This paper presents the performance analysis hardware/software co-design of iterative methods for solving linear systems in terms of speed, iteration and toleration. Three iterative methods, Jacobi, Gauss-Seidel (GS) and conjugate gradient (CG), are implemented using Xilink EDK (embedded development kit). For comparison purposes, the same methods are also implemented in pure software using Xilink EDK. Comparisons of the implementations of three methods in software and hardware/software codesign show that the CG method benefits most from the hardware implementation of the matrix multiplication block. These indicators can be used for trade-off of area and speed during the development of reconfigurable hardware accelerator system for large scale circuit and neural simulations.
展开▼
