Techniques for Optimizing FEM/MoM Codes

This paper presents techniques for optimizing the performance of FEM/MoM codes. The process of computing MoM matrices can be accelerated by choosing the order of Gaussian Quadrature calculations adaptively and taking advantage of the symmetric nature of the impedance (or admittance) matrix. Optimizing loop index order and buffering loop indices can significantly improve the performance of matrix operations. Unrolling techniques can be utilized to exploit the power of high-speed cache memory and reduce data access time. Preconditioning techniques can enhance the spectral properties of the FEM/MoM matrix equations and dramatically improve the convergence rate of iterative solvers. Index Terms Hybrid FEM/MoM, code performance, loop index order, unrolling, buffering, preconditioned iterative solvers.