up new solutions, as for MOM, we store impedance matrix in main memory instead of video memory, then read the required data to video memory in sequence, in this way, we can apparently handle more unknowns, but the acceleration ratio drops; as for MLFMA, in the multipole expansion to multipole expansion step, the threads that read data from video memory changed to read data from shared memory, in this way, we improved the reading speed, but bring about redundant data in shared memory, in the local expansion to local expansion step, the angular spectrum expansion becomes parallel after arrange groups. after the steps above, the overall acceleration ratio becomes 14.
Keywords/Search Tags:GPU, CUDA programming model, parallel MOM, parallel MLFMA
PDF Full Text Request
Related items