Study On GPU Acceleration Of Electromagnetic Scattering From Rough Surface Using Method Of Moment

Computational electromagnetic techniques using massively parallel processorshave gained much support with software tools such as CUDA (Compute UnifiedDevice Architecture) which greatly simplifying the process of programming with GPUin recent years. Accelerations of electromagnetic scattering from rough surface usingMethod of Moment with GPU are presented. The assembly and subsequent solution tothe matrix equation are involved. For the solution to the matrix equation, basic linearalgebra subroutine (cuBLAS) is used. According to the data obtained by ourcalculations, both the assembly and solution of the matrix equation implementationsresult in significant acceleration over CPU implementations, with up to50and100times, respectively. Main works of this thesis are as follows:1. Introductions to the programming model of CUDA, debugging tool: NVIDIAParallel Nsight and performance profiling tool: NVIDIA Visual Profiler are presented.2. Based fundamental of MoM, integral equations and the matrix equations forcalculating the EM scattering from a1-D PEC/dielectric rough surface are provided indetail.3. The matrix-assembly and solution are performed by using GPU in calculatingEM scattering from1-D PEC rough surface, and comparisons of the numericalprecision and calculating time are given between CPU and GPU implementations.4. The matrix-solution are solved with GPU in calculating EM scattering from1-D dielectric surface, and comparisons of the calculating time and numerical precisionare presented by using CPU and GPU.